research paper topics on cigarette smoking

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

The PMC website is updating on October 15, 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Nature Portfolio

Health effects associated with smoking: a Burden of Proof study

Xiaochen dai.

1 Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA USA

2 Department of Health Metrics Sciences, School of Medicine, University of Washington, Seattle, WA USA

Gabriela F. Gil

Marissa b. reitsma, noah s. ahmad, jason a. anderson, catherine bisignano, sinclair carr, rachel feldman, simon i. hay, vincent iannucci, hilary r. lawlor, matthew j. malloy, laurie b. marczak, susan a. mclaughlin, larissa morikawa, erin c. mullany, sneha i. nicholson, erin m. o’connell, chukwuma okereke, reed j. d. sorensen, joanna whisnant, aleksandr y. aravkin.

3 Department of Applied Mathematics, University of Washington, Seattle, WA USA

Christopher J. L. Murray

Emmanuela gakidou, associated data.

The findings from the present study are supported by data available in the published literature. Data sources and citations for each risk–outcome pair can be downloaded using the ‘download’ button on each risk curve page currently available at https://vizhub.healthdata.org/burden-of-proof . Study characteristics and citations for all input data used in the analyses are also provided in Supplementary Table 3 , and Supplementary Table 2 provides a template of the data collection form.

All code used for these analyses is publicly available online ( https://github.com/ihmeuw-msca/burden-of-proof ).

As a leading behavioral risk factor for numerous health outcomes, smoking is a major ongoing public health challenge. Although evidence on the health effects of smoking has been widely reported, few attempts have evaluated the dose–response relationship between smoking and a diverse range of health outcomes systematically and comprehensively. In the present study, we re-estimated the dose–response relationships between current smoking and 36 health outcomes by conducting systematic reviews up to 31 May 2022, employing a meta-analytic method that incorporates between-study heterogeneity into estimates of uncertainty. Among the 36 selected outcomes, 8 had strong-to-very-strong evidence of an association with smoking, 21 had weak-to-moderate evidence of association and 7 had no evidence of association. By overcoming many of the limitations of traditional meta-analyses, our approach provides comprehensive, up-to-date and easy-to-use estimates of the evidence on the health effects of smoking. These estimates provide important information for tobacco control advocates, policy makers, researchers, physicians, smokers and the public.

A meta-analysis using the Burden of proof method reported consistent evidence supporting harmful associations between smoking and 28 different health outcomes.

Among both the public and the health experts, smoking is recognized as a major behavioral risk factor with a leading attributable health burden worldwide. The health risks of smoking were clearly outlined in a canonical study of disease rates (including lung cancer) and smoking habits in British doctors in 1950 and have been further elaborated in detail over the following seven decades 1 , 2 . In 2005, evidence of the health consequences of smoking galvanized the adoption of the first World Health Organization (WHO) treaty, the Framework Convention on Tobacco Control, in an attempt to drive reductions in global tobacco use and second-hand smoke exposure 3 . However, as of 2020, an estimated 1.18 billion individuals globally were current smokers and 7 million deaths and 177 million disability-adjusted life-years were attributed to smoking, reflecting a persistent public health challenge 4 . Quantifying the relationship between smoking and various important health outcomes—in particular, highlighting any significant dose–response relationships—is crucial to understanding the attributable health risk experienced by these individuals and informing responsive public policy.

Existing literature on the relationship between smoking and specific health outcomes is prolific, including meta-analyses, cohort studies and case–control studies analyzing the risk of outcomes such as lung cancer 5 – 7 , chronic obstructive pulmonary disease (COPD) 8 – 10 and ischemic heart disease 11 – 14 due to smoking. There are few if any attempts, however, to systematically and comprehensively evaluate the landscape of evidence on smoking risk across a diverse range of health outcomes, with most current research focusing on risk or attributable burden of smoking for a specific condition 7 , 15 , thereby missing the opportunity to provide a comprehensive picture of the health risk experienced by smokers. Furthermore, although evidence surrounding specific health outcomes, such as lung cancer, has generated widespread consensus, findings about the attributable risk of other outcomes are much more heterogeneous and inconclusive 16 – 18 . These studies also vary in their risk definitions, with many comparing dichotomous exposure measures of ever smokers versus nonsmokers 19 , 20 . Others examine the distinct risks of current smokers and former smokers compared with never smokers 21 – 23 . Among the studies that do analyze dose–response relationships, there is large variation in the units and dose categories used in reporting their findings (for example, the use of pack-years or cigarettes per day) 24 , 25 , which complicates the comparability and consolidation of evidence. This, in turn, can obscure data that could inform personal health choices, public health practices and policy measures. Guidance on the health risks of smoking, such as the Surgeon General’s Reports on smoking 26 , 27 , is often based on experts’ evaluation of heterogenous evidence, which, although extremely useful and well suited to carefully consider nuances in the evidence, is fundamentally subjective.

The present study, as part of the Global Burden of Diseases, Risk Factors, and Injuries Study (GBD) 2020, re-estimated the continuous dose–response relationships (the mean risk functions and associated uncertainty estimates) between current smoking and 36 health outcomes (Supplementary Table 1 ) by identifying input studies using a systematic review approach and employing a meta-analytic method 28 . The 36 health outcomes that were selected based on existing evidence of a relationship included 16 cancers (lung cancer, esophageal cancer, stomach cancer, leukemia, liver cancer, laryngeal cancer, breast cancer, cervical cancer, colorectal cancer, lip and oral cavity cancer, nasopharyngeal cancer, other pharynx cancer (excluding nasopharynx cancer), pancreatic cancer, bladder cancer, kidney cancer and prostate cancer), 5 cardiovascular diseases (CVDs: ischemic heart disease, stroke, atrial fibrillation and flutter, aortic aneurysm and peripheral artery disease) and 15 other diseases (COPD, lower respiratory tract infections, tuberculosis, asthma, type 2 diabetes, Alzheimer’s disease and related dementias, Parkinson’s disease, multiple sclerosis, cataracts, gallbladder diseases, low back pain, peptic ulcer disease, rheumatoid arthritis, macular degeneration and fractures). Definitions of the outcomes are described in Supplementary Table 1 . We conducted a separate systematic review for each risk–outcome pair with the exception of cancers, which were done together in a single systematic review. This approach allowed us to systematically identify all relevant studies indexed in PubMed up to 31 May 2022, and we extracted relevant data on risk of smoking, including study characteristics, following a pre-specified template (Supplementary Table 2 ). The meta-analytic tool overcomes many of the limitations of traditional meta-analyses by incorporating between-study heterogeneity into the uncertainty of risk estimates, accounting for small numbers of studies, relaxing the assumption of log(linearity) applied to the risk functions, handling differences in exposure ranges between comparison groups, and systematically testing and adjusting for bias due to study designs and characteristics. We then estimated the burden-of-proof risk function (BPRF) for each risk–outcome pair, as proposed by Zheng et al. 29 ; the BPRF is a conservative risk function defined as the 5th quantile curve (for harmful risks) that reflects the smallest harmful effect at each level of exposure consistent with the available evidence. Given all available data for each outcome, the risk of smoking is at least as harmful as the BPRF indicates.

We used the BPRF for each risk–outcome pair to calculate risk–outcome scores (ROSs) and categorize the strength of evidence for the association between smoking and each health outcome using a star rating from 1 to 5. The interpretation of the star ratings is as follows: 1 star (*) indicates no evidence of association; 2 stars (**) correspond to a 0–15% increase in risk across average range of exposures for harmful risks; 3 stars (***) represent a 15–50% increase in risk; 4 stars (****) refer to >50–85% increase in risk; and 5 stars (*****) equal >85% increase in risk. The thresholds for each star rating were developed in consultation with collaborators and other stakeholders.

The increasing disease burden attributable to current smoking, particularly in low- and middle-income countries 4 , demonstrates the relevance of the present study, which quantifies the strength of the evidence using an objective, quantitative, comprehensive and comparative framework. Findings from the present study can be used to support policy makers in making informed smoking recommendations and regulations focusing on the associations for which the evidence is strongest (that is, the 4- and 5-star associations). However, associations with a lower star rating cannot be ignored, especially when the outcome has high prevalence or severity. A summary of the main findings, limitations and policy implications of the study is presented in Table ​ Table1 1 .

Policy summary

We evaluated the mean risk functions and the BPRFs for 36 health outcomes that are associated with current smoking 30 (Table ​ (Table2). 2 ). Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 31 for each of our systematic reviews, we identified studies reporting relative risk (RR) of incidence or mortality from each of the 36 selected outcomes for smokers compared with nonsmokers. We reviewed 21,108 records, which were identified to have been published between 1 May 2018 and 31 May 2022; this represents the most recent time period since the last systematic review of the available evidence for the GBD at the time of publication. The meta-analyses reported in the present study for each of the 36 health outcomes are based on evidence from a total of 793 studies published between 1970 and 2022 (Extended Data Fig. ​ Fig.1 1 – 5 and Supplementary Information 1.5 show the PRISMA diagrams for each outcome). Only prospective cohort and case–control studies were included for estimating dose–response risk curves, but cross-sectional studies were also included for estimating the age pattern of smoking risk on cardiovascular and circulatory disease (CVD) outcomes. Details on each, including the study’s design, data sources, number of participants, length of follow-up, confounders adjusted for in the input data and bias covariates included in the dose–response risk model, can be found in Supplementary Information 2 and 3 . The theoretical minimum risk exposure level used for current smoking was never smoking or zero 30 .

Strength of the evidence for the relationship between current smoking and the 36 health outcomes analyzed

The ROS represents the signed value of the log(BPRF) averaged across the 15th–85th percentiles of exposure. The BPRF corresponds to the lower (if harmful) or higher (if protective) UI—inclusive of between-study heterogeneity—for each risk–outcome pair’s RR curve. ROSs are directly comparable across outcomes and each risk–outcome pair receives an ROS based on the final formulation of the risk curve. For Parkinson’s disease, the ROS reflects a protective effect of smoking, whereas for the other outcomes it reflects a harmful effect. Negative ROSs indicate that a conservative interpretation of the available evidence suggests that there may be no association between risk and outcome. For ease of interpretation, we have transformed the ROS and BPRF into a star rating (1–5), with a higher rating representing a larger effect and stronger evidence. Average BPRF, which is the exponential ROS for harmful effects (or exponential negative ROS for protective effects), is the conservative exposure-averaged RR consistent with all the available data. Average increased risk, which equates to (average BPRF − 1) × 100% for harmful effects or (1 − average BPRF) × 100% for protective effects, refers to the percentage increase in RR based on a conservative interpretation of the evidence. For harmful risks, this percentage is positive and, for protective risks, negative, indicating the percentage decrease in RR. The average increased risk is not applicable for pairs with negative ROSs. N/A, not available; Pub., Publication; ref., reference.

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and lung cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and prostate cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Five-star associations

When the most conservative interpretation of the evidence, that is, the BPRF, suggests that the average exposure (15th–85th percentiles of exposure) of smoking increases the risk of a health outcome by >85% (that is, ROS > 0.62), smoking and that outcome are categorized as a 5-star pair. Among the 36 outcomes, there are 5 that have a 5-star association with current smoking: laryngeal cancer (375% increase in risk based on the BPRF, 1.56 ROS), aortic aneurysm (150%, 0.92), peripheral artery disease (137%, 0.86), lung cancer (107%, 0.73) and other pharynx cancer (excluding nasopharynx cancer) (92%, 0.65).

Results for all 5-star risk–outcome pairs are available in Table ​ Table2 2 and Supplementary Information 4.1 . In the present study, we provide detailed results for one example 5-star association: current smoking and lung cancer. We extracted 371 observations from 25 prospective cohort studies and 53 case–control studies across 25 locations (Supplementary Table 3 ) 5 , 6 , 32 – 107 . Exposure ranged from 1 pack-year to >112 pack-years, with the 85th percentile of exposure being 50.88 pack-years (Fig. ​ (Fig.1a 1a ).

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on the x axis and the estimated s.d. that includes reported s.d. and between-study heterogeneity on the y axis.

We found a very strong and significant harmful relationship between pack-years of current smoking and the RR of lung cancer (Fig. ​ (Fig.1b). 1b ). The mean RR of lung cancer at 20 pack-years of smoking was 5.11 (95% uncertainty interval (UI) inclusive of between-study heterogeneity = 1.84–14.99). At 50.88 pack-years (85th percentile of exposure), the mean RR of lung cancer was 13.42 (2.63–74.59). See Table ​ Table2 2 for mean RRs at other exposure levels. The BPRF, which represents the most conservative interpretation of the evidence (Fig. ​ (Fig.1a), 1a ), suggests that smoking in the 15th–85th percentiles of exposure increases the risk of lung cancer by an average of 107%, yielding an ROS of 0.73.

The relationship between pack-years of current smoking and RR of lung cancer is nonlinear, with diminishing impact of further pack-years of smoking, particularly for middle-to-high exposure levels (Fig. ​ (Fig.1b). 1b ). To reduce the effect of bias, we adjusted observations that did not account for more than five confounders, including age and sex, because they were the significant bias covariates identified by the bias covariate selection algorithm 29 (Supplementary Table 7 ). The reported RRs across studies were very heterogeneous. Our meta-analytic method, which accounts for the reported uncertainty in both the data and between-study heterogeneity, fit the data and covered the estimated residuals well (Fig. ​ (Fig.1c). 1c ). After trimming 10% of outliers, we still detected publication bias in the results for lung cancer. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 5-star pairs.

Four-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of a health outcome by 50–85% (that is, ROS > 0.41–0.62), smoking is categorized as having a 4-star association with that outcome. We identified three outcomes with a 4-star association with smoking: COPD (72% increase in risk based on the BPRF, 0.54 ROS), lower respiratory tract infection (54%, 0.43) and pancreatic cancer (52%, 0.42).

In the present study, we provide detailed results for one example 4-star association: current smoking and COPD. We extracted 51 observations from 11 prospective cohort studies and 4 case–control studies across 36 locations (Supplementary Table 3 ) 6 , 8 – 10 , 78 , 108 – 116 . Exposure ranged from 1 pack-year to 100 pack-years, with the 85th percentile of exposure in the exposed group being 49.75 pack-years.

We found a strong and significant harmful relationship between pack-years of current smoking and RR of COPD (Fig. ​ (Fig.2b). 2b ). The mean RR of COPD at 20 pack-years was 3.17 (1.60–6.55; Table ​ Table2 2 reports RRs at other exposure levels). At the 85th percentile of exposure, the mean RR of COPD was 6.01 (2.08–18.58). The BPRF suggests that average smoking exposure raises the risk of COPD by an average of 72%, yielding an ROS of 0.54. The results for the other health outcomes that have an association with smoking rated as 4 stars are shown in Table ​ Table2 2 and Supplementary Information 4.2 .

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on th e x axis and the estimated s.d. that includes the reported s.d. and between-study heterogeneity on the y axis.

The relationship between smoking and COPD is nonlinear, with diminishing impact of further pack-years of current smoking on risk of COPD, particularly for middle-to-high exposure levels (Fig. ​ (Fig.2a). 2a ). To reduce the effect of bias, we adjusted observations that did not account for age and sex and/or were generated for individuals aged >65 years 116 , because they were the two significant bias covariates identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was large heterogeneity in the reported RRs across studies, and our meta-analytic method fit the data and covered the estimated residuals well (Fig. ​ (Fig.2b). 2b ). Although we trimmed 10% of outliers, publication bias was still detected in the results for COPD. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for reported RR data and alternative exposures across studies for the remaining health outcomes that have a 4-star association with smoking.

Three-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of a health outcome by 15–50% (or, when protective, decreases the risk of an outcome by 13–34%; that is, ROS >0.14–0.41), the association between smoking and that outcome is categorized as having a 3-star rating. We identified 15 outcomes with a 3-star association: bladder cancer (40% increase in risk, 0.34 ROS); tuberculosis (31%, 0.27); esophageal cancer (29%, 0.26); cervical cancer, multiple sclerosis and rheumatoid arthritis (each 23–24%, 0.21); lower back pain (22%, 0.20); ischemic heart disease (20%, 0.19); peptic ulcer and macular degeneration (each 19–20%, 0.18); Parkinson's disease (protective risk, 15% decrease in risk, 0.16); and stomach cancer, stroke, type 2 diabetes and cataracts (each 15–17%, 0.14–0.16).

We present the findings on smoking and type 2 diabetes as an example of a 3-star risk association. We extracted 102 observations from 24 prospective cohort studies and 4 case–control studies across 15 locations (Supplementary Table 3 ) 117 – 144 . The exposure ranged from 1 cigarette to 60 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 26.25 cigarettes smoked per day.

We found a moderate and significant harmful relationship between cigarettes smoked per day and the RR of type 2 diabetes (Fig. ​ (Fig.3b). 3b ). The mean RR of type 2 diabetes at 20 cigarettes smoked per day was 1.49 (1.18–1.90; see Table ​ Table2 2 for other exposure levels). At the 85th percentile of exposure, the mean RR of type 2 diabetes was 1.54 (1.20–2.01). The BPRF suggests that average smoking exposure raises the risk of type 2 diabetes by an average of 16%, yielding an ROS of 0.15. See Table ​ Table2 2 and Supplementary Information 4.3 for results for the additional health outcomes with an association with smoking rated as 3 stars.

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on the x axis and the estimated s.d. that includes the reported s.d. and between-study heterogeneity on the y axis.

The relationship between smoking and type 2 diabetes is nonlinear, particularly for high exposure levels where the mean risk curve becomes flat (Fig. ​ (Fig.3a). 3a ). We adjusted observations that were generated in subpopulations, because it was the only significant bias covariate identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was moderate heterogeneity in the observed RR data across studies and our meta-analytic method fit the data and covered the estimated residuals extremely well (Fig. 3b,c ). After trimming 10% of outliers, we still detected publication bias in the results for type 2 diabetes. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 3-star pairs.

Two-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of an outcome by 0–15% (that is, ROS 0.0–0.14), the association between smoking and that outcome is categorized as a 2-star rating. We identified six 2-star outcomes: nasopharyngeal cancer (14% increase in risk, 0.13 ROS); Alzheimer’s and other dementia (10%, 0.09); gallbladder diseases and atrial fibrillation and flutter (each 6%, 0.06); lip and oral cavity cancer (5%, 0.05); and breast cancer (4%, 0.04).

We present the findings on smoking and breast cancer as an example of a 2-star association. We extracted 93 observations from 14 prospective cohort studies and 9 case–control studies across 14 locations (Supplementary Table 3 ) 84 , 87 , 145 – 165 . The exposure ranged from 1 cigarette to >76 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 34.10 cigarettes smoked per day.

We found a weak but significant relationship between pack-years of current smoking and RR of breast cancer (Extended Data Fig. ​ Fig.6). 6 ). The mean RR of breast cancer at 20 pack-years was 1.17 (1.04–1.31; Table ​ Table2 2 reports other exposure levels). The BPRF suggests that average smoking exposure raises the risk of breast cancer by an average of 4%, yielding an ROS of 0.04. See Table ​ Table2 2 and Supplementary Information 4.4 for results on the additional health outcomes for which the association with smoking has been categorized as 2 stars.

a , log-relative risk function. b , relative risk function. c , A modified funnel plot showing the residuals (relative to 0) on the x-axis and the estimated standard deviation (SD) that includes reported SD and between-study heterogeneity on the y-axis.

The relationship between smoking and breast cancer is nonlinear, particularly for high exposure levels where the mean risk curve becomes flat (Extended Data Fig. ​ Fig.6a). 6a ). To reduce the effect of bias, we adjusted observations that were generated in subpopulations, because it was the only significant bias covariate identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was heterogeneity in the reported RRs across studies, but our meta-analytic method fit the data and covered the estimated residuals (Extended Data Fig. ​ Fig.6b). 6b ). After trimming 10% of outliers, we did not detect publication bias in the results for breast cancer. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 2-star pairs.

One-star associations

When average exposure to smoking does not significantly increase (or decrease) the risk of an outcome, once between-study heterogeneity and other sources of uncertainty are accounted for (that is, ROS < 0), the association between smoking and that outcome is categorized as 1 star, indicating that there is not sufficient evidence for the effect of smoking on the outcome to reject the null (that is, there may be no association). There were seven outcomes with an association with smoking that rated as 1 star: colorectal and kidney cancer (each –0.01 ROS); leukemia (−0.04); fractures (−0.05); prostate cancer (−0.06); liver cancer (−0.32); and asthma (−0.64).

We use smoking and prostate cancer as examples of a 1-star association. We extracted 78 observations from 21 prospective cohort studies and 1 nested case–control study across 15 locations (Supplementary Table 3 ) 157 , 160 , 166 – 185 . The exposure among the exposed group ranged from 1 cigarette to 90 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 29.73 cigarettes smoked per day.

Based on our conservative interpretation of the data, we did not find a significant relationship between cigarettes smoked per day and the RR of prostate cancer (Fig. ​ (Fig.4B). 4B ). The exposure-averaged BPRF for prostate cancer was 0.94, which was opposite null from the full range of mean RRs, such as 1.16 (0.89–1.53) at 20 cigarettes smoked per day. The corresponding ROS was −0.06, which is consistent with no evidence of an association between smoking and increased risk of prostate cancer. See Table ​ Table2 2 and Supplementary Information 4.5 for results for the additional outcomes that have a 1-star association with smoking.

The relationship between smoking and prostate cancer is nonlinear, particularly for middle-to-high exposure levels where the mean risk curve becomes flat (Fig. ​ (Fig.4a). 4a ). We did not adjust for any bias covariate because no significant bias covariates were selected by the algorithm (Supplementary Table 7 ). The RRs reported across studies were very heterogeneous, but our meta-analytic method fit the data and covered the estimated residuals well (Fig. 4b,c ). The ROS associated with the BPRF is −0.05, suggesting that the most conservative interpretation of all evidence, after accounting for between-study heterogeneity, indicates an inconclusive relationship between smoking exposure and the risk of prostate cancer. After trimming 10% of outliers, we still detected publication bias in the results for prostate cancer, which warrants further studies using sample populations. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 1-star pairs.

Age-specific dose–response risk for CVD outcomes

We produced age-specific dose–response risk curves for the five selected CVD outcomes ( Methods ). The ROS associated with each smoking–CVD pair was calculated based on the reference risk curve estimated using all risk data regardless of age information. Estimation of the BPRF, calculation of the associated ROS and star rating of the smoking–CVD pairs follow the same rules as the other non-CVD smoking–outcome pairs (Table ​ (Table1 1 and Supplementary Figs. 2 – 4 ). Once we had estimated the reference dose–response risk curve for each CVD outcome, we determined the age group of the reference risk curve. The reference age group is 55–59 years for all CVD outcomes, except for peripheral artery disease, the reference age group for which is 60–64 years. We then estimated the age pattern of smoking on all CVD outcomes (Supplementary Fig. 2 ) and calculated age attenuation factors of the risk for each age group by comparing the risk of each age group with that of the reference age group, using the estimated age pattern (Supplementary Fig. 3 ). Last, we applied the draws of age attenuation factors of each age group to the dose–response risk curve for the reference age group to produce the age group-specific dose–response risk curves for each CVD outcome (Supplementary Fig. 4 ).

Using our burden-of-proof meta-analytic methods, we re-estimated the dose–response risk of smoking on 36 health outcomes that had previously been demonstrated to be associated with smoking 30 , 186 . Using these methods, which account for both the reported uncertainty of the data and the between-study heterogeneity, we found that 29 of the 36 smoking–outcome pairs are supported by evidence that suggests a significant dose–response relationship between smoking and the given outcome (28 with a harmful association and 1 with a protective association). Conversely, after accounting for between-study heterogeneity, the available evidence of smoking risk on seven outcomes (that is, colon and rectum cancer, kidney cancer, leukemia, prostate cancer, fractures, liver cancer and asthma) was insufficient to reject the null or draw definitive conclusions on their relationship to smoking. Among the 29 outcomes that have evidence supporting a significant relationship to smoking, 8 had strong-to-very-strong evidence of a relationship, meaning that, given all the available data on smoking risk, we estimate that average exposure to smoking increases the risk of those outcomes by >50% (4- and 5-star outcomes). The currently available evidence for the remaining 21 outcomes with a significant association with current smoking was weak to moderate, indicating that smoking increases the risk of those outcomes by at least >0–50% (2- and 3-star associations).

Even under our conservative interpretation of the data, smoking is irrefutably harmful to human health, with the greatest increases in risk occurring for laryngeal cancer, aortic aneurysm, peripheral artery disease, lung cancer and other pharynx cancer (excluding nasopharynx cancer), which collectively represent large causes of death and ill-health. The magnitude of and evidence for the associations between smoking and its leading health outcomes are among the highest currently analyzed in the burden-of-proof framework 29 . The star ratings assigned to each smoking–outcome pair offer policy makers a way of categorizing and comparing the evidence for a relationship between smoking and its potential health outcomes ( https://vizhub.healthdata.org/burden-of-proof ). We found that, for seven outcomes in our analysis, there was insufficient or inconsistent evidence to demonstrate a significant association with smoking. This is a key finding because it demonstrates the need for more high-quality data for these particular outcomes; availability of more data should improve the strength of evidence for whether or not there is an association between smoking and these health outcomes.

Our systematic review approach and meta-analytic methods have numerous benefits over existing systematic reviews and meta-analyses on the same topic that use traditional random effects models. First, our approach relaxes the log(linear) assumption, using a spline ensemble to estimate the risk 29 . Second, our approach allows variable reference groups and exposure ranges, allowing for more accurate estimates regardless of whether or not the underlying relative risk is log(linear). Furthermore, it can detect outliers in the data automatically. Finally, it quantifies uncertainty due to between-study heterogeneity while accounting for small numbers of studies, minimizing the risk that conclusions will be drawn based on spurious findings.

We believe that the results for the association between smoking and each of the 36 health outcomes generated by the present study, including the mean risk function, BPRF, ROS, average excess risk and star rating, could be useful to a range of stakeholders. Policy makers can formulate their decisions on smoking control priorities and resource allocation based on the magnitude of the effect and the consistency of the evidence relating smoking to each of the 36 outcomes, as represented by the ROS and star rating for each smoking–outcome association 187 . Physicians and public health practitioners can use the estimates of average increased risk and the star rating to educate patients and the general public about the risk of smoking and to promote smoking cessation 188 . Researchers can use the estimated mean risk function or BPRF to obtain the risk of an outcome at a given smoking exposure level, as well as uncertainty surrounding that estimate of risk. The results can also be used in the estimation of risk-attributable burden, that is, the deaths and disability-adjusted life-years due to each outcome that are attributable to smoking 30 , 186 . For the general public, these results could help them to better understand the risk of smoking and manage their health 189 .

Although our meta-analysis was comprehensive and carefully conducted, there are limitations to acknowledge. First, the bias covariates used, although carefully extracted and evaluated, were based on observable study characteristics and thus may not fully capture unobserved characteristics such as study quality or context, which might be major sources of bias. Second, if multiple risk estimates with different adjustment levels were reported in a given study, we included only the fully adjusted risk estimate and modeled the adjustment level according to the number of covariates adjusted for (rather than which covariates were adjusted for) and whether a standard adjustment for age and sex had been applied. This approach limited our ability to make full use of all available risk estimates in the literature. Third, although we evaluated the potential for publication bias in the data, we did not test for other forms of bias such as when studies are more consistent with each other than expected by chance 29 . Fourth, our analysis assumes that the relationships between smoking and health outcomes are similar across geographical regions and over time. We do not have sufficient evidence to quantify how the relationships may have evolved over time because the composition of smoking products has also changed over time. Perhaps some of the heterogeneity of the effect sizes in published studies reflects this; however, this cannot be discerned with the currently available information.

In the future, we plan to include crude and partially adjusted risk estimates in our analyses to fully incorporate all available risk estimates, to model the adjusted covariates in a more comprehensive way by mapping the adjusted covariates across all studies comprehensively and systematically, and to develop methods to evaluate additional forms of potential bias. We plan to update our results on a regular basis to provide timely and up-to-date evidence to stakeholders.

To conclude, we have re-estimated the dose–response risk of smoking on 36 health outcomes while synthesizing all the available evidence up to 31 May 2022. We found that, even after factoring in the heterogeneity between studies and other sources of uncertainty, smoking has a strong-to-very-strong association with a range of health outcomes and confirmed that smoking is irrefutably highly harmful to human health. We found that, due to small numbers of studies, inconsistency in the data, small effect sizes or a combination of these reasons, seven outcomes for which some previous research had found an association with smoking did not—under our meta-analytic framework and conservative approach to interpreting the data—have evidence of an association. Our estimates of the evidence for risk of smoking on 36 selected health outcomes have the potential to inform the many stakeholders of smoking control, including policy makers, researchers, public health professionals, physicians, smokers and the general public.

For the present study, we used a meta-analytic tool, MR-BRT (metaregression—Bayesian, regularized, trimmed), to estimate the dose–response risk curves of the risk of a health outcome across the range of current smoking levels along with uncertainty estimates 28 . Compared with traditional meta-analysis using linear mixed effect models, MR-BRT relaxes the assumption of a log(linear) relationship between exposure and risk, incorporates between-study heterogeneity into the uncertainty of risk estimates, handles estimates reported across different exposure categories, automatically identifies and trims outliers, and systematically tests and adjusts for bias due to study designs and characteristics. The meta-analytic methods employed by the present study followed the six main steps proposed by Zheng et al. 28 , 29 , namely: (1) enacting a systematic review approach and data extraction following a pre-specified and standardized protocol; (2) estimating the shape of the relationship between exposure and RR; (3) evaluating and adjusting for systematic bias as a function of study characteristics and risk estimation; (4) quantifying between-study heterogeneity while adjusting for within-study correlation and the number of studies; (5) evaluating potential publication or reporting biases; and (6) estimating the mean risk function and the BPRF, calculating the ROS and categorizing smoking–outcome pairs using a star-rating scheme from 1 to 5.

The estimates for our primary indicators of this work—mean RRs across a range of exposures, BRPFs, ROSs and star ratings for each risk–outcome pair—are not specific to or disaggregated by specific populations. We did not estimate RRs separately for different locations, sexes (although the RR of prostate cancer was estimated only for males and of cervical and breast cancer only for females) or age groups (although this analysis was applied to disease endpoints in adults aged ≥30 years only and, as detailed below, age-specific estimates were produced for the five CVD outcomes).

The present study complies with the PRISMA guidelines 190 (Supplementary Tables 9 and 10 and Supplementary Information 1.5 ) and Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations 191 (Supplementary Table 11 ). The study was approved by the University of Washington Institutional Review Board (study no. 9060). The systematic review approach was not registered.

Selecting health outcomes

In the present study, current smoking is defined as the current use of any smoked tobacco product on a daily or occasional basis. Health outcomes were initially selected using the World Cancer Research Fund criteria for convincing or probable evidence as described in Murray et al. 186 . The 36 health outcomes that were selected based on existing evidence of a relationship included 16 cancers (lung cancer, esophageal cancer, stomach cancer, leukemia, liver cancer, laryngeal cancer, breast cancer, cervical cancer, colorectal cancer, lip and oral cavity cancer, nasopharyngeal cancer, other pharynx cancer (excluding nasopharynx cancer), pancreatic cancer, bladder cancer, kidney cancer and prostate cancer), 5 CVDs (ischemic heart disease, stroke, atrial fibrillation and flutter, aortic aneurysm and peripheral artery disease) and 15 other diseases (COPD, lower respiratory tract infections, tuberculosis, asthma, type 2 diabetes, Alzheimer’s disease and related dementias, Parkinson’s disease, multiple sclerosis, cataracts, gallbladder diseases, low back pain, peptic ulcer disease, rheumatoid arthritis, macular degeneration and fracture). Definitions of the outcomes are described in Supplementary Table 1 .

Step 1: systematic review approach to literature search and data extraction

Informed by the systematic review approach we took for the GBD 2019 (ref. 30 ), for the present study we identified input studies in the literature using a systematic review approach for all 36 smoking–outcome pairs using updated search strings to identify all relevant studies indexed in PubMed up to 31 May 2022 and extracted data on smoking risk estimates. Briefly, the studies that were extracted represented several types of study design (for example, cohort and case–control studies), measured exposure in several different ways and varied in their choice of reference categories (where some compared current smokers with never smokers, whereas others compared current smokers with nonsmokers or former smokers). All these study characteristics were catalogued systematically and taken into consideration during the modeling part of the analysis.

In addition, for CVD outcomes, we also estimated the age pattern of risk associated with smoking. We applied a systematic review of literature approach for smoking risk for the five CVD outcomes. We developed a search string to search for studies reporting any association between binary smoking status (that is, current, former and ever smokers) and the five CVD outcomes from 1 January 1970 to 31 May 2022, and included only studies reporting age-specific risk (RR, odds ratio (OR), hazard ratio (HR)) of smoking status. The inclusion criteria and results of the systematic review approach are reported in accordance with PRISMA guidelines 31 . Details for each outcome on the search string used in the systematic review approach, refined inclusion and exclusion criteria, data extraction template and PRISMA diagram are given in Supplementary Information 1 . Title and/or abstract screening, full text screening and data extraction were conducted by 14 members of the research team and extracted data underwent manual quality assurance by the research team to verify accuracy.

Selecting exposure categories

Cumulative exposure in pack-years was the measure of exposure used for COPD and all cancer outcomes except for prostate cancer, to reflect the risk of both duration and intensity of current smoking on these outcomes. For prostate cancer, CVDs and all the other outcomes except for fractures, we used cigarette-equivalents smoked per day as the exposure for current smoking, because smoking intensity is generally thought to be more important than duration for these outcomes. For fractures, we used binary exposure, because there were few studies examining intensity or duration of smoking on fractures. The smoking–outcome pairs and the corresponding exposures are summarized in Supplementary Table 4 and are congruent with the GBD 2019 (refs. 30 , 186 ).

Steps 2–5: modeling dose–response RR of smoking on the selected health outcomes

Of the six steps proposed by Zheng et al. 29 , steps 2–5 cover the process of modeling dose–response risk curves. In step 2, we estimated the shape (or the ‘signal’) of the dose–response risk curves, integrating over different exposure ranges. To relax the log(linear) assumption usually applied to continuous dose–response risk and make the estimates robust to the placement of spline knots, we used an ensemble spline approach to fit the functional form of the dose–response relationship. The final ensemble model was a weighted combination of 50 models with random knot placement, with the weight of each model proportional to measures of model fit and total variation. To avoid the influence of extreme data and reduce publication bias, we trimmed 10% of data for each outcome as outliers. We also applied a monotonicity constraint to ensure that the mean risk curves were nondecreasing (or nonincreasing in the case of Parkinson’s disease).

In step 3, following the GRADE approach 192 , 193 , we quantified risk of bias across six domains, namely, representativeness of the study population, exposure, outcome, reverse causation, control for confounding and selection bias. Details about the bias covariates are provided in Supplementary Table 4 . We systematically tested for the effect of bias covariates using metaregression, selected significant bias covariates using the Lasso approach 194 , 195 and adjusted for the selected bias covariates in the final risk curve.

In step 4, we quantified between-study heterogeneity accounting for within-study correlation, uncertainty of the heterogeneity, as well as small number of studies. Specifically, we used a random intercept in the mixed-effects model to account for the within-study correlation and used a study-specific random slope with respect to the ‘signal’ to capture between-study heterogeneity. As between-study heterogeneity can be underestimated or even zero when the number of studies is small 196 , 197 , we used Fisher’s information matrix to estimate the uncertainty of the heterogeneity 198 and incorporated that uncertainty into the final results.

In step 5, in addition to generating funnel plots and visually inspecting for asymmetry (Figs. ​ (Figs.1c, 1c , ​ ,2c, 2c , ​ ,3c 3c and ​ and4c 4c and Extended Data Fig. ​ Fig.6c) 6c ) to identify potential publication bias, we also statistically tested for potential publication or reporting bias using Egger’s regression 199 . We flagged potential publication bias in the data but did not correct for it, which is in line with the general literature 10 , 200 , 201 . Full details about the modeling process have been published elsewhere 29 and model specifications for each outcome are in Supplementary Table 6 .

Step 6: estimating the mean risk function and the BPRF

In the final step, step 6, the metaregression model inclusive of the selected bias covariates from step 3 (for example, the highest adjustment level) was used to predict the mean risk function and its 95% UI, which incorporated the uncertainty of the mean effect, between-study heterogeneity and the uncertainty in the heterogeneity estimate accounting for small numbers of studies. Specifically, 1,000 draws were created for each 0.1 level of doses from 0 pack-years to 100 pack-years or cigarette-equivalents smoked per day using the Bayesian metaregression model. The mean of the 1,000 draws was used to estimate the mean risk at each exposure level, and the 25th and 95th draws were used to estimate the 95% UIs for the mean risk at each exposure level.

The BPRF 29 is a conservative estimate of risk function consistent with the available evidence, correcting for both between-study heterogeneity and systemic biases related to study characteristics. The BPRF is defined as either the 5th (if harmful) or 95th (if protective) quantile curve closest to the line of log(RR) of 0, which defines the null (Figs. ​ (Figs.1a, 1a , ​ ,2b, 2b , ​ ,3a 3a and ​ and4a). 4a ). The BPRF represents the smallest harmful (or protective) effect of smoking on the corresponding outcome at each level of exposure that is consistent with the available evidence. A BPRF opposite null from the mean risk function indicates that insufficient evidence is available to reject null, that is, that there may not be an association between risk and outcome. Likewise, the further the BPRF is from null on the same side of null as the mean risk function, the higher the magnitude and evidence for the relationship. The BPRF can be interpreted as indicating that, even accounting for between-study heterogeneity and its uncertainty, the log(RR) across the studied smoking range is at least as high as the BPRF (or at least as low as the BPRF for a protective risk).

To quantify the strength of the evidence, we calculated the ROS for each smoking–outcome association as the signed value of the log(BPRF) averaged between the 15th and 85th percentiles of observed exposure levels for each outcome. The ROS is a single summary of the effect of smoking on the outcome, with higher positive ROSs corresponding to stronger and more consistent evidence and a higher average effect size of smoking and a negative ROS, suggesting that, based on the available evidence, there is no significant effect of smoking on the outcome after accounting for between-study heterogeneity.

For ease of communication, we further classified each smoking–outcome association into a star rating from 1 to 5. Briefly, 1-star associations have an ROS <0, indicating that there is insufficient evidence to find a significant association between smoking and the selected outcome. We divided the positive ROSs into ranges 0.0–0.14 (2-star), >0.14–0.41 (3-star), >0.41–0.62 (4-star) and >0.62 (5-star). These categories correspond to excess risk ranges for harmful risks of 0–15%, >15–50%, >50–85% and >85%. For protective risks, the ranges of exposure-averaged decreases in risk by star rating are 0–13% (2 stars), >13–34% (3 stars), >34–46% (4 stars) and >46% (5 stars).

Among the 36 smoking–outcome pairs analyzed, smoking fracture was the only binary risk–outcome pair, which was due to limited data on the dose–response risk of smoking on fracture 202 . The estimation of binary risk was simplified because the RR was merely a comparison between current smokers and nonsmokers or never smokers. The concept of ROS for continuous risk can naturally extend to binary risk because the BPRF is still defined as the 5th percentile of the effect size accounting for data uncertainty and between-study heterogeneity. However, binary ROSs must be divided by 2 to make them comparable with continuous ROSs, which were calculated by averaging the risk over the range between the 15th and the 85th percentiles of observed exposure levels. Full details about estimating mean risk functions, BPRFs and ROSs for both continuous and binary risk–outcome pairs can be found elsewhere 29 .

Estimating the age-specific risk function for CVD outcomes

For non-CVD outcomes, we assumed that the risk function was the same for all ages and all sexes, except for breast, cervical and prostate cancer, which were assumed to apply only to females or males, respectively. As the risk of smoking on CVD outcomes is known to attenuate with increasing age 203 – 206 , we adopted a four-step approach for GBD 2020 to produce age-specific dose–response risk curves for CVD outcomes.

First, we estimated the reference dose–response risk of smoking for each CVD outcome using dose-specific RR data for each outcome regardless of the age group information. This step was identical to that implemented for the other non-CVD outcomes. Once we had generated the reference curve, we determined the age group associated with it by calculating the weighted mean age across all dose-specific RR data (weighted by the reciprocal of the s.e.m. of each datum). For example, if the weighted mean age of all dose-specific RR data was 56.5, we estimated the age group associated with the reference risk curve to be aged 55–59 years. For cohort studies, the age range associated with the RR estimate was calculated as a mean age at baseline plus the mean/median years of follow-up (if only the maximum years of follow-up were reported, we would halve this value and add it to the mean age at baseline). For case–control studies, the age range associated with the OR estimate was simply the reported mean age at baseline (if mean age was not reported, we used the midpoint of the age range instead).

In the third step, we extracted age group-specific RR data and relevant bias covariates from the studies identified in our systematic review approach of age-specific smoking risk on CVD outcomes, and used MR-BRT to model the age pattern of excess risk (that is, RR-1) of smoking on CVD outcomes with age group-specific excess RR data for all CVD outcomes. We modeled the age pattern of smoking risk on CVDs following the same steps we implemented for modeling dose–response risk curves. In the final model, we included a spline on age, random slope on age by study and the bias covariate encoding exposure definition (that is, current, former and ever smokers), which was picked by the variable selection algorithm 28 , 29 . When predicting the age pattern of the excess risk of smoking on CVD outcomes using the fitted model, we did not include between-study heterogeneity to reduce uncertainty in the prediction.

In the fourth step, we calculated the age attenuation factors of excess risk compared with the reference age group for each CVD outcome as the ratio of the estimated excess risk for each age group to the excess risk for the reference age group. We performed the calculation at the draw level to obtain 1,000 draws of the age attenuation factors for each age group. Once we had estimated the age attenuation factors, we carried out the last step, which consisted of adjusting the risk curve for the reference age group from step 1 using equation (1) to produce the age group-specific risk curves for each CVD outcome:

We implemented the age adjustment at the draw level so that the uncertainty of the age attenuation factors could be naturally incorporated into the final adjusted age-specific RR curves. A PRISMA diagram detailing the systematic review approach, a description of the studies included and the full details about the methods are in Supplementary Information 1.5 and 5.2 .

Estimating the theoretical minimum risk exposure level

The theoretical minimum risk exposure level for smoking was 0, that is, no individuals in the population are current or former smokers.

Model validation

The validity of the meta-analytic tool has been extensively evaluated by Zheng and colleagues using simulation experiments 28 , 29 . For the present study, we conducted two additional sensitivity analyses to examine how the shape of the risk curves was impacted by applying a monotonicity constraint and trimming 10% of data. We present the results of these sensitivity analyses in Supplementary Information 6 . In addition to the sensitivity analyses, the dose–response risk estimates were also validated by plotting the mean risk function along with its 95% UI against both the extracted dose-specific RR data from the studies included and our previous dose–response risk estimates from the GBD 2019 (ref. 30 ). The mean risk functions along with the 95% UIs were validated based on data fit and the level, shape and plausibility of the dose–response risk curves. All curves were validated by all authors and reviewed by an external expert panel, comprising professors with relevant experience from universities including Johns Hopkins University, Karolinska Institute and University of Barcelona; senior scientists working in relevant departments at the WHO and the Center for Disease Control and Prevention (CDC) and directors of nongovernmental organizations such as the Campaign for Tobacco-Free Kids.

Statistical analysis

Analyses were carried out using R v.3.6.3, Python v.3.8 and Stata v.16.

Statistics and reproducibility

The study was a secondary analysis of existing data involving systematic reviews and meta-analyses. No statistical method was used to predetermine sample size. As the study did not involve primary data collection, randomization and blinding, data exclusions were not relevant to the present study, and, as such, no data were excluded and we performed no randomization or blinding. We have made our data and code available to foster reproducibility.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Research reporting summaries, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-022-01978-x.

Supplementary information

Supplementary Information 1: Data source identification and assessment. Supplementary Information 2: Data inputs. Supplementary Information 3: Study quality and bias assessment. Supplementary Information 4: The dose–response RR curves and their 95% UIs for all smoking–outcome pairs. Supplementary Information 5: Supplementary methods. Supplementary Information 6: Sensitivity analysis. Supplementary Information 7: Binary smoking–outcome pair. Supplementary Information 8: Risk curve details. Supplementary Information 9: GATHER and PRISMA checklists.

Acknowledgements

Research reported in this publication was supported by the Bill & Melinda Gates Foundation and Bloomberg Philanthropies. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funders. The study funders had no role in study design, data collection, data analysis, data interpretation, writing of the final report or the decision to publish.

We thank the Tobacco Metrics Team Advisory Group for their valuable input and review of the work. The members of the Advisory Group are: P. Allebeck, R. Chandora, J. Drope, M. Eriksen, E. Fernández, H. Gouda, R. Kennedy, D. McGoldrick, L. Pan, K. Schotte, E. Sebrie, J. Soriano, M. Tynan and K. Welding.

Extended data

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and chronic obstructive pulmonary disease conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and type 2 diabetes conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and breast cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Author contributions

X.D., S.I.H., S.A.M., E.C.M., E.M.O., C.J.L.M. and E.G. managed the estimation or publications process. X.D. and G.F.G. wrote the first draft of the manuscript. X.D. and P.Z. had primary responsibility for applying analytical methods to produce estimates. X.D., G.F.G., N.S.A., J.A.A., S.C., R.F., V.I., M.J.M., L.M., S.I.N., C.O., M.B.R. and J.W. had primary responsibility for seeking, cataloguing, extracting or cleaning data, and for designing or coding figures and tables. X.D., G.F.G., M.B.R., N.S.A., H.R.L., C.O. and J.W. provided data or critical feedback on data sources. X.D., J.H., R.J.D.S., A.Y.A., P.Z., C.J.L.M. and E.G. developed methods or computational machinery. X.D., G.F.G., M.B.R., S.I.H., J.H., R.J.D.S., A.Y.A., P.Z., C.J.L.M. and E.G. provided critical feedback on methods or results. X.D., G.F.G., M.B.R., C.B., S.I.H., L.B.M., S.A.M., A.Y.A. and E.G. drafted the work or revised it critically for important intellectual content. X.D., S.I.H., L.B.M., E.C.M., E.M.O. and E.G. managed the overall research enterprise.

Peer review

Peer review information.

Nature Medicine thanks Frederic Sitas and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Jennifer Sargent and Ming Yang, in collaboration with the Nature Medicine team.

Data availability

Code availability, competing interests.

The authors declare no competing interests.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

are available for this paper at 10.1038/s41591-022-01978-x.

The online version contains supplementary material available at 10.1038/s41591-022-01978-x.

• Open access
• Published: 30 July 2021

Impact of tobacco and/or nicotine products on health and functioning: a scoping review and findings from the preparatory phase of the development of a new self-report measure

• Esther F. Afolalu   ORCID: orcid.org/0000-0001-8866-4765 1 ,
• Erica Spies 1 ,
• Agnes Bacso 1 ,
• Emilie Clerc 1 ,
• Linda Abetz-Webb 2 ,
• Sophie Gallot 1 &
• Christelle Chrea 1  

Harm Reduction Journal volume  18 , Article number:  79 ( 2021 ) Cite this article

17k Accesses

4 Citations

11 Altmetric

Metrics details

Measuring self-reported experience of health and functioning is important for understanding the changes in the health status of individuals switching from cigarettes to less harmful tobacco and/or nicotine products (TNP) or reduced-risk products (RRP) and for supporting tobacco harm reduction strategies.

This paper presents insights from three research activities from the preparatory phase of the development of a new self-report health and functioning measure. A scoping literature review was conducted to identify the positive and negative impact of TNP use on health and functioning. Focus groups ( n  = 29) on risk perception and individual interviews ( n  = 40) on perceived dependence in people who use TNPs were reanalyzed in the context of health and functioning, and expert opinion was gathered from five key opinion leaders and five technical consultants.

Triangulating the findings of the review of 97 articles, qualitative input from people who use TNPs, and expert feedback helped generate a preliminary conceptual framework including health and functioning and conceptually-related domains impacted by TNP use. Domains related to the future health and functioning measurement model include physical health signs and symptoms, general physical appearance, functioning (physical, sexual, cognitive, emotional, and social), and general health perceptions.

Conclusions

This preliminary conceptual framework can inform future research on development and validation of new measures for assessment of overall health and functioning impact of TNPs from the consumers’ perspective.

As a leading cause of preventable morbidity and mortality worldwide, smoking remains a major public health problem. Compared with those who do not smoke, people who smoke are significantly more likely to develop heart diseases, lung cancer, chronic obstructive pulmonary disease (COPD), and other diseases [ 1 , 2 ]. It is well established that the best way to avoid the health risks associated with smoking is for people to never start and for those who smoke to quit [ 1 , 3 ]. Tobacco harm reduction is one way to alleviate the health risk for individuals who choose not to quit smoking [ 4 ], by providing less harmful tobacco and/or nicotine products (TNP), such as reduced-risk products (RRP) (used here to refer to products that present, are likely to present, or have the potential to present, less risk of harm to people who smoke and switch to these products versus continued smoking) or modified risk tobacco products (MRTP).

Several smokeless tobacco products and a heated tobacco product were recently authorized for marketing with modified risk claims through the United States (US) Food and Drug Administration (FDA) MRTP pathway [ 5 ]. The guidance on MRTP applications [ 6 ] specifies that health outcomes should be assessed during premarket evaluation and postmarket surveillance of modified risk TNPs such as these. These health outcomes comprise not only objective clinical and biological measures but also self-reported outcomes [ 6 , 7 ]. Studies and reports have recently started providing evidence on the health impact of new TNPs [ 8 ]. For instance, recent papers have investigated the effects of e-cigarettes and heated tobacco products on cardiopulmonary outcomes [ 9 , 10 , 11 , 12 , 13 , 14 ]. However, the papers have mainly focused on clinical measurements, such as spirometry and other lung function tests; consumer perception is rarely explored or the focus of the research. Measuring self-reported experience is important for understanding the changes in the health status of individuals switching from cigarettes to RRPs and is a key component of tobacco harm reduction strategies [ 7 ]. Self-reported ratings of RRP effectiveness or adverse events might differ from clinical measures and provide another perspective as useful as the clinicians. In addition, consumer perception of positive changes in health status, functioning and other behavioral outcomes will also subsequently influence use behaviors and switching to RRPs rather than continuing smoking.

Self-perceived health status is a complex concept to define and measure, particularly within the context of TNP use [ 15 ]. While generic health status measures, such as the Medical Outcomes Study 36-item Short-Form Health Survey (SF-36), have been used to evaluate the health status of people who smoke [ 16 , 17 ], comparisons have mainly been made between those who currently smoke, those who used to smoke, and those who never smoked [ 18 , 19 ]. Results from these studies strongly suggest that, in healthy populations, existing generic measures are not sensitive enough to detect change over time when stopping or switching from cigarettes to other TNPs, owing to high ceiling effects [ 20 ]. While a few smoking-specific quality of life measures have been developed, these measures have not been widely implemented or standardized [ 15 , 17 , 21 , 22 ], and the application of these smoking-specific measures to different TNP use across the risk continuum is scarce [ 20 ].

As part of the A ssessment of B ehavioral OU tcomes related to T obacco and Nicotine Products (ABOUT™) Toolbox initiative [ 23 ], the present project aims at developing a new self-report measure (ABOUT™— Health and Functioning ) to address the current gap and assess the impact of TNPs on health and functioning (including health status, functional status and other health-related quality of life constructs). This paper presents insights from three research activities [ 24 , 25 ] from the preparatory phase of development of the measure—that is, a scoping literature review, reanalysis of consumer focus groups/interviews, and expert opinion. These three activities serve as background research to support the development of a preliminary conceptual framework of health and functioning associated with the use of TNPs.

Scoping literature review

The purpose of the review was to address two main questions among individuals who use TNPs:

What are the positive and negative health and functioning impacts of TNP use?

What concepts are evaluated by measures used to assess the positive and negative impacts of TNP use?

Given the nature and breadth of the research questions and the number of potentially relevant publications, a scoping literature review was used as it provides a means of identifying the literature and mapping the concepts and evidence on a topic by using an informative and iterative research process [ 26 ]. The scoping review involved a PubMed search (August 2018) and application of Sciome’s rapid Evidence Mapping (rEM) [ 27 ], followed by additional manual screening and review. rEM is a proprietary methodology developed by Sciome ( https://www.sciome.com/ ) to rapidly summarize and produce a quantitative representation of the available body of scientific evidence in a particular area. The study by Lam et al. demonstrated a proof-of-concept application of the rEM methodology [ 27 ]. The PubMed search terms targeted qualitative and quantitative research among people who use TNPs (Table 1 ). This was supplemented by a second, parallel step of manually identifying relevant literature through other known sources. Table 2 describes the general inclusion and exclusion criteria that were applied to the scoping literature review.

After the initial rEM exercise, two reviewers (EC, SG) further manually screened the titles and abstracts of the articles identified through the automated rEM exercise against the inclusion and exclusion criteria. Finally, the selected publications underwent a full screening by two reviewers (VL and DF) for determining their relevance to the research questions for data extraction and one of the co-authors (LA-W) cross-checked the screening and resolved differences in opinion among the reviewers.

The World Health Organization (WHO) International Classification of Functioning, Disability and Health (ICF) [ 28 ] framework and the revised Wilson and Cleary [ 29 , 30 ] model were used as a guide to broadly inform categories for data extraction from the literature on TNP use and health and functioning. These established models enable the conceptualization and description of health status and functioning (the combination of which is often referred to as health-related quality of life) [ 31 , 32 ], and related outcomes and determinants. To complement and refine this and to ensure relevance to those who use TNPs, the data extracted from the literature was also grouped and labeled based on the contents of the literature reviewed.

The elements extracted from the selected papers were as follows:

Author, citation details, and publication type

Objectives and/or research questions

Sample type, size, and principle demographics

Type(s) of TNP and definitions of levels of consumption

Methodology, questionnaires, and statistical methods used

Main results

Results grouped in broad categories: Health Signs and Symptoms; General Health Perceptions; Quality of Life, Health-Related Quality of Life, and Functional Status; Individual Characteristics; Environmental and Social Characteristics; Biomarkers and Biological Endpoints.

Reanalysis of focus groups/in-depth interviews

The objective of the secondary analyses of existing qualitative data in people who use TNPs was to inform the drafting of the initial conceptual framework, as well as interview guides for planned concept elicitation qualitative studies to identify concepts and develop items to detect what is relevant to measure in this context. Two sets of qualitative data containing information related to health and functioning were reanalyzed and participants had consented for their data to be used in future studies. The first was from 29 focus groups (total number of participants n  = 229) that were originally designed to discuss perceived risk, appeal, and intent to use TNPs [ 33 , 34 ]. The focus groups—stratified by smoking status—were conducted in the United States (US; n  = 12), Japan ( n  = 4), Italy ( n  = 4), and the United Kingdom (UK; n  = 9) between December 2012 and August 2013. The second dataset included 40 in-depth interviews conducted in North Carolina, USA, with people who use TNPs, to discuss issues centered on perceived dependence on TNPs [ 35 ]. While 21 interviewees were people who were poly-TNPs users, 19 were people who were exclusive users of one of the following types of TNPs: cigarettes ( n  = 5), smokeless tobacco ( n  = 5), e-cigarettes ( n  = 5), or another type of TNP (pipes, waterpipes, or nicotine replacement therapy [NRT] products; n  = 4). These interviews were conducted in August 2017. The demographics of both data sets are presented in Table 3 . For reanalyzing the data, an initial codebook guided by the literature review data extraction categories was developed; however, new codes were created to complement these categories based on the thematic content analysis of the transcripts. The qualitative analysis software Quirkos [ 36 ] was used for the reanalysis.

Expert panel review

An expert panel consisting of five key opinion leaders (KOL) and five technical consultants was convened in August 28, 2018, in Neuchâtel, Switzerland. The KOLs were subject matter experts in the fields of nicotine and smoking cessation ( n  = 1), Patients Reported Outcomes (PRO) evaluation and scale development ( n  = 3), and health economics ( n  = 1). The consultants were experts on nicotine dependence ( n  = 1), psychometric validation ( n  = 2), market research ( n  = 1), and PRO development and validation ( n  = 1). The meeting followed an agenda and semi-structured discussion guide to facilitate conversations. First, the panel was presented with the principles underlying the tobacco harm reduction assessment strategy [ 4 ]. This session was followed by an open elicitation phase, during which two experienced moderators asked the panel to identify and discuss concepts related to health and functioning in people who use TNPs that different stakeholders might find important. Then, the panel was asked to review and respond to the concepts identified in the literature review and in the qualitative research reanalysis. These findings were discussed in depth to arrive at a consolidated preliminary conceptual framework. Each concept was presented, and the experts were asked to rank and agree on concepts to be included and how the concepts should be grouped by domains in the framework. In generating the framework, the project team and expert panel considered the themes and concepts identified under each of the categories from the scoping literature review, specific concepts from the secondary analyses of the qualitative data, and the expert panel meeting. The authors then synthesized and re-organized concepts emerging from the different preparatory phase activities under main health and functioning and conceptually-related domains. The participants also provided their input on the best strategies for planned qualitative studies to inform and support the development and validity of the proposed health and functioning measure.

The literature search identified 4761 articles. Figure  1 (flow diagram) depicts the results of the search and screening process. Titles and abstracts were screened by the rEM exercise until the machine learning algorithms predicted 97.7% relevant references; thus, 707 abstracts were not screened. After applying the inclusion/exclusion criteria to the remaining 4,054 abstracts, 281 were identified as part of the rEM exercise. After additional manual screening and review of the abstracts and articles against the inclusion/exclusion criteria, 90 full-text articles were included for data extraction [ 20 , 37 – 125 ]. Seven additional full-text articles were also included on the basis of a manual search [ 126 , 127 , 128 , 129 , 130 , 131 , 132 ]. Findings are summarized in Table 4 and a detailed description and data extracted from all the articles from the literature review is presented in Additional File 1 .

Flow diagram Sciome’s rapid Evidence Mapping (rEM) and manual screening processes of the scoping literature review

Fifty-six publications (56/97; 58%) presented data related to health signs and symptoms . These are grouped under five core areas: mental health and cognitive functioning (28/97; 29%); pain and physical trauma (6/97; 6%); respiratory, cardiovascular and inflammatory conditions (5/97; 5%); “other” health conditions , which included insomnia, liver disease, eye health, and hearing loss (5/97; 5%); and oral health (4/97; 4%). There were also eight publications related to the effects of smoking cessation on health signs and symptoms, mostly benefits of cessation but also including perceived dependence, addiction, and withdrawal symptoms (8/97; 8%). Overall, the burden and impact of cigarette smoking on both physical and mental health symptoms was negative and generally worse among people who smoke relative to those who do not smoke. On the other hand, quitting smoking was accompanied by improvements in general physical health and psychological wellbeing. However, in spite of the positive impact of quitting smoking, loss of moments of pleasure, struggle to manage stress, the social aspects of smoking, and withdrawal symptoms were seen as barriers to quitting.

The general health perceptions of various adults who use TNPs were reported in 18 of the 97 articles (18%), with 9 of them detailing the general health perceptions related to cigarettes and 9 being related to e-cigarettes and other TNPs. Perceptions were determined by questionnaires and focus groups for evaluating the health impacts, fear of diseases, harm to others and self, social impacts (both positive [e.g., inclusion and looking “cool”] and negative [e.g., stigma and exclusion]), and other reasons for taking up or considering/attempting smoking cessation.

Quality of life, health-related quality of life, and functional status was studied in 9 of the 97 included articles (9%). These studies mostly demonstrated with generic and specific QoL, HRQoL, or functional status questionnaires that cigarette smoking was associated with a worse quality of life and that smoking cessation often resulted in an improved quality of life. However, in some cases, the use of TNPs also reportedly enabled individuals to manage their levels of anxiety and improve some aspects of social engagement and functional status.

Individual, environmental and social characteristics were found to influence the decision to smoke and/or consider or attempt to quit smoking or switching to other TNPs, as reported in 8 (8%) and 11 (11%) of 97 publications, respectively. Some key characteristics and determinants of smoking behavior included low socioeconomic status, male sex, living alone, family, and close social environment, societal stigma, and local regulations.

Finally, 12 of the 97 publications (12%) were related to studies on biomarkers and biological endpoints in people who use TNPs and showed that smoking cigarettes negatively influenced cardiovascular, respiratory, oral, renal, stress, metabolic, and inflammatory-related biomarkers and physiological assessments.

The themes from this reanalysis are summarized below and organized on the basis of the narrative of the participants of their experiences.

Perceived negative impact of smoking

Other than health, the biggest and most salient reported negative impact of smoking was the perceived lack of control related to addiction and emotional health and wellbeing. Participants reported feeling that cigarette smoking was running their lives or “holding them hostage.” They reported that this perceived lack of a sense of control or willpower often led to feelings of weakness or a feeling that they were a “slave” to cigarettes. Many respondents reported smoking even when they did not necessarily want to and experiencing feelings of obsession and craving.

Perceived lack of control and addiction were also related to the activities of the participants throughout the day. People who smoke often reported altering their activities to smoke because of patterns of behavior or routine and the experienced need for a smoke. They reported that the “need for a smoke” sensation would cause them to leave work or social events early, not attend events if smoking was not allowed, interrupt what they were doing to smoke, and get up in the middle of the night.

Fear of withdrawal symptoms, with a strong emphasis on mental/emotional health, was also prominent among reported negative impacts of smoking. This fear was often reported as limiting the willingness of individuals to try to quit smoking or facilitating a return to prior smoking behavior. Individuals reported fearing the following symptoms they associated with withdrawal: mood swings and irritability, violent or aggressive behavior, inability to concentrate, anxiety, anger, and weight gain.

Perceived benefits of smoking

Several perceived benefits were identified that keep individuals smoking or using cigarettes. These included perceptions of enhanced cognitive functioning, relaxation, a way to take a break, use as a coping strategy, a social function, a weight management tool, the perception that it feels good, and being part of one’s identity. It is also important to note that the perceived benefits of smoking often outweighed the risks and the feeling of lack of control in the participant discussions. Even people who used to smoke noted they missed the relaxation and breaks they associated with smoking.

Recognition of symptoms/diseases related to smoking

Table 5 summarizes the negative symptoms and diseases related to smoking recognized by participants in both the focus groups and interviews. These were mostly related to six main body systems (cardiovascular, digestive, oral, neurological, reproductive, and respiratory).

Impacts on physical functioning

The participants noted how smoking impacts their physical functioning. In particular, they noted how their exercise capacity during running, playing sports, walking upstairs, and general physical activity was diminished. They also reported reduced stamina and endurance, decreased physical strength, and feeling tired more easily.

Effects on emotional health

The participants also described how smoking impacts their emotional health and wellbeing. People who smoke reported feelings of shame, guilt, weakness, and a lack of control or powerlessness. They also reported feelings of depression and anxiety associated with worry about health risks. Furthermore, the participants indicated that they experienced a fear of going to places where they could not smoke, being a bad role model for their children, and (in case of people who used to smoke) going back to smoking.

Positive and negative social impacts

Smoking was perceived to have both negative and positive impacts on the social lives of participants. Smoking impacted life negatively when it was not allowed in certain environments, such as in homes, at work, and in cars and airplanes. Stigma was also associated with smoking in an environment where peers and family members do not smoke, but it was also seen as a source of group identity within social networks that had a higher prevalence of smoking behaviors. Participants reported that smoking had some positive impacts on their social interaction, because it facilitated work breaks and increased communication with peers.

Reasons people decided to try to quit

Throughout the focus groups and interviews, individuals identified several reasons why they tried to quit smoking. These included: health, diagnosis of cancer (self, family, or friend), gum disease, pregnancy, hospital stay, worry that it will “kill me,” dislike of taste or odor, social reasons, change in surroundings (fewer smoking spaces), and price.

Reasons people do not like alternatives to cigarettes

The participants’ reasons for not liking alternatives to cigarettes (i.e., less harmful TNPs/RRPs) included perceptions that the alternatives did not work (i.e., the participants still had cravings and experienced withdrawal symptoms), made them feel or get ill (nausea and vomiting), were not “the same” as cigarettes in terms of the ritual, taste, or “feeling,” or were inconvenient/too big to carry.

The conclusions of the expert panel widely supported the findings of the literature review and the input from the reanalyzed focus groups and interviews. Some of the experts working in field of tobacco and nicotine provided additional insights based on their extensive experience with people who use TNPs; they highlighted the importance of the enjoyment of smoking for people who find it difficult to quit, the positive immediate benefits of quitting, and the smoking-related biomarkers that might be on a causal pathway between switching and changes in health and functioning status.

The following main areas were discussed and agreed during the meeting: (1) utility of use, referring to the perceived satisfaction and enjoyment of smoking (e.g., craving relief, weight control, and social affiliation); (2) signs and symptoms of withdrawal (e.g., anxiety, depression, and anger) and the positive immediate physical health effects of quitting smoking (e.g., better general and oral hygiene, less coughing, and improved exercise capacity); (3) functioning, including cognitive, physical, sexual, social, emotional, and role functioning; (4) worry associated with smoking and smoking-related diseases; (5) general health perceptions and quality of life; (6) association with smoking-related biomarkers that could be on the causal pathway between switching and changes in health and functioning; and (7) TNP use patterns and maintenance of switching to RRPs.

Generation of the preliminary conceptual framework

Triangulation of the findings from the literature review, qualitative input from people who use TNPs, and expert panel feedback helped generate a preliminary descriptive conceptual framework that includes the health and functioning and conceptually-related domains impacted by TNP use (Fig.  2 ).

Health and functioning conceptual framework related to tobacco and/or nicotine product use from the preparatory phase research findings

Four domains related to the future health and functioning measurement model for TNP use are indicated in grey rectangular boxes and include (moving down from proximal to distal parameters) physical health symptoms (e.g., oral and respiratory symptoms), general physical condition (e.g., appearance and hygiene), functioning (physical, sexual, cognitive, emotional, and social functioning), and general health perceptions, which will be the most distal measure of health and functioning. The preparatory phase research also identified six conceptually-related domains (in dashed rectangular boxes), which are not direct indicators of health status but might influence the impact of TNP use on health and functioning. These include attitudinal variables (worry about the health risks of using TNPs and perceived dependence/fear of withdrawal symptoms associated with quitting smoking), and utilitarian ones (perceived appeal, satisfaction, and benefits of TNP use). In addition, personal factors (e.g., sociodemographic) and environmental factors (e.g., peer/family influence, policies and regulations and sociocultural context) are also reflected in the conceptual framework as indirect indicators of health and functioning.

The framework further indicates that specific behavioral indicators (i.e., TNP use patterns over time) might influence any impact of TNP use on health and functioning. Whilst other causal and reciprocal relationships and hierarchies might exist within the domains, these are not explicitly characterized in this initial draft of the framework and will have to be tested with further empirical data. Finally, identified biomarkers of potential harm (in italics and dashed box) are also integrated in this conceptual framework as part of the conceptually-related domains, because they are on a causal pathway between TNP use and changes in health and functioning [ 133 , 134 ]. Biomarkers are not part of the measurement model that will be considered for a new self-report measure; however, because they are the most proximal parameters to health and functioning, they will be assessed independently as appropriate endpoints by objective clinical or biological analyses.

Triangulation of published literature, reanalysis of qualitative data, and expert opinion helped develop the presented preliminary conceptual framework as the foundation for a new measure to assess the impact of TNPs on self-reported health and functioning. This is essential for identifying relevant concepts and understanding what is important to measure in people who use TNPs. The findings reveal the importance of not only the perceived impacts of TNP use on physical health and physical functioning, but also on aspects of mental health and social interactions and functioning, and general perceptions of health and health-related quality of life.

For the literature review, the WHO ICF [ 28 ] and Wilson and Cleary model [ 29 , 30 ] served as useful guides to develop categories for data abstraction. The scoping literature review yielded 97 articles on TNP use and the relationship to health, perceptions of health, social and individual functioning, and quality of life. Overall, most studies had focused on the known negative effects of cigarette smoking (e.g., mental, respiratory, and oral health) and the rationale and motivation to quit smoking. The WHO ICF and Wilson and Clearly models were not always sufficient for identifying the breadth of relevant concepts, especially from the perspective of TNP use. Development of new codes for the reanalysis of existing qualitative data allowed for the development, extension, and exploration of the topic and provided valuable insights reported in the qualitative data reanalysis, such as the perceived benefits/satisfaction from cigarette smoking, and the rationale for quitting smoking or switching to an RRP. The findings show how this manner of secondary analysis can be valuable in health-related fields where the topic is broad and an existing body of knowledge can contribute by offering a different perspective [ 135 ].

The presentation of the preliminary conceptual framework from this preparatory phase is specific to TNP use and marks a slight departure from the established norms and characterization of the variables typically observed in existing generic health and functioning and health-related quality of life models, such as the WHO ICF and Wilson and Clearly models. Notably, specific hypothesized relationships and the hierarchy between domains are not explicitly characterized in the current draft of the framework. The framework provided an exploratory representation of the current findings to reflect a measurement instrument in people who use TNPs that would ideally be able to assess and demonstrate improvements in self-reported health and functioning status, stability of perceived positive aspects of using TNPs, and no worsening in key areas of physical and emotional health and functioning upon switching to RRPs. Nevertheless, the framework could still undergo further refinement to support the development and validation of a new measure and to further characterize and test the relationships and hierarchies between domains.

This work is not without limitations. For the scoping literature review, among the reviewed articles, not many reported on the use of e-cigarettes and other alternative tobacco or nicotine-delivery devices, because most studies had focused exclusively on cigarettes. It is possible that concepts associated with health and functioning that are relevant to other TNPs were not identified. This is most likely the consequence of the large number of publications related to cigarette use. Some concepts might also have been missed, given the large evidence base on health and functioning-related themes and concepts. However, this was also not a systematic literature search; a scoping review is generally broader than a systematic review in terms of the former having a less-defined research question, broader inclusion and exclusion criteria, and no systematic appraisal of study quality [ 26 ]. Nevertheless, the present scoping review methodology provides a lens on the overall evidence base, and regular updates on the search—specifically related to RRPs and novel TNPs and their health and functioning impacts—could be considered for fully understanding the evolving state of the art in this context. The reanalysis of existing qualitative data also has limitations related to data fit and completeness of preexisting data [ 136 ]. The insights collected from these reanalyzed studies were originally for a different purpose several years prior to the present research, and this might not completely and accurately reflect the objectives of the new project.

Considering the findings of the current research, the development of a health and functioning measure can continue to follow the FDA’s Guidance on PRO measures. As specified within the guideline, gaining input directly from the intended use populations through concept elicitation is a critical activity for ensuring content validity during the development of any new self-reported measure [ 137 ]. Continuous engagement with an expert panel can also support the refinement of the conceptual framework as well as the development of the draft and final measure.

The goal of this research was to identify from varied research activities key concepts and aspects of health and functioning and related changes associated with the use of TNPs. The resulting preliminary conceptual framework provides the basis for informing future research to further understand health and functioning concepts important to measure in individual who switch to RRPs and to develop a new self-report measure to assess this from the consumers’ perspective.

Availability of data and materials

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Abbreviations

Assessment of Behavioral OUtcomes related to Tobacco and Nicotine Products Toolbox

Chronic obstructive pulmonary disease

Food and Drug Administration

Health-related quality of life

International Classification of Functioning, Disability and Health

• Modified risk tobacco products

Nicotine replacement therapy

Patient-Reported Outcomes

Quality of life

Reduced-risk products

Rapid Evidence Mapping

• Tobacco and/or nicotine products

United Kingdom

United States

36-Item Short-Form Health Survey

World Health Organization

U.S. Department of Health and Human Services. The health consequences of smoking: 50 years of progress. A report of the surgeon general. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2014.

Google Scholar  

Danaei G, Ding EL, Mozaffarian D, Taylor B, Rehm J, Murray CJL, et al. The preventable causes of death in the United States: comparative risk assessment of dietary, lifestyle, and metabolic risk factors. PLoS Med. 2009;6(4):e1000058.

Article   PubMed   PubMed Central   Google Scholar  

Bilano V, Gilmour S, Moffiet T, d’Espaignet ET, Stevens GA, Commar A, et al. Global trends and projections for tobacco use, 1990–2025: an analysis of smoking indicators from the WHO Comprehensive Information Systems for Tobacco Control. Lancet. 2015;385(9972):966–76.

Article   PubMed   Google Scholar  

Zeller M, Hatsukami D. The strategic dialogue on tobacco harm reduction: a vision and blueprint for action in the US. Tob Control. 2009;18(4):324–32.

Food and Drug Administration. Modified risk orders 2020. https://www.fda.gov/tobacco-products/advertising-and-promotion/modified-risk-orders .

Food and Drug Administration. Modified risk tobacco product applications: draft guidance for industry 2012. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/modified-risk-tobacco-product-applications .

Institute of Medicine. Scientific standards for studies on modified risk tobacco products. Washington, DC: The National Academies Press; 2012. p. 370.

Stratton K, Kwan LY, Eaton DL, editors. National Academies of Sciences Engineering and Medicine. Public health consequences of e-cigarettes. Washington, DC: The National Academies Press; 2018. p. 774.

Lappas AS, Tzortzi AS, Konstantinidi EM, Teloniatis SI, Tzavara CK, Gennimata SA, et al. Short-term respiratory effects of e-cigarettes in healthy individuals and smokers with asthma. Respirology. 2018;23(3):291–7.

Meo SA, Ansary MA, Barayan FR, Almusallam AS, Almehaid AM, Alarifi NS, et al. Electronic cigarettes: impact on lung function and fractional exhaled nitric oxide among healthy adults. Am J Men’s Health. 2019;13(1):1557988318806073.

Article   Google Scholar  

Polosa R, Cibella F, Caponnetto P, Maglia M, Prosperini U, Russo C, et al. Health impact of E-cigarettes: a prospective 3.5-year study of regular daily users who have never smoked. Sci Rep. 2017;7(1):13825.

Article   PubMed   PubMed Central   CAS   Google Scholar  

Sharman A, Zhussupov B, Sharman D, Kim I, Yerenchina E. Lung function in users of a smoke-free electronic device with HeatSticks (iQOS) versus smokers of conventional cigarettes: protocol for a longitudinal cohort observational study. JMIR Res Protoc. 2018;7(11):e10006.

Skotsimara G, Antonopoulos AS, Oikonomou E, Siasos G, Ioakeimidis N, Tsalamandris S, et al. Cardiovascular effects of electronic cigarettes: a systematic review and meta-analysis. Eur J Prev Cardiol. 2019;26(11):1219–28.

Wang JB, Olgin JE, Nah G, Vittinghoff E, Cataldo JK, Pletcher MJ, et al. Cigarette and e-cigarette dual use and risk of cardiopulmonary symptoms in the Health eHeart Study. PLoS ONE. 2018;13(7):e0198681.

Ware JE Jr, Gandek B, Kulasekaran A, Guyer R. Evaluation of smoking-specific and generic quality of life measures in current and former smokers in Germany and the United States. Health Qual Life Outcomes. 2015;13:128.

Frendl DM, Ware JE Jr. Patient-reported functional health and well-being outcomes with drug therapy: a systematic review of randomized trials using the SF-36 health survey. Med Care. 2014;52(5):439–45.

Goldenberg M, Danovitch I, IsHak WW. Quality of life and smoking. Am J Addict. 2014;23(6):540–62.

Olufade AO, Shaw JW, Foster SA, Leischow SJ, Hays RD, Coons SJ. Development of the smoking cessation quality of life questionnaire. Clin Ther. 1999;21(12):2113–30.

Article   CAS   PubMed   Google Scholar  

Sarna L, Bialous SA, Cooley ME, Jun HJ, Feskanich D. Impact of smoking and smoking cessation on health-related quality of life in women in the Nurses’ Health Study. Qual Life Res. 2008;17(10):1217–27.

Kulasekaran A, Proctor C, Papadopoulou E, Shepperd CJ, Guyer R, Gandek B, et al. Preliminary evaluation of a new german translated tobacco quality of life impact tool to discriminate between healthy current and former smokers and to explore the effect of switching smokers to a reduced toxicant prototype cigarette. Nicotine Tob Res. 2015;17(12):1456–64.

Edelen MO. The PROMIS smoking assessment toolkit–background and introduction to supplement. Nicotine Tob Res. 2014;16(Suppl 3):S170–4.

Shaw JW, Coons SJ, Foster SA, Leischow SJ, Hays RD. Responsiveness of the Smoking Cessation Quality of Life (SCQoL) questionnaire. Clin Ther. 2001;23(6):957–69.

Chrea C, Acquadro C, Afolalu EF, Spies E, Salzberger T, Abetz-Webb L, et al. Developing fit-for-purpose self-report instruments for assessing consumer responses to tobacco and nicotine products: the ABOUT Toolbox initiative. F1000Res. 2018;7:1878.

Food and Drug Administration. Patient-reported outcome measures: use in medical product development to support labeling claims 2009. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/patient-reported-outcome-measures-use-medical-product-development-support-labeling-claims .

Food and Drug Administration. Patient-focused drug development: methods to identify what is important to patients guidance for industry, Food and Drug Administration Staff, and Other Stakeholders 2019. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/patient-focused-drug-development-methods-identify-what-important-patients-guidance-industry-food-and .

Munn Z, Peters MDJ, Stern C, Tufanaru C, McArthur A, Aromataris E. Systematic review or scoping review? Guidance for authors when choosing between a systematic or scoping review approach. BMC Med Res Methodol. 2018;18(1):143.

Lam J, Howard BE, Thayer K, Shah RR. Low-calorie sweeteners and health outcomes: a demonstration of rapid evidence mapping (rEM). Environ Int. 2019;123:451–8.

World Health Organisation. International classification of functioning, disability and health (ICF) 2001. https://www.who.int/classifications/icf/en/ .

Ferrans CE, Zerwic JJ, Wilbur JE, Larson JL. Conceptual model of health-related quality of life. J Nurs Scholarsh. 2005;37(4):336–42.

Wilson IB, Cleary PD. Linking clinical variables with health-related quality of life. A conceptual model of patient outcomes. JAMA. 1995;273(1):59–65.

Karimi M, Brazier J. Health, health-related quality of life, and quality of life: what is the difference? Pharmacoeconomics. 2016;34(7):645–9.

Moons P. Why call it health-related quality of life when you mean perceived health status? Eur J Cardiovasc Nurs. 2004;3(4):275–7.

Cano S, Chrea C, Salzberger T, Alfieri T, Emilien G, Mainy N, et al. Development and validation of a new instrument to measure perceived risks associated with the use of tobacco and nicotine-containing products. Health Qual Life Outcomes. 2018;16(1):192.

Salzberger T, Chrea C, Cano SJ, Martin M, Atkison M, Emilien G, et al. Perceived risks associated with the use of tobacco and nicotine-containing products: findings from qualitative research. Tobacco Sci Technol. 2017;50:32–42.

Chrea C, Salzberger T, Abetz-Webb L, Afolalu EF, Cano S, Rose J, et al. PRM183—development of a fit-for-purpose tobacco and nicotine products dependence instrument. Barcelona: ISPOR Europe; 2018.

Book   Google Scholar  

Quirkos 2.3.1 [Computer Software] 2020. https://www.quirkos.com .

Abu-Helalah MA, Alshraideh HA, Al-Serhan AA, Nesheiwat AI, Da’na M, Al-Nawafleh A. Epidemiology, attitudes and perceptions toward cigarettes and hookah smoking amongst adults in Jordan. Environ Health Prev Med. 2015;20(6):422–33.

Akturk E, Yağmur J, Açıkgöz N, Ermi N, Cansel M, Karaku Y, et al. Assessment of atrial conduction time by tissue Doppler echocardiography and P-wave dispersion in smokers. J Interv Card Electrophysiol. 2012;34(2):247–53.

Aubin HJ, Peiffer G, Stoebner-Delbarre A, Vicaut E, Jeanpetit Y, Solesse A, et al. The French Observational Cohort of Usual Smokers (FOCUS) cohort: French smokers perceptions and attitudes towards smoking cessation. BMC Public Health. 2010;10(100):1–8.

Becoña E, Vázquez MI, Míguez MC, Fernández del Río E, López-Durán A, Martínez Ú, et al. Smoking habit profile and health-related quality of life. Psicothema. 2013;25(4):421–6.

PubMed   Google Scholar  

Bennasar Veny M, Pericas Beltrán J, González Torrente S, Segui González P, Aguiló Pons A, Tauler RP. Self-perceived factors associated with smoking cessation among primary health care nurses: a qualitative study. Rev Latino-Am Enfermagem. 2011;19(6):1437–44.

Bommelé J, Schoenmakers TM, Kleinjan M, van Straaten B, Wits E, Snelleman M, et al. Perceived pros and cons of smoking and quitting in hard-core smokers: a focus group study. BMC Public Health. 2014;14:175–85.

Borland R, Yong HH, O’Connor RJ, Hyland A, Thompson ME. The reliability and predictive validity of the Heaviness of Smoking Index and its two components: findings from the International Tobacco Control Four Country study. Nicotine Tob Res. 2010;12(Suppl 1):S45-50.

Bot M, Vink J, Milaneschi Y, Smit JH, Kluft C, Neuteboom J, et al. Plasma cotinine levels in cigarette smokers: impact of mental health and other correlates. Eur Addict Res. 2014;20(4):183–91.

Brody AL, Olmstead RE, Abrams AL, Costello MR, Khan A, Kozman D, et al. Effect of a history of major depressive disorder on smoking-induced dopamine release. Biol Psychiatry. 2009;66(9):898–901.

Article   CAS   PubMed   PubMed Central   Google Scholar  

Brook DW, Rubenstone E, Zhang C, Brook JS. Trajectories of cigarette smoking in adulthood predict insomnia among women in late mid-life. Sleep Med. 2012;13(9):1130–7.

Brotman RM, He X, Gajer P, Fadrosh D, Sharma E, Mongodin EF, et al. Association between cigarette smoking and the vaginal microbiota: a pilot study. BMC Infect Dis. 2014;14:471–82.

Brown-Johnson CG, Cataldo JK, Orozco N, Lisha NE, Hickman NJ 3rd, Prochaska JJ. Validity and reliability of the Internalized Stigma of Smoking Inventory: an exploration of shame, isolation, and discrimination in smokers with mental health diagnoses. Am J Addict. 2015;24(5):410–8.

Bush T, Hsu C, Levine MD, Magnusson B, Miles L. Weight gain and smoking: perceptions and experiences of obese quitline participants. BMC Public Health. 2014;14:1229.

Caldirola D, Cavedini P, Riva A, Di Chiaro NV, Perna G. Cigarette smoking has no pro-cognitive effect in subjects with obsessive-compulsive disorder: a preliminary study. Psychiatr Danub. 2016;28(1):86–90.

Caldirola D, Daccò S, Grassi M, Citterio A, Menotti R, Cavedini P, et al. Effects of cigarette smoking on neuropsychological performance in mood disorders: a comparison between smoking and nonsmoking inpatients. J Clin Psychiatry. 2013;74(2):e130–6.

Carpenter MJ, Gray KM. A pilot randomized study of smokeless tobacco use among smokers not interested in quitting: changes in smoking behavior and readiness to quit. Nicotine Tob Res. 2010;12(2):136–43.

Croucher R, Haque MF, Kassim S. Oral pain before and after smokeless tobacco cessation in U.K.-resident Bangladeshi women: cross-sectional analyses. Nicotine Tobacco Res. 2013;15(5):896–903.

Article   CAS   Google Scholar  

Depp CA, Bowie CR, Mausbach BT, Wolyniec P, Thornquist MH, Luke JR, et al. Current smoking is associated with worse cognitive and adaptive functioning in serious mental illness. Acta Psychiatr Scand. 2015;131(5):333–41.

Ditre JW, Zale EL, Heckman BW, Hendricks PS. A measure of perceived pain and tobacco smoking interrelations: pilot validation of the pain and smoking inventory. Cogn Behav Ther. 2017;46(6):339–51.

Doiron M, Dupré N, Langlois M, Provencher P, Simard M. Smoking history is associated to cognitive impairment in Parkinson’s disease. Aging Ment Health. 2017;21(3):322–6.

Gonzalez A, Zvolensky MJ, Vujanovic AA, Leyro TM, Marshall EC. An evaluation of anxiety sensitivity, emotional dysregulation, and negative affectivity among daily cigarette smokers: relation to smoking motives and barriers to quitting. J Psychiatr Res. 2008;43(2):138–47.

Grogan S, Fry G, Gough B, Conner M. Smoking to stay thin or giving up to save face? Young men and women talk about appearance concerns and smoking. Br J Health Psychol. 2009;14(1):175–86.

Hawari FI, Obeidat NA, Ghonimat IM, Ayub HS, Dawahreh SS. The effect of habitual waterpipe tobacco smoking on pulmonary function and exercise capacity in young healthy males: A pilot study. Respir Med. 2017;122:71–5.

Heffernan TM, O’Neill TS, Moss M. Smoking-related prospective memory deficits in a real-world task. Drug Alcohol Depend. 2012;120(1–3):1–6.

Highland KB, McChargue DE. Stress-induced cardiovascular reactivity among African American smokers. Am J Health Behav. 2011;35(1):51–9.

Holley AL, Law EF, Tham SW, Myaing M, Noonan C, Strachan E, et al. Current smoking as a predictor of chronic musculoskeletal pain in young adult twins. J Pain. 2013;14(10):1131–9.

Ichikawa Y, Kitagawa K, Kato S, Dohi K, Hirano T, Ito M, et al. Altered coronary endothelial function in young smokers detected by magnetic resonance assessment of myocardial blood flow during the cold pressor test. Int J Cardiovasc Imaging. 2014;30:73–80.

Javed F, Abduljabbar T, Vohra F, Malmstrom H, Rahman I, Romanos GE. Comparison of periodontal parameters and self-perceived oral symptoms among cigarette smokers, individuals vaping electronic cigarettes, and never-smokers. J Periodontol. 2017;88(10):1059–65.

Johnson AL, McLeish AC. Differences in panic psychopathology between smokers with and without asthma. Psychol Health Med. 2017;22(1):110–20.

Kahler CW, Daughters SB, Leventhal AM, Rogers ML, Clark MA, Colby SM, et al. Personality, psychiatric disorders, and smoking in middle-aged adults. Nicotine Tob Res. 2009;11(7):833–41.

Karadoğan D, Önal Ö, Şahin D, Yazıcı S, Kanbay Y. Evaluation of school teachers’ sociodemographic characteristics and quality of life according to their cigarette smoking status: a cross-sectional study from eastern Black Sea region of Turkey. Tuberkuloz ve toraks. 2017;65(1):18–24.

Kim J, Gall SL, Dewey HM, Macdonell RA, Sturm JW, Thrift AG. Baseline smoking status and the long-term risk of death or nonfatal vascular event in people with stroke: a 10-year survival analysis. Stroke. 2012;43(12):3173–8.

Kim J, Lee S. Using focus group interviews to analyze the behavior of users of new types of tobacco products. J Prev Med Public Health. 2017;50(5):336–46.

Kralikova E, Novak J, West O, Kmetova A, Hajek P. Do e-cigarettes have the potential to compete with conventional cigarettes?: a survey of conventional cigarette smokers’ experiences with e-cigarettes. Chest. 2013;144(5):1609–14.

Lao XQ, Jiang CQ, Zhang WS, Adab P, Lam TH, Cheng KK, et al. Smoking, smoking cessation and inflammatory markers in older Chinese men: The Guangzhou Biobank Cohort Study. Atherosclerosis. 2009;203(1):304–10.

Lappan S, Thorne CB, Long D, Hendricks PS. Longitudinal and reciprocal relationships between psychological well-being and smoking. Nicotine Tob Res. 2018;22(1):18–23.

Article   PubMed Central   Google Scholar  

Lynch ME, Johnson KC, Kable JA, Carroll J, Coles CD. Smoking in pregnancy and parenting stress: maternal psychological symptoms and socioeconomic status as potential mediating variables. Nicotine Tob Res. 2011;13(7):532–9.

Lyvers M, Carlopio C, Honours VB, Edwards MS. Mood, mood regulation, and frontal systems functioning in current smokers, long-term abstinent ex-smokers, and never-smokers. J Psychoactive Drugs. 2014;46(2):133–9.

Martin LM, Sayette MA. A review of the effects of nicotine on social functioning. Exp Clin Psychopharmacol. 2018;26(5):425–39.

Martinasek MP, McDermott RJ, Bryant CA. Antecedents of university students’ hookah smoking intention. Am J Health Behav. 2013;37(5):599–609.

Mattey DL, Dawson SR, Healey EL, Packham JC. Relationship between smoking and patient-reported measures of disease outcome in ankylosing spondylitis. J Rheumatol. 2011;38(12):2608–15.

McCann SJ. Subjective well-being, personality, demographic variables, and American state differences in smoking prevalence. Nicotine Tob Res. 2010;12(9):895–904.

McDonald EA, Ling PM. One of several “toys” for smoking: young adult experiences with electronic cigarettes in New York City. Tob Control. 2015;24(6):588–93.

McLeish AC, Zvolensky MJ, Del Ben KS, Burke RS. Anxiety sensitivity as a moderator of the association between smoking rate and panic-relevant symptoms among a community sample of middle-aged adult daily smokers. Am J Addict. 2009;18(1):93–9.

Melis M, Lobo SL, Ceneviz C, Ruparelia UN, Zawawi KH, Chandwani BP, et al. Effect of cigarette smoking on pain intensity of TMD patients: a pilot study. Cranio. 2010;28(3):187–92.

Memon A, Barber J, Rumsby E, Parker S, Mohebati L, de Visser RO, et al. What factors are important in smoking cessation and relapse in women from deprived communities? A qualitative study in Southeast England. Public Health. 2016;134:39–45.

Mendiondo MS, Alexander LA, Crawford T. Health profile differences for menthol and non-menthol smokers: findings from the National Health Interview Survey. Addiction. 2010;105(Suppl 1):124–40.

Miyatake N, Numata T, Nishii K, Sakano N, Suzue T, Hirao T, et al. Influence of cigarette smoking on estimated glomerular filtration rate (eGFR) in Japanese male workers. Acta Med Okayama. 2010;64:385–90.

CAS   PubMed   Google Scholar  

Mohammadi S, Mazhari MM, Mehrparvar AH, Attarchi MS. Cigarette smoking and occupational noise-induced hearing loss. Eur J Public Health. 2009;20(4):452–5.

Mohammadnezhad M, Tsourtos G, Wilson C, Ratcliffe J, Ward P. “I have never experienced any problem with my health. So far, it hasn’t been harmful”: older Greek-Australian smokers’ views on smoking: a qualitative study. BMC Public Health. 2015;15:304–15.

Morris MC, Mielock AS, Rao U. Salivary stress biomarkers of recent nicotine use and dependence. Am J Drug Alcohol Abuse. 2016;42(6):640–8.

Muir S, Marshall B. Changes in health perceptions of male prisoners following a smoking cessation program. J Correct Health Care. 2016;22(3):247–56.

Nelson JP, Pederson LL, Lewis J. Tobacco use in the Army: illuminating patterns, practices, and options for treatment. Mil Med. 2009;174(2):162–9.

Nemeth JM, Liu ST, Klein EG, Ferketich AK, Kwan MP, Wewers ME. Factors influencing smokeless tobacco use in rural Ohio Appalachia. J Community Health. 2012;37(6):1208–17.

Nikcevic AV, Spada MM. Metacognitions about smoking: a preliminary investigation. Clin Psychol Psychother. 2010;17(6):536–42.

Nunes SO, Vargas HO, Brum J, Prado E, Vargas MM, de Castro MR, et al. A comparison of inflammatory markers in depressed and nondepressed smokers. Nord J Psychiatry. 2012;14:540–6.

CAS   Google Scholar  

Oh DL, Heck JE, Dresler C, Allwright S, Haglund M, Del Mazo SS, et al. Determinants of smoking initiation among women in five European countries: a cross-sectional survey. BMC Public Health. 2010;10(74):1–11.

Pasco JA, Williams LJ, Jacka FN, Ng F, Henry MJ, Nicholson GC, et al. Tobacco smoking as a risk factor for major depressive disorder: population-based study. Br J Psychiatry. 2008;193(4):322–6.

Pina JA, Namba MD, Leyrer-Jackson JM, Cabrera-Brown G, Gipson CD. Social influences on nicotine-related behaviors. Int Rev Neurobiol. 2018;140:1–32.

Poole-Di Salvo E, Liu YH, Brenner S, Weitzman M. Adult household smoking is associated with increased child emotional and behavioral problems. J Dev Behav Pediatr. 2010;31(2):107–15.

Rahman MA, Mahmood MA, Spurrier N, Rahman M, Choudhury SR, Leeder S. Why do Bangladeshi people use smokeless tobacco products? Asia-Pacific J Public Health. 2015;27(2):NP2197–209.

Rollini F, Franchi F, Cho JR, Degroat C, Bhatti M, Ferrante E, et al. Cigarette smoking and antiplatelet effects of aspirin monotherapy versus clopidogrel monotherapy in patients with atherosclerotic disease: results of a prospective pharmacodynamic study. J Cardiovasc Transl Res. 2014;7(1):53–63.

Romijnders K, van Osch L, de Vries H, Talhout R. Perceptions and reasons regarding e-cigarette use among users and non-users: a narrative literature review. Int J Environ Res Public Health. 2018;15(6):1190.

Rooke C, Cunningham-Burley S, Amos A. Smokers’ and ex-smokers’ understanding of electronic cigarettes: a qualitative study. Tob Control. 2016;25:e60–6.

Roth B, Bengtsson M, Ohlsson B. Diarrhoea is not the only symptom that needs to be treated in patients with microscopic colitis. Eur J Intern Med. 2013;24(6):573–8.

Sales MP, Oliveira MI, Mattos IM, Viana CM, Pereira ED. The impact of smoking cessation on patient quality of life. J Bras Pneumol. 2009;35(5):436–41.

Schane RE, Prochaska JJ, Glantz SA. Counseling nondaily smokers about secondhand smoke as a cessation message: a pilot randomized trial. Nicotine Tob Res. 2013;15(2):334–42.

Schnoll RA, Goren A, Annunziata K, Suaya JA. The prevalence, predictors and associated health outcomes of high nicotine dependence using three measures among US smokers. Addiction. 2013;108(11):1989–2000.

Segarra R, Zabala A, Eguíluz JI, Ojeda N, Elizagarate E, Sánchez P, et al. Cognitive performance and smoking in first-episode psychosis: the self-medication hypothesis. Eur Arch Psychiatry Clin Neurosci. 2011;261(4):241–50.

Sherratt FC, Newson L, Marcus MW, Field JK, Robinson J. Perceptions towards electronic cigarettes for smoking cessation among Stop Smoking Service users. Br J Health Psychol. 2015;21(2):421–33.

Soneji SS, Sung HY, Primack BA, Pierce JP, Sargent JD. Quantifying population-level health benefits and harms of e-cigarette use in the United States. PLoS ONE. 2018;13(3):e0193328.

Talati A, Wickramaratne PJ, Keyes KM, Hasin DS, Levin FR, Weissman MM. Smoking and psychopathology increasingly associated in recent birth cohorts. Drug Alcohol Depend. 2013;133(2):724–32.

Tan QH. Smoking spaces as enabling spaces of wellbeing. Health Place. 2013;24:173–82.

Tatullo M, Gentile S, Paduano F, Santacroce L, Marrelli M. Crosstalk between oral and general health status in e-smokers. Medicine. 2016;95(49):e5589.

Taylor G, McNeill A, Girling A, Farley A, Lindson-Hawley N, Aveyard P. Change in mental health after smoking cessation: systematic review and meta-analysis. BMJ. 2014;348:g1151-g.

Thompson TP, Greaves CJ, Ayres R, Aveyard P, Warren FC, Byng R, et al. An exploratory analysis of the smoking and physical activity outcomes from a pilot randomized controlled trial of an exercise assisted reduction to stop smoking intervention in disadvantaged groups. Nicotine Tob Res. 2016;18(3):289–97.

Torigian DA, Green-McKenzie J, Liu X, Shofer FS, Werner T, Smith CE, et al. A study of the feasibility of FDG-PET/CT to systematically detect and quantify differential metabolic effects of chronic tobacco use in organs of the whole body—a prospective pilot study. Acad Radiol. 2017;24:930–40.

Ulvik A, Ebbing M, Hustad S, Midttun Ø, Nygård O, Vollset SE, et al. Long- and short-term effects of tobacco smoking on circulating concentrations of B vitamins. Clin Chem. 2010;56(5):755–63.

Vidrine JI, Businelle MS, Cinciripini P, Li Y, Marcus MT, Waters AJ, et al. Associations of mindfulness with nicotine dependence, withdrawal, and agency. Subst Abus. 2009;30(4):318–27.

Volkman JE, DeRycke EC, Driscoll MA, Becker WC, Brandt CA, Mattocks KM, et al. Smoking status and pain intensity among OEF/OIF/OND veterans. Pain Med. 2015;16(9):1690–6.

Vujanovic AA, Marshall-Berenz EC, Beckham JC, Bernstein A, Zvolensky MJ. Posttraumatic stress symptoms and cigarette deprivation in the prediction of anxious responding among trauma-exposed smokers: a laboratory test. Nicotine Tob Res. 2010;12(11):1080–8.

Wadsworth E, Neale J, McNeill A, Hitchman SC. How and why do smokers start using e-cigarettes? Qualitative study of Vapers in London, UK. Int J Environ Res Public Health. 2016;13(7):661–74.

Article   PubMed Central   CAS   Google Scholar  

Xie X, Dijkstra AE, Vonk JM, Oudkerk M, Vliegenthart R, Groen HJ. Chronic respiratory symptoms associated with airway wall thickening measured by thin-slice low-dose CT. Am J Roentgenol. 2014;203(4):W383–90.

Yang T, Shiffman S, Rockett IR, Cui X, Cao R. Nicotine dependence among Chinese city dwellers: a population-based cross-sectional study. Nicotine Tob Res. 2011;13(7):556–64.

Yu A, Cai X, Zhang Z, Shi H, Liu D, Zhang P, et al. Effect of nicotine dependence on opioid requirements of patients after thoracic surgery. Acta Anaesthesiol Scand. 2015;59(1):115–22.

Zhang X, Kahende J, Fan AZ, Barker L, Thompson TJ, Mokdad AH, et al. Smoking and visual impairment among older adults with age-related eye diseases. Prev Chronic Dis. 2011;8(4):A84.

PubMed   PubMed Central   Google Scholar  

Zhao L, Xu L, Lai Y, Che C, Zhou Y. Temporal changes of smoking status and motivation in Chinese patients with hepatitis B: relationship with anxiety and depression. J Clin Nurs. 2012;21(15–16):2193–201.

Zorlu N, Cropley VL, Zorlu PK, Delibas DH, Adibelli ZH, Baskin EP, et al. Effects of cigarette smoking on cortical thickness in major depressive disorder. J Psychiatr Res. 2017;84:1–8.

Miyatake N, Numata T, Nishii K, Sakano N, Suzue T, Hirao T, et al. Relation between cigarette smoking and ventilatory threshold in the Japanese. Environ Health Prev Med. 2011;16(3):185–90.

Bjørngaard JH, Gunnell D, Elvestad MB, Davey Smith G, Skorpen F, Krokan H, et al. The causal role of smoking in anxiety and depression: a Mendelian randomization analysis of the HUNT study. Psychol Med. 2012;43(4):711–9.

Chen SC, Chen HF, Peng HL, Lee LY, Chiang TY, Chiu HC. Psychometric testing of the Chinese-Version Glover-Nilsson Smoking Behavioral Questionnaire (GN-SBQ-C) for the identification of nicotine dependence in adult smokers in Taiwan. Int J Behav Med. 2017;24(2):272–9.

Doran N, Spring B, McChargue D, Pergadia M, Richmond M. Impulsivity and smoking relapse. Nicotine Tob Res. 2004;6:641–7.

Haller CS, Etter JF, Courvoisier DS. Trajectories in cigarette dependence as a function of anxiety: a multilevel analysis. Drug Alcohol Depend. 2014;139:115–20.

Jamal M, Van der Does AJW, Cuijpers P, Penninx BW. Association of smoking and nicotine dependence with severity and course of symptoms in patients with depressive or anxiety disorder. Drug Alcohol Depend. 2012;126(1–2):138–46.

Krishnan-Sarin S, Reynolds B, Duhig A, Smith A, Liss T, McFetridge A, et al. Behavioral impulsivity predicts treatment outcome in a smoking cessation program for adolescent smokers. Drug Alcohol Depend. 2007;88:79–82.

Tidey JW, Pacek LR, Koopmeiners JS, Vandrey R, Nardone N, Drobes DJ, et al. Effects of 6-week use of reduced-nicotine content cigarettes in smokers with and without elevated depressive symptoms. Nicotine Tob Res. 2017;19(Suppl 1):59–67.

Chang CM, Cheng YC, Cho TM, Mishina EV, Del Valle-Pinero AY, van Bemmel DM, et al. Biomarkers of potential harm: summary of an FDA-sponsored public workshop. Nicotine Tob Res. 2019;21(1):3–13.

Haziza C, de La Bourdonnaye G, Donelli A, Skiada D, Poux V, Weitkunat R, et al. Favorable changes in biomarkers of potential harm to reduce the adverse health effects of smoking in smokers switching to the menthol tobacco heating system 2.2 for three months (part 2). Nicotine Tob Res. 2019;22(4):549–59.

Bishop L, Kuula-Luumi A. Revisiting qualitative data reuse: a decade on. SAGE Open. 2017;7(1):1–15.

Sherif V. Evaluating preexisting qualitative research data for secondary analysis. In: Forum: qualitative social research, vol. 19, no. 2. 2018.

Patrick D, Burke L, Gwaltney C, Leidy N, Martin M, Molsen E, et al. Content validity—establishing and reporting the evidence in newly developed patient-reported outcomes (PRO) instruments for medical product evaluation: ISPOR PRO good research practices task force report: part 1—eliciting concepts for a new PRO instrument. Value Health. 2011;14(8):967–77.

Download references

Acknowledgements

We thank the team at Sciome LLC for their assistance and contribution to the literature review. We thank Vivienne Law and David Floyd for their contributions to the literature review, reanalysis of qualitative data, and assistance with review of the draft manuscript. We thank Catherine Acquadro for her review of the draft manuscript. We also thank John Ware, Jed Rose, Ashley Slagle, Donald Patrick, Karl Fagerström, Stefan Cano, and Thomas Salzberger for their input and review.

Philip Morris International is the sole source of funding and sponsor of this research.

Author information

Authors and affiliations.

PMI R&D, Philip Morris Product S.A., Quai Jeanrenaud 5, 2000, Neuchâtel, Switzerland

Esther F. Afolalu, Erica Spies, Agnes Bacso, Emilie Clerc, Sophie Gallot & Christelle Chrea

Patient-Centered Outcomes Assessments Ltd., 1 Springbank, Bollington, Macclesfield, Cheshire, SK10 5LQ, UK

Linda Abetz-Webb

You can also search for this author in PubMed   Google Scholar

Contributions

EA, ES and CC performed conceptualization. EA, ES and LA-W performed methodology. EA, ES, SG, EC and LA-W were involved in the investigation. EA and ES were involved in writing—original draft. EA, EC, LA-W and CC were involved in writing—review & editing. EA performed visualization. ES and CC performed supervision. AB, EC and SG were involved in data curation. AB and EC were involved in project administration. LA-W performed formal analysis. CC was involved in funding acquisition. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Esther F. Afolalu .

Ethics declarations

Ethics approval and consent to participate.

Not applicable.

Consent for publication

Competing interests.

Esther F. Afolalu, Emilie Clerc, and Christelle Chrea are employees of Philip Morris International. Agnes Bacso, Erica Spies, and Sophie Gallot completed the work during prior employment with Philip Morris International. Linda Abetz-Webb is a consultant for Philip Morris International.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Work completed during prior affiliation with PMI R&D: Erica Spies, Agnes Bacso, Sophie Gallot.

Supplementary Information

Additional file 1..

Summary tables of results of scoping literature review

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ . The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Cite this article.

Afolalu, E.F., Spies, E., Bacso, A. et al. Impact of tobacco and/or nicotine products on health and functioning: a scoping review and findings from the preparatory phase of the development of a new self-report measure. Harm Reduct J 18 , 79 (2021). https://doi.org/10.1186/s12954-021-00526-z

Download citation

Received : 14 October 2020

Accepted : 21 July 2021

Published : 30 July 2021

DOI : https://doi.org/10.1186/s12954-021-00526-z

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Health and functioning
• Scoping review
• Qualitative research
• Conceptual framework

Harm Reduction Journal

ISSN: 1477-7517

• Submission enquiries: [email protected]

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here .

Loading metrics

Open Access

Peer-reviewed

Research Article

Perceptions of health risks of cigarette smoking: A new measure reveals widespread misunderstanding

Affiliation Department of Communication, Stanford University, Stanford, California, United States of America

Affiliation Graduate School of Business, Stanford University, Stanford, California, United States of America

* E-mail: [email protected]

Affiliations Department of Political Science, Vanderbilt University, Nashville, Tennessee, United States of America, Hoover Institution, Stanford University, Stanford, California, United States of America

Affiliation U.S. Department of Treasury, Washington, D.C., United States of America

Affiliation LinChiat Chang Consulting, LLC, San Francisco, California, United States of America

Affiliation Department of Communication Studies, University of Michigan, Ann Arbor, Michigan, United States of America

Affiliation GfK Custom Research North America, New York City, New York, United States of America

• Jon A. Krosnick, 
• Neil Malhotra, 
• Cecilia Hyunjung Mo, 
• Eduardo F. Bruera, 
• LinChiat Chang, 
• Josh Pasek, 
• Randall K. Thomas

• Published: August 14, 2017
• https://doi.org/10.1371/journal.pone.0182063
• Reader Comments

15 Feb 2019: Krosnick JA, Malhotra N, Mo CH, Bruera EF, Chang L, et al. (2019) Correction: Perceptions of health risks of cigarette smoking: A new measure reveals widespread misunderstanding. PLOS ONE 14(2): e0212705. https://doi.org/10.1371/journal.pone.0212705 View correction

Most Americans recognize that smoking causes serious diseases, yet many Americans continue to smoke. One possible explanation for this paradox is that perhaps Americans do not accurately perceive the extent to which smoking increases the probability of adverse health outcomes. This paper examines the accuracy of Americans’ perceptions of the absolute risk, attributable risk, and relative risk of lung cancer, and assesses which of these beliefs drive Americans’ smoking behavior. Using data from three national surveys, statistical analyses were performed by comparing means, medians, and distributions, and by employing Generalized Additive Models. Perceptions of relative risk were associated as expected with smoking onset and smoking cessation, whereas perceptions of absolute risk and attributable risk were not. Additionally, the relation of relative risk with smoking status was stronger among people who held their risk perceptions with more certainty. Most current smokers, former smokers, and never-smokers considerably underestimated the relative risk of smoking. If, as this paper suggests, people naturally think about the health consequences of smoking in terms of relative risk, smoking rates might be reduced if public understanding of the relative risks of smoking were more accurate and people held those beliefs with more confidence.

Citation: Krosnick JA, Malhotra N, Mo CH, Bruera EF, Chang L, Pasek J, et al. (2017) Perceptions of health risks of cigarette smoking: A new measure reveals widespread misunderstanding. PLoS ONE 12(8): e0182063. https://doi.org/10.1371/journal.pone.0182063

Editor: Raymond Niaura, Legacy, Schroeder Institute for Tobacco Research and Policy Studies, UNITED STATES

Received: May 7, 2016; Accepted: June 20, 2017; Published: August 14, 2017

This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Data Availability: Data are available at: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/JP2JHH , doi: 10.7910/DVN/JP2JHH .

Funding: LC and RKT have commercial affiliations with LinChiat Chang Consulting and GfK Custom Research North America, respectively. These companies provided support in the form of salaries for authors LC and RKT, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.

Competing interests: LC and RKT have commercial affiliations with LinChiat Chang Consulting and GfK Custom Research North America, respectively. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Introduction

Despite a constant flow of messages reminding Americans of the health risks of cigarette smoking, and despite a steady decline in the proportion of Americans who smoke during the last 50 years, more than 20% of Americans continue to smoke regularly today [ 1 ]. This paper explores whether the continued prevalence of smoking may, in part, stem from a failure to acknowledge these risks. At first blush, this assertion may seem patently implausible; much research indicates that increasingly large proportions of Americans recognize the various dangers of smoking, and some studies even suggest that most Americans overestimate the proportion of smokers who suffer from certain smoking-related ailments [ 2 ]. Nonetheless, it is possible that people underestimate the magnitude of some of the health risks caused by smoking. Because individuals seem to base their decisions about whether to smoke on how they believe the act of smoking changes the risk of contracting specific diseases, correcting any underestimation of risk may yield future reductions in smoking onset and increases in cessation [ 3 ]. To explore these possibilities, we conducted three studies of national samples of American adults documenting risk perceptions and their relations to smoking behavior.

Challenges in the study of risk perception

One way to gauge the accuracy of people’s perceptions of the health dangers of smoking is to focus simply on the list of maladies that become more likely as a result of smoking. This list includes various cancers, heart diseases, respiratory diseases, premature death, and more [ 4 , 5 ]. By asking representative national samples of American adults to identify which diseases and medical conditions on a provided list are linked with smoking, researchers have illuminated three interesting patterns. First, since the 1950s, the proportion of Americans who failed to identify any health risks of smoking dropped consistently [ 6 ]. Second, according to Gallup [ 7 ], a sizable proportion of Americans still fails to recognize a link between smoking and some related ailments (see S1 Fig ). Other contemporary surveys support these same conclusions [ 8 – 10 ]. The proportion of American adults who associate smoking with a particular ailment varies considerably across ailments, from a high of 81% who report a link between smoking and cancer to single-digit proportions who identify links with asthma, hypertension, bronchitis, and stroke [ 11 ]. Thus, even today, Americans apparently underestimate the breadth of the danger.

A more refined way to gauge the accuracy of perceptions is to focus on the amount of increased risk of each malady that results from smoking. According to epidemiological studies, each of these increases is a function of many attributes, including age of smoking onset, number of years of regular smoking, number of cigarettes consumed per day, and more [ 4 , 5 ]. Therefore, actual risks must be expressed as variables that are functions of such factors, and perceptions of these risks must be ascertained specifying such factors.

Furthermore, even holding constant age of onset, length of smoking, and dosage, a smoking-related risk can be perceived in three different ways: (1) absolute risk (i.e., “what is the chance that a person will get lung cancer if he/she smokes?”), (2) attributable risk (i.e., “how much does smoking raise the chances that a person will get lung cancer compared to not smoking?”), and (3) relative risk (i.e., “how much more likely is a person to get lung cancer if he/she smokes?”) [ 12 , 13 ]. Mausner and Bahn [ 14 ] provide a thorough review of how epidemiologists calculate and use each of these different measures of risk. Assessing Americans’ perceptions of all three seems most sensible in order to determine whether people tend to perceive all types of risk accurately, overestimate all types of risk, underestimate all types of risk, or overestimate some while underestimating others. Attributable fraction is another measure of risk perceptions, but we do not investigate this measure in this study [ 15 ].

One way to think about the goal of such an investigation is to identify any ways in which people underestimate risk, so that public health education campaigns can correct this misunderstanding. But it could turn out that people underestimate one particular type of risk (e.g., absolute risk) and yet do not use that particular perception of risk in their decision-making about whether to start or stop smoking. Therefore, efforts to correct the public’s misunderstanding would not translate into changes in smoking behavior. So to draw out implications of measurements of perceived risk, we need evidence indicating which perceptions may be behaviorally consequential.

The research described in this paper set out to do so by gauging perceptions of absolute risk, attributable risk, and relative risk with a focus specifically on lung cancer. And we explored which of these risk perceptions might drive smoking onset and cessation. We focus on lung cancer specifically rather than all health risks associated with smoking following Viscusi’s seminal work on smoking-related risks [ 2 ]. While the share of American adults who associate smoking with a particular health malady varies across maladies [ 11 ], an assessment of which type of risk perception—absolute risk, attributable risk, and relative risk—impacts Americans’ smoking behavior the most should not be sensitive to the health malady of interest. In other words, if perceptions of relative risk of lung cancer affects smoking behavior more than perceptions of absolute and attributable risk of lung cancer, then perceptions of relative risk of another disease should similarly be most effective at driving smoking behavior.

Prior studies of perceptions of the magnitude of risk

A number of past studies have attempted to measure perceptions of the magnitude of the risk of smoking in representative samples of American adults, but their methodologies entailed a series of limitations, as we outline next. It is worth noting that this paper focuses on the U.S. and therefore does not discuss the many interesting studies of smoking-related risk perceptions that have been done in countries other than the U.S [ 16 – 18 ].

We also do not discuss studies that examined people’s perceptions of their own personal smoking-related risks (e.g., Boney-McCoy et al. [ 19 ]; Strecher et al. [ 20 ]) because our focus is on Americans’ perceptions of the risk of smoking to people in general. Many studies have produced interesting results involving people’s perceptions of their own personal risks of smoking-related health problems (e.g., [ 19 , 21 – 27 ]). However, according to Gigerenzer [ 28 ], people naturally think about the population rather than personal chance, and perceptions of personal risk likely mediate the relationship between general risk and behavior.

Because this paper is focused on the beliefs of adults, we also do not discuss the findings of many interesting studies of youth. For example, Romer and Jamieson [ 29 ] asked questions similar to Viscusi’s [ 2 ] of a national sample of 14- and 15-year-olds: “Out of every 100 cigarette smokers, how many do you think will: (a) get lung cancer because they smoke? (b) have heart problems, like a heart attack, because they smoke? (c) die from a smoking-related illness?” Their results mirror Viscusi’s [ 2 ]: on average; respondents said 61.4% of smokers would develop lung cancer, much higher than the true rate. Likewise, a representative sample of 20–22 year olds said 52.6% on average. Many other studies have explored the beliefs of children and adolescents as well [ 21 , 30 – 37 ].

Some past studies have asked people to describe their perceptions of the magnitude of a smoking-related risk of some malady by asking people to select a point on a rating scale with a small number of verbally labeled response options. For example, Weinstein et al. [ 27 ] asked “How likely do you think it is that (the average male cigarette smoker/the average female cigarette smoker/you) will develop lung cancer in the future?” and offered a 5-point scale ranging from “very low” to “very high.” Similarly, Romer and Jamieson [ 29 ] asked respondents “In your opinion, is smoking very risky for a person’s health, somewhat risky, only a little risky, or not risky at all?” It is not clear whether “somewhat risky” or “very risky” is an overestimate or underestimate of risk. In other words, measures that assess perceptions of smoking’s dangers on these non-numeric subjective probability scales do not permit assessing the degree to which magnitudes of perceived risk reflect true numeric risk levels.

Other studies have measured perceptions of risks quantitatively but did not specify the population of people being described or the dosage of smoking being addressed. For example, in a survey conducted by Audits & Surveys Worldwide, respondents were asked, “Among 100 cigarette smokers, how many of them do you think will get lung cancer because they smoke?” [ 2 ]. The characteristics of a smoker are important contextual considerations with regards to actual health risks a given smoker faces. The probabilities of various smoking-related ailments differ for occasional and daily smokers and depend on the age of a smoker as well as the duration of smoking. Because this type of question does not specify what population is to be described or how much smoking was done for how long, it is impossible to gauge the accuracy of responses by comparing them with the results of epidemiological studies, which show risk to vary across populations and age, smoking duration, and dosage. Some scholarly work has begun to remedy this issue, specifying the exact quantity of cigarettes smoked per day [ 38 ].

Another potential limitation of the Audits & Surveys question is the phrase “because they smoke.” This phrase was presumably meant to lead respondents to estimate the number of lung cancer cases completely attributable to smoking. As Slovic [ 36 ] observed, this phrase can be interpreted in various different ways. Specifically, people may believe that smoking, along with other factors, enhances the chances of contracting lung cancer, leading them to respond that smoking is partially responsible for some lung cancer cases. This, too, makes it difficult to identify the appropriate true rate of smoking-induced lung cancer cases to which to compare risk perceptions.

Finally, the notions of “subadditivity” and “the focus of judgment effect” point to another potential problem with the Audits & Surveys question [ 39 – 41 ]. The question, “Among 100 cigarette smokers, how many of them do you think will get lung cancer because they smoke?” focuses respondents’ attention on just one possible outcome of smoking: getting lung cancer. This approach typically leads to overestimation of the probability of the event in question. Asking respondents instead to report the number of smokers who will not get lung cancer would focus attention on that outcome instead, probably leading to overstatement of that probability. So the sum of the average answers to these two forms of the question would most likely total more than 100. A more desirable measurement approach would overcome the bias induced by arbitrarily asking about only one outcome (e.g., either getting lung cancer or not getting lung cancer).

The present research

To overcome the limitations of past studies, we conducted three surveys measuring Americans’ beliefs about smoking-related health risks in different ways. To gauge perceived risk, we asked two questions: one about the risk to nonsmokers, and the other about the risk to smokers. This approach is advantageous if a researcher wants to measure perceptions of attributable risk or relative risk, because (1) subadditivity is likely to bias both reports upward, so subtracting or dividing one judgment from or by the other will minimize the impact of overestimation, (2) answers to these questions can be used to generate assessments of perceived absolute risk, attributable risk, and relative risk, and (3) this approach employs the principle of decomposition, which enhances the accuracy of measures of people’s beliefs [ 15 ]. It is worth noting one limitation of our research is the fact that we only ask about lung cancer, and do not consider other health risks linked with smoking. However, most likely people’s perceptions of risk across multiple disease categories would be positively correlated. Consequently, our general conclusions about lung cancer would likely be similar if respondents were forced to consider multiple disease categories.

In decomposition, a single, global judgment is broken down into a series of sub-judgments, each of which a respondent must make in the process of generating the global judgment. Here, in order to gauge people’s perceptions of relative risk, we could ask, “how many more times likely is a smoker to get lung cancer than a nonsmoker?” To answer the global question, a respondent must estimate both the likelihood a nonsmoker will get lung cancer and estimate the likelihood that a smoker will get lung cancer, and then mentally compute the ratio of the probabilities. Because respondents can accidentally make a computational error when executing that last step, surveyors can more accurately measure people’s beliefs by asking directly about the sub-judgments, leaving the researcher to compute the ratio. The same logic applies to the measurement of perceived attributable risk (see S1 Appendix for a discussion of measuring probabilities and numeracy).

When measuring perceptions of the lung cancer risks of nonsmokers and smokers, we expressed specifically a volume of smoking and at what age it began, so we could more accurately gauge the extent to which people overestimated or under-estimated the health risks of smoking. And rather than asking survey respondents to report probabilities, we asked them to report frequencies, since a variety of studies suggest that people think more naturally in terms of frequencies [ 42 , 43 ].

We compared the three risk perception measures (absolute, attributable, and relative risk) in terms of their associations with cessation among a sample of current and former smokers. We also compared the risk perception measures in terms of their associations with the desire to quit among current smokers. Although previous studies have found positive and significant correlations between risk perceptions and the desire to quit, none of these studies compared different risk perception measures to one another or analyzed numerical risk estimates [ 27 , 44 , 45 ].

Such associations can occur for at least two reasons. First, beliefs about the health risks of smoking may be instigators of smoking cessation (for a review of this literature, see S2 Appendix ). Second, people may adjust their beliefs about smoking’s health risks in order to rationalize their status as a smoker or a non-smoker [ 46 – 48 ]. If perceptions of health risks are motivators of smoking cessation and/or if quitting smoking induces people to inflate their risk perceptions, then perceived risk should be lower among people who currently smoke than among people who have quit. That is, the higher a person’s perceived risk, the more likely he or she is to have quit. Likewise, the higher a current smoker’s perception of risk, the more motivated he or she should be to quit smoking. Therefore, the more strongly a risk perception measure is associated with whether a person has quit smoking and a smoker’s desire to quit, the more likely that risk perception is to capture the way people naturally think about risk in this arena.

Many possible patterns of risk perception types could be found in a population. The most heterogeneous pattern would be one in which one-third of people think in terms of absolute risk, while another one-third of people think in terms of attributable risk, and the remaining people think in terms of relative risk. The most homogeneous case would be one in which everyone thinks in terms of just one of these risk perceptions to make behavioral choices regarding smoking. Our analyses explored the extent of use of each of the three risk perception measures.

We also explored whether people who felt more certain about risk perceptions manifested stronger relations of those perceptions with cessation and desire to quit. Psychological research on attitude strength suggests that people hold beliefs and attitudes with varying degrees of certainty, and beliefs held with more certainty are more likely to shape thinking and action [ 49 ]. Therefore, we explored whether any of the risk perceptions were more strongly related to cessation among people who held their risk perceptions with more certainty.

Three studies

Our three studies explored five main questions: (1) How many people overestimate and underestimate absolute risk, attributable risk, and relative risk of lung cancer due to smoking? (2) How strongly are perceived absolute risk, attributable risk, and relative risk related to quitting? (3) How strongly are perceived absolute risk, attributable risk, and relative risk related to desire to quit among current smokers? (4) Are the relations between risk perceptions and quitting strongest among respondents who are most certain about their risk perceptions? (5) How strongly are perceived absolute risk, attributable risk, and relative risk related to having initiated smoking?

Study 1 was a random digit dial telephone survey of a nationally representative sample of American adults who were current or former smokers, conducted in 2000 by Schulman, Ronca, and Bucuvalas, Inc. (hereafter SRBI). Study 2 was a 2006 survey of a national non-representative sample of current and former smokers who volunteered to complete Internet surveys for Harris Interactive in exchange for points that could be redeemed for gifts. Study 3 was a 2009 survey of a nationally representative sample of all Americans, including people who had never smoked, via the Face-to-Face Recruited Internet Survey Platform (the FFRISP; see S3 Appendix for descriptions of the methodologies of the three studies, and see S4 Appendix for the demographic characteristics of the three samples).

The telephone survey respondents who were current or former smokers were asked:

(1) “Next, I'd like to turn to a different topic: what you personally think about the effect of cigarette smoking on people's health. I'm going to read these next two questions very slowly to let you think about each part of them, and I can repeat each question as many times as you like before you answer, so you can be sure they are clear to you. First, if we were to randomly choose one thousand American adults who never smoked cigarettes at all during their lives, how many of those one thousand people do you think would get lung cancer sometime during their lives?” (2) “And if we were to randomly choose one thousand American adults who each smoked one pack of cigarettes a day every day for 20 years starting when they were 20 years old, how many of those one thousand people do you think would get lung cancer sometime during their lives?” (3) “You said that smokers are [more likely/as likely/less likely] to get lung cancer than nonsmokers. How certain are you about this? Extremely certain, very certain, moderately certain, slightly certain, or not certain at all?”

We ask respondents to assess the prospect of lung cancer incidence generally like Viscusi [ 2 ]. We emphasized “personally” so that people would feel comfortable providing their own best guess of a fact, specifically general population risk of contracting lung cancer. This wording is designed to avoid the question seeming like a “quiz” (or their guess of what a public health authority might say), but rather their personal assessment of risk. For the two Internet surveys, the wording was adapted for self-administration. In all three studies, the response choices for the last question were presented in descending order for a randomly chosen half of the respondents and in ascending order for the other half. By implementing the same internally valid research design three separate times, it is possible to assess whether our findings are replicable.

Each of the three studies discussed above were deemed as suitable for exempt IRB review status by Stanford University’s review board, as no identifying information on the respondents was retained, and disclosure of answers to the survey questions would not place the respondents at risk. Informed consent for Study 1 was provided verbally given that Study 1 was a telephone survey. Written informed consent was provided for both Study 2 and Study 3, and Stanford’s IRB approved use of oral consent in Study 1 and written consent in Study 2 and 3.

Actual risk

We used data reported by Peto et al. [ 50 ] to compute the actual absolute risk, attributable risk, and relative risk of contracting lung cancer for one-pack-a-day smokers who started smoking at age 20 and smoked for 20 years. To do so, we divided the absolute risk of mortality due to lung cancer among these smokers (about 3%) by the absolute risk of mortality due to lung cancer among non-smokers (about 0.4%, yielding a relative risk of about 7). Although Peto et al. [ 50 ] examined mortality instead of incidence, the probability of dying from lung cancer conditional on developing lung cancer is 74.4% within a thirteen-year period according to Marcus et al. [ 51 ], and even higher among smokers [ 52 ]. If relative risk is higher, then our results understate the proportion of Americans who underestimate this relative risk. According to these figures, the attributable risk of lung cancer due to smoking is then about 3% (3% minus 0.4%, rounds to 3%). It is worth noting that although one might imagine that it is difficult to estimate risk rates because of complex functional forms, interactions of smoking with other risk factors, cohort effects, and other complications, research suggests that in fact, risk rates are largely robust to some potential complexities [ 53 – 55 ].

Perceived risk

In Study 1, the mean of current and former smokers’ perceptions of absolute risk of lung cancer among smokers was 48% (i.e., 480.1 smokers out of 1,000 smokers would get lung cancer); the median was 50% (see columns 1 and 2 of Table 1 ). 10.3% of respondents perceived absolute risks between 0% and 5.0%, and the remaining respondents gave answers above 5.0%. 99.5% of respondents overestimated absolute risk, only about 0.3% estimated it correctly (by giving an answer of 30), and 0.2% underestimated it (by giving an answer less than 30).

• PPT PowerPoint slide
• PNG larger image
• TIFF original image

https://doi.org/10.1371/journal.pone.0182063.t001

As expected, the mean and median perceived absolute risk of nonsmokers getting lung cancer were less: 22% and 10%, respectively. Thirty-six percent of respondents gave answers between 0% and 5.0%. Thus, most people vastly overestimated this absolute risk.

Only 5.2% of respondents thought smokers were less likely to get lung cancer than nonsmokers (a belief revealed by attributable risks less than 0; see columns 1 and 2 of Table 2 ). Attributable risk was calculated by subtracting each respondent’s answer to the question about nonsmokers from his or her answer to the question about smokers. 9.6% of respondents thought smokers and nonsmokers were equally likely to contract lung cancer, reporting an attributable risk of 0. A large majority, 85.2% of respondents, reported that smokers were more likely than nonsmokers to contract lung cancer. 76.1% overestimated attributable risk by reporting figures greater than 4%. The mean perceived attributable risk was about 27%, and the median was 20%.

https://doi.org/10.1371/journal.pone.0182063.t002

In contrast, a large majority of respondents (74.6%) underestimated relative risk, because they reported perceptions that implied a relative risk less than 7 (see columns 1 and 2 of Table 3 ). Relative risk was computed by dividing each respondent's answer to the question about 1,000 smokers by his or her answer to the question about 1,000 nonsmokers. Because this quantity is undefined for respondents who said none of the 1,000 nonsmokers would get lung cancer (because the denominator would be zero), 1 was added to these respondents’ answers to the questions about smokers and nonsmokers to allow the relative risk quantity to be defined for all respondents. Note that re-computing all analyses reported below treating these people as having missing data on the relative risk measure had negligible impact on the reported results. 54.6% of the respondents could be said to have vastly underestimated relative risk, because their reports implied a value less than 3. Only about 1.5% of respondents perceived relative risk approximately correctly (e.g., 7), and only 23.9% of respondents overestimated relative risk. 5.2% of respondents perceived a relative risk of less than 1, meaning they thought smokers developed lung cancer less often than nonsmokers, and 9.6% of the sample perceived a relative risk of 1.0, meaning they thought smokers and nonsmokers were equally likely to develop lung cancer. Mean perceived relative risk was 26.7, much higher than the true value, and the median was 2.5, lower than the true value. Thus, relative risk tells a very different story about the prevalent errors in risk perceptions than does attributable risk: most people overestimated the latter, whereas most people underestimated the former.

https://doi.org/10.1371/journal.pone.0182063.t003

Compared to the representative sample of current and formers smokers interviewed in Study 1, Study 2’s non-probability sample of current and former smokers reported: (1) lower perceived absolute risk of lung cancer among nonsmokers and smokers (e.g., 49.5% and 25.7%, respectively, gave answers between 0 and 50 out of 1,000 who would get lung cancer, compared to 36.0% and 10.3% in Study 1; see seventh and eighth columns in Table 1 ); (2) lower perceived attributable risk (e.g., 50.9% had a value of 99 or less, compared to 30.7% of the Study 1 respondents; see the eighth column of Table 2 ); and (3) lower perceived relative risk (e.g., 59.5% had values of 2.99 or less, as compared with 54.6% of the Study 1 respondents; see the eighth column of Table 3 ).

Using all three risk measures, Study 3’s representative sample of current and former smokers perceived less risk than the Study 1’s respondents did 9 years earlier. Study 3’s current and former smokers reported lower absolute risk among nonsmokers (mean = 11.9%, median = 5%) than did the Study 1 respondents (mean = 21.5%, median = 10%; see columns nine and one, respectively, of Table 1 ). Study 3’s current and former smokers perceived lower absolute risk for smokers than did the Study 1 respondents (means = 33.1% vs. 48.0%; medians = 30.0% vs. 50.0%; see columns ten and two, respectively, of Table 1 ). And Study 3’s current and former smokers perceived lower attributable risk of smoking than did the Study 1 respondents (means = 21.1% vs. 26.7%; medians = 11.5% vs. 20.0%; see columns nine and one, respectively, of Table 2 ) and lower relative risk than did the Study 1 respondents (means = 12.9 vs. 26.7; medians = 2.5 vs. 2.5; see columns 9 and 1, respectively, of Table 3 ).

Study 3 suggests that the perceived risk of lung cancer may have declined among current and former smokers between 2000 and 2009. That is, the two representative sample surveys indicated that respondents’ assessments of the absolute risk of lung cancer for both smokers and non-smokers became notably more accurate during this period.

Comparing risk measures

Which of these measures is an appropriate focus for claims about public risk perceptions and their accuracy? One way to answer this question is to determine which of these risk perceptions drives people’s decisions about whether or not to smoke. Many possible patterns of risk perception use are possible in any population. The most heterogeneous pattern would be one in which some people decide whether to smoke or quit based upon their perceptions of the attributable risk, while others make this decision with reference to perceptions of relative risk, and still others make their decisions based on perceptions of absolute risk, with the three groups being of roughly equal size. The most homogeneous case is that in which everyone uses just one of these risk perceptions to make their behavioral choices regarding smoking. By gauging which risk perceptions have how much impact for how many people, we can begin to understand whether smoking behavior overall in a population is driven mostly by perceptions that overestimate risk, mostly by perceptions that underestimate risk, or by a mixture of perceptions that sometimes overestimate and other times underestimate.

The data of all three studies allowed us to explore whether perceptions of attributable risk, relative risk, and absolute risk inspire people to quit smoking by comparing current and former smokers. If perceptions of health risks are indeed a principal motivator of smoking cessation, then perceived risk should be lower among people who currently smoke than among people who used to smoke but have quit. In other words, the higher a person’s perceived risk, the more likely he or she should be to have quit smoking. Based upon this assumption, the better a risk perception measure predicts whether a person has quit smoking, the more likely that risk perception is to have driven quitting decisions.

To adjudicate whether absolute risk, attributable risk, or relative risk drove people’s decisions to quit, we estimated the parameters of generalized additive models (GAMs) comparing current smokers to former smokers by using a Gaussian link function predicting a binary variable representing whether a respondent was a current or former smoker using the various measures of perceived risk and the weights for unequal probability of selection and demographic post-stratification (see S5 Appendix for more details on GAMs). GAMs are especially useful for estimating models containing two highly correlated predictors (as we have here) because relaxing the assumption of linearity prevents model misspecification, allowing for better isolation of the unique relations of different risk perceptions with other variables.

Using this flexible approach, we first estimated a model in which relative and attributable risk predicted quitting (more precisely, having quit). It might seem appealing to estimate GAMs predicting quitting using all three measures, but non-independence among the three measures of perceived risk makes that impossible. When examining Study 1’s data, we see that perceptions of relative risk were sensibly correlated with diminished chances of remaining a smoker (see the top-left panel of S2 Fig ). The dark line in the figure represents the estimated relation, and the two light lines demark the bounds of the 95% confidence interval around the estimates. The small vertical lines at the bottom of the figure (called “rugmarks”) indicate whether one or more respondents provided a data point at each point along the x-axis. Increasing perceived relative risk was associated with decreased log-odds of remaining a smoker. Movement from the 25 th percentile to the 75 th percentile (weighted) of relative risk increased the probability of quitting by 13.8 percentage points (see the first row of the first column of Table 4 ).

https://doi.org/10.1371/journal.pone.0182063.t004

In contrast, over the range of the bulk of the data (where the majority of the rugmarks on the x-axis are located), the relation between attributable risk and quitting was fairly flat (see bottom-left panel of S2 Fig ). Movement across the interquartile range of attributable risk increased the probability of quitting negligibly, by only 0.3% (see second row of the first column of Table 4 ).

To more formally gauge and compare these relations, we estimated a set of nested GAMs. First, we estimated a model predicting quitting using only attributable risk and then observed the improvement in goodness of fit of the model when we added relative risk as a predictor. A likelihood ratio (hereafter LR) test comparing the log likelihood of the two-variable model to the nested one-variable model indicated that the addition of the extra variable resulted in a significantly better fit (p=.03), meaning that relative risk was a reliable unique predictor of quitting (see third row of the first column of Table 4 ). Next, we estimated a model predicting quitting using only relative risk and then estimated the improvement in goodness of fit when attributable risk was added as a predictor. This addition did not improve the model’s fit significantly (p=.64; see fourth row of the first column of Table 4 ). Thus, relative risk perceptions appear to have been related to decisions to quit smoking, whereas perceptions of attributable risk were not.

To explore whether absolute risk outperforms relative risk, we estimated a GAM in which quitting was predicted by both measures. As shown in the right panels of S2 Fig , relative risk was again sensibly related to quitting (with probability of remaining a smoker declining smoothly as perceived risk increased), whereas absolute risk was not. Again, adding relative risk to a model fitted with only absolute risk improved the fit significantly (p=.002), whereas adding absolute risk to a model with relative risk did not yield a significant improvement in fit (p=.15; see rows seven and eight of the first column of Table 4 ). Movement across the interquartile range of absolute risk was associated with a 10.5% decrease in the chances of quitting, whereas movement across the interquartile range of relative risk was associated with a sizable and more reasonable 15.2% increase in the likelihood of quitting (see rows five and six of the first column of Table 4 ). As shown in columns two and three of Table 4 (as well as S3 and S4 Figs), these same results were replicated in Studies 2 and 3.

There may be an illusion hidden in these results. When people are asked to report a probability but do not know the answer, they sometimes answer “50,” meaning “fifty-fifty” or “I don’t know,” rather than meaning a 50% chance [ 56 ]. To explore the impact of this potential source of measurement error on our conclusions, we re-estimated the logistic GAM by: (1) dropping the respondents who answered “500” to the question about nonsmokers or to the question about smokers; (2) replacing the 500s with values generated by multiple imputation; and (3) replacing the 500s with answers obtained by a follow-up probe. The results supported the above conclusions even more strongly (for details of these approaches and results, see S6 Appendix ).

Next, we explored whether certainty moderated the associations of risk perceptions with quitting behavior. In Study 1, as expected, the correlation of relative risk with quitting was significantly stronger among high certainty respondents (people who were extremely certain, 27% of the sample) than among lower certainty respondents. Among the high certainty respondents, the probability of quitting increased over the interquartile range of relative risk by 23.7 percentage points (p=.008), a much larger increase than among the low certainty respondents, whose positive change was just 10.5 percentage points (p=.054). Accounting for certainty significantly improved the goodness of fit of the model (p=.03).

Likewise, in Study 2, the positive relation between perceived relative risk and quitting was significantly stronger among high certainty respondents than among low certainty respondents (p=.009). Among the high certainty respondents (18% of the sample), movement across the interquartile range of relative risk increased the probability of quitting by 44.1% (p<.001), whereas movement across this interquartile range in the low certainty group was associated with an increase in quitting probability of only 13.6% (p<.001). Accounting for certainty significantly improved the goodness of fit of the model (p=.009).

In Study 3, among high certainty individuals (30.5% of the sample), movement across the interquartile range of relative risk was associated with an increased probability of quitting smoking of 15.8% (p=.06), whereas movement across this interquartile range in the low certainty group was associated with an increase in quitting probability of 11.1% (p=.03). Accounting for certainty again significantly improved the goodness of fit of the model (p=.03).

Desire to quit.

Next, we examined whether current smokers’ risk perceptions were associated with their desire to quit. While a desire to quit does not automatically translate to smoking cessation, a strong desire to quit is predictive of subsequent quitting behavior, and is a necessary condition for quitting [ 57 ]. In Study 1, adding relative risk to a GAM model predicting desire to quit among current smokers with attributable risk caused a marginally non-significant improvement in fit (p=.09; see the third row of column four in Table 4 ). Movement from the 25 th to the 75 th percentile of relative risk raised the probability of wanting to quit by 17.0% (see the first row of column four in Table 4 ). But adding attributable risk to a model predicting desire to quit with relative risk did not improve fit significantly (p=.27; see row four of column four in Table 4 ). Movement across the interquartile range of attributable risk slightly lowered desire to quit by 1.1% (see row two of column four in Table 4 ). Likewise, adding relative risk to a model including absolute risk yielded a significant improvement in fit (p=.046; see row seven of column four in Table 4 ). Movement across the interquartile range of relative risk increased desire to quit by 13.9% (see row five in Table 4 ). But adding absolute risk to a model including relative risk marginally significantly decreased desire to quit (interquartile range movement = 15.6%, p=.06; see rows six and eight of column four in Table 4 ). The data from Studies 2 and 3 yielded similar results (see columns five and six of Table 4 ). This further supports the contention that people think in terms of relative risk perceptions.

Smoking onset.

We observed the expected results when we used the three measures in Study 3 to explore whether perceived risk was greater among people who ever smoked than among people who never smoked. Comparing the distributions in the ninth and tenth columns in Table 1 with the distributions in the last two columns of the table, we see that: (1) both groups had similar expectations for the proportion of nonsmokers who would get lung cancer (mean = 11% for people who never smoked vs. 12% for people who ever smoked), but (2) the expected proportion of smokers who would get lung cancer was higher among people who had never smoked (mean = 43.3%) than among people who ever smoked (mean = 33.1%).

Also as expected, people who never smoked perceived higher attributable risk of smoking than did people who ever smoked (see the last two columns in Table 2 ): (1) 3.9% thought that smokers were less likely to contract lung cancer than nonsmokers (attributable risk of less than 0); (2) 6.3% thought that smokers and nonsmokers were equally likely to get lung cancer (attributable risk of 0); and (3) 89.7% thought that smokers were more likely to contract lung cancer than nonsmokers. Respondents who never smoked thought smokers were 32 percentage points more likely than nonsmokers to get lung cancer, on average (see columns 11 and 12 of Table 2 ). Thus, these individuals perceived a higher attributable risk than did current and former smokers (21.1 percentage points; see column nine of Table 2 ). Likewise, respondents who never smoked also perceived higher relative risk than did current and former smokers (compare the last two columns of Table 3 with the ninth and tenth columns of that table).

As expected, perceptions of relative risk were strongly associated with status as a never smoker vs. a current smoker in GAMs (see the left panels of S5 Fig ). Adding relative risk to a model predicting current smoking with attributable risk considerably improved fit (p<.001), whereas adding attributable risk to a model with relative risk did not significantly improve fit (p=.57). Movement across the interquartile range of relative risk yielded a 22.7 percentage point decrease in the likelihood that respondents were smokers. Movement across the interquartile range of attributable risk yielded a decrease in the probability of being a smoker of only 0.7 percentage points.

Likewise, adding relative risk to a model with only absolute risk improved fit significantly (p<.001), whereas adding absolute risk to a model including relative risk was associated with only a marginally significant improvement in fit (p=.07). Movement across the interquartile range of relative risk (when controlling for absolute risk) was associated with a 22.3 percentage point decrease in the probability of ever having smoked (see the right panels of S5 Fig ). In contrast, movement across the interquartile range of absolute risk (when controlling for relative risk) produced only an 8.5 percentage point decrease in the likelihood of ever having smoked.

Summary and implications

Taken together, this evidence suggests that while Americans have overestimated the absolute risk and risk difference of lung cancer associated with cigarette smoking, Americans have generally underestimated the relative risk. Furthermore, this evidence suggests that people may think more about smoking health risks in terms of relative risk than in terms of absolute risk or risk difference. The relations we saw here may result from the influence of health risk beliefs on decisions to quit smoking, decisions to start smoking, and regret about smoking, or these relations may occur because people rationalize their smoking status by adjusting their risk perceptions, or from some other process. Having seen here that these are possibilities, we look forward to future research exploring them to characterize the basis for the relations we observed.

Communication of risk has been a difficult task for medical professionals, and our findings encourage consideration of a different approach to communicating health risks than has been typical on American cigarette packages and in other prominent health communications [ 58 , 59 ]. There are a large number of studies that show that the design of and warnings on cigarette packs can influence perceptions of the risks of smoking [ 60 – 68 ]. However, much constructive work can perhaps still be done by informing individuals about how much smoking increases their health risks. If the findings reported here are correct in suggesting that people use perceptions of relative risk when deciding whether to quit smoking, and if relative risk is indeed underestimated by most current and former smokers, corrective steps in this regard might be consequential. More specifically, if public health efforts are initiated in the future to encourage Americans to more accurately recognize the magnitudes of relative risks for various undesirable health outcomes of cigarette consumption, this may well lead to a reduction in the nation’s smoking rate and a consequent reduction in smoking-related morbidity and mortality. This may be why quantitative information about relative risk on cigarette packages in Australia (e.g., “Tobacco smoking causes more than four times the number of deaths caused by car accidents.”) appears to have been effective in encouraging smoking cessation [ 69 ].

Future research could explore these possibilities with experiments gauging the effects of different ways of describing risks on cigarette packages and other health communication mediums like television advertisements, poster campaigns, and doctor-patient communication [ 70 ]. Our findings suggest that when conducting such experiments, it may be desirable to attempt to alter people’s perceptions of relative risk in order to most directly address people’s natural approach to thinking about health risks in this arena. Perceptions of relative risk might be changed best by making such direct statements. But it may also be that such perceptions can be changed even more effectively by inducing affective reactions or in other non-quantitative ways, while simultaneously maximizing trust in the source of the information [ 71 , 72 ]. It is important to bear in mind that even successful efforts to change risk perceptions may not produce changes in behavior, so it will be important for future investigations to assess whether risk perception changes are translated into action [ 73 ].

In addition to their applied value, the findings reported here are interesting in basic psychological terms. By distinguishing between absolute, attributable, and relative risk, the present findings encourage future study with such measures to understand how people make many types of risky decisions and, more generally, how people trade off probabilities when making choices. And many important questions remain regarding risk perceptions involving smoking, such as how people arrive at their perceptions of relative, attributable, and absolute risk, and when and why some people use one measure rather than another to make behavioral decisions. Future studies of these sorts of issues seem merited, both in the smoking and other domains.

Resonance with other findings

Various findings reported here resonate with findings of some past studies. For example, Viscusi [ 2 ] and Borland [ 69 ] found that people overestimated the absolute risk of smoking. Khwaja et al. [ 74 ] found that both smokers and non-smokers overestimated their risks of dying from all sorts of causes [ 69 ]. When Weinstein et al. [ 27 ] asked respondents to assess the relative risk of smoking (“Would you say the average smoker has about the same lung cancer risk as a nonsmoker, a little higher lung cancer risk than a nonsmoker, twice the nonsmoker’s risk, five times the nonsmoker’s risk, or ten times the nonsmoker’s risk?”), smokers offered underestimates.

Boney-McCoy et al. [ 19 ] found that current smokers perceived the absolute risk of smoking to be significantly lower than that perceived by former smokers. This is consistent with the evidence reported here that when considered alone, absolute risk perceptions are related to quitting in the same way. However, when controlling for relative risk, the relation of quitting to absolute risk perceptions was close to zero in the present data.

Antoñanzas et al. [ 75 ] found distributions of Spaniards’ perceptions of attributable and relative risk (regarding the impact of cigarette smoking on lung cancer and heart disease) very similar to those reported here. Viscusi et al. [ 76 ] found that each of these risk perceptions predicted Spaniards’ status as a smoker or nonsmoker when considered alone, and relative risk was a considerably stronger predictor than attributable risk, though Viscusi et al. [ 76 ] did not assess the predictive abilities of perceived attributable risk and relative risk in a single regression equation.

The present evidence that people seem to think in terms of relative risk rather than attributable or absolute risk resonates with research on effective ways to communicate risks to patients [ 77 , 78 ]. For example, Malenka et al. [ 13 ] asked respondents to imagine they had a disease and could choose to take one of two medications—one described in terms of its impact on relative risk (“reduces risk of dying by 80%”) and the other (statistically equivalent) described in terms of impact on attributable risk (“can prevent 8 deaths per 100 people”). Most respondents preferred the medication described in terms of relative risk, perhaps because this portrayal resonated with people’s natural way of thinking about medication benefits found that relative risk information had more impact than did attributable risk information [ 79 – 83 ]. These findings contrast with Saitz’s [ 84 ] and Gigerenzer et al.’s [ 85 ] speculations that people will respond as well or better to attributable risk information (presented as two absolute risks) than to relative risk information, a finding challenged by our data as well.

A preference for thinking about health risks in terms of relative risk is also apparent in news media stories. In one study, 83% of such stories reported benefits of medications in terms of relative risk only, 2% reported benefits in terms of attributable risk only, and 15% reported benefits in terms of both indicators [ 86 ]. Similarly, medical journal articles tend to focus on reports of relative risk rather than attributable risk [ 87 ].

Other directions for further research

Future research might gain more insight into people’s natural ways of thinking about health risks by asking people to describe the health risks of smoking with whatever language they wish. With enough probing, open-ended data gathering might reveal whether people naturally use language evoking absolute risk, attributable risks, or relative risk levels, or a non-numeric representation, and such evidence is worthwhile to collect in future research [ 37 , 88 ]. Future work should also incorporate how much life is lost when calculating risk (see Viscusi [ 38 ] for a discussion of how this might affect an understanding of these results).

Generalizing beyond lung cancer

The focus of the analyses reported here has been people’s perceptions of the risk of getting lung cancer due to smoking. Because lung cancer is one of the best-known health risks of smoking [ 11 ], Americans may be less likely to underestimate the relative risk of lung cancer than of other diseases that are known to be caused by smoking. If we had asked survey questions about heart disease, oral cancers, or stroke instead of lung cancer, the prevalence of underestimation of relative risk may have been even greater than was observed for lung cancer. Correcting these misunderstandings may decrease the expected smoking rate even more. Future studies can explore these possibilities.

Implications regarding other domains of risk perception.

Differentiating perceived relative risk from perceived attributable risk may be useful in other health domains as well. For example, Meltzer and Egleston [ 89 ] reported that patients with diabetes vastly overestimated their own absolute risk of experiencing various complications. But perhaps their perceptions of relative risk are more accurate.

Implications for health education.

Psychological research on health counseling communication has revealed errors in people’s understanding of risk information [ 90 – 92 ]. However, educational efforts can present risk rates in various different ways, and some presentation approaches can cause misunderstandings [ 93 , 92 ]. The present evidence bolsters the conclusions of some past studies suggesting that future research may be most successful when presenting relative risk information to yield better quality decisions [ 94 – 99 ].

Supporting information

S1 fig. proportions of americans who failed to assert that smoking is dangerous to human health: gallup organization surveys..

https://doi.org/10.1371/journal.pone.0182063.s001

S2 Fig. Generalized Additive Models predicting the probability of being a current smoker: SRBI Survey (n = 456).

https://doi.org/10.1371/journal.pone.0182063.s002

S3 Fig. Generalized Additive Models predicting the probability of being a current smoker: Harris Interactive Survey (n = 795).

https://doi.org/10.1371/journal.pone.0182063.s003

S4 Fig. Generalized Additive Models predicting the probability of being a current smoker vs. former smoker: FFRISP (n = 471).

https://doi.org/10.1371/journal.pone.0182063.s004

S5 Fig. Generalized Additive Models predicting the probability of being a current smoker vs. never smoker: FFRISP (n = 714).

https://doi.org/10.1371/journal.pone.0182063.s005

S1 Appendix. Measuring risk.

https://doi.org/10.1371/journal.pone.0182063.s006

S2 Appendix. Literature on the relation of health risk perceptions with quitting smoking.

https://doi.org/10.1371/journal.pone.0182063.s007

S3 Appendix. Survey methodology.

https://doi.org/10.1371/journal.pone.0182063.s008

S4 Appendix. Demographics of current and former smokers in the SRBI Survey, current and former smokers in the Harris Interactive Survey, all individuals in the FFRISP Survey, and the nation’s population.

https://doi.org/10.1371/journal.pone.0182063.s009

S5 Appendix. GAMs.

https://doi.org/10.1371/journal.pone.0182063.s010

S6 Appendix. Exploring responses of 500.

https://doi.org/10.1371/journal.pone.0182063.s011

S7 Appendix. References for supporting information.

https://doi.org/10.1371/journal.pone.0182063.s012

Acknowledgments

The first survey described in this paper was funded by Empire Blue Cross/Blue Shield of New York. The third data set described was collected via the Face-to-Face Recruited Internet Survey Platform (FFRISP), funded by NSF Grant 0619956, Jon A. Krosnick, Principal Investigator. The authors thank Geoffrey Fong and Paul Slovic for very helpful suggestions. The authors acknowledge the excellent research assistance of Virginia Lovison. Jon Krosnick is University Fellow at Resources for the Future.

Author Contributions

• Conceptualization: JAK LC.
• Data curation: JAK NM CHM LC JP RKT.
• Formal analysis: NM CHM LC JP.
• Funding acquisition: JAK RKT.
• Investigation: JAK LC RKT.
• Methodology: NM LC JP.
• Project administration: JAK NM CHM.
• Resources: JAK RKT.
• Software: NM CHM LC JP RKT.
• Supervision: JAK.
• Validation: NM CHM JP.
• Visualization: NM CHM LC JP.
• Writing – original draft: JAK NM CHM EFB JP.
• Writing – review & editing: JAK NM CHM EFB JP.
• 1. Centers for Disease Control and Prevention. Cigarette smoking among adults—United States, 2007.
• 2. Viscusi WK. Smoking: Making the risky decision. New York: Oxford University Press; 1992.
• View Article
• Google Scholar
• PubMed/NCBI
• 6. Newport F, Moore DW, Saad L (Gallup O. Long term Gallup Poll trends: A portrait of American public opinion through the century [Internet]. http://www.gallup.com/poll/3400/longterm-gallup-poll-trends-portrait-american-public-opinion.aspx
• 7. Gallup Organization. National survey. Retrieved from iPOLL Databank, The Roper Center for Public Opinion Research, University of Connecticut. http://www.ropercenter.uconn.edu/ipoll.html . Accessed August 1, 2014.
• 8. Department of Health and Human Services, Public Health Service, Office of the Assistant Secretary for Health, Office on Smoking and Health. Use of Tobacco Survey (ARC Identifier 607143),1986.
• 9. American Lung Association and Gallup Organization. National survey, June 1987. Retrieved from iPOLL Databank, The Roper Center for Public Opinion Research, University of Connecticut. http://www.ropercenter.uconn.edu/ipoll.html . Accessed.
• 12. Manski CF. Identification problems in the social sciences. Cambridge, MA: Harvard University Press; 1995.
• 14. Mausner JS, Bahn JK. Epidemiology: An introductory text. Philadelphia, PA: Saunders; 1974.
• 47. Festinger L. A theory of cognitive dissonance. Stanford, CA: Stanford University Press; 1957.
• 49. Petty RE, Krosnick JA. Attitude strength: Antecedents and consequences. Hillsdale, NJ: Erlbaum; 1995.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 10 October 2022

Health effects associated with smoking: a Burden of Proof study

• Xiaochen Dai   ORCID: orcid.org/0000-0002-0289-7814 1 , 2 ,
• Gabriela F. Gil 1 ,
• Marissa B. Reitsma 1 ,
• Noah S. Ahmad 1 ,
• Jason A. Anderson 1 ,
• Catherine Bisignano 1 ,
• Sinclair Carr 1 ,
• Rachel Feldman 1 ,
• Simon I. Hay   ORCID: orcid.org/0000-0002-0611-7272 1 , 2 ,
• Jiawei He 1 , 2 ,
• Vincent Iannucci 1 ,
• Hilary R. Lawlor 1 ,
• Matthew J. Malloy 1 ,
• Laurie B. Marczak 1 ,
• Susan A. McLaughlin 1 ,
• Larissa Morikawa   ORCID: orcid.org/0000-0001-9749-8033 1 ,
• Erin C. Mullany 1 ,
• Sneha I. Nicholson 1 ,
• Erin M. O’Connell 1 ,
• Chukwuma Okereke 1 ,
• Reed J. D. Sorensen 1 ,
• Joanna Whisnant 1 ,
• Aleksandr Y. Aravkin 1 , 3 ,
• Peng Zheng 1 , 2 ,
• Christopher J. L. Murray   ORCID: orcid.org/0000-0002-4930-9450 1 , 2 &
• Emmanuela Gakidou   ORCID: orcid.org/0000-0002-8992-591X 1 , 2  

Nature Medicine volume  28 ,  pages 2045–2055 ( 2022 ) Cite this article

29k Accesses

56 Citations

169 Altmetric

Metrics details

• Risk factors

Matters Arising to this article was published on 14 April 2023

As a leading behavioral risk factor for numerous health outcomes, smoking is a major ongoing public health challenge. Although evidence on the health effects of smoking has been widely reported, few attempts have evaluated the dose–response relationship between smoking and a diverse range of health outcomes systematically and comprehensively. In the present study, we re-estimated the dose–response relationships between current smoking and 36 health outcomes by conducting systematic reviews up to 31 May 2022, employing a meta-analytic method that incorporates between-study heterogeneity into estimates of uncertainty. Among the 36 selected outcomes, 8 had strong-to-very-strong evidence of an association with smoking, 21 had weak-to-moderate evidence of association and 7 had no evidence of association. By overcoming many of the limitations of traditional meta-analyses, our approach provides comprehensive, up-to-date and easy-to-use estimates of the evidence on the health effects of smoking. These estimates provide important information for tobacco control advocates, policy makers, researchers, physicians, smokers and the public.

Similar content being viewed by others

A systematic review and network meta-analysis of population-level interventions to tackle smoking behaviour

The Burden of Proof studies: assessing the evidence of risk

Health effects associated with exposure to secondhand smoke: a Burden of Proof study

Among both the public and the health experts, smoking is recognized as a major behavioral risk factor with a leading attributable health burden worldwide. The health risks of smoking were clearly outlined in a canonical study of disease rates (including lung cancer) and smoking habits in British doctors in 1950 and have been further elaborated in detail over the following seven decades 1 , 2 . In 2005, evidence of the health consequences of smoking galvanized the adoption of the first World Health Organization (WHO) treaty, the Framework Convention on Tobacco Control, in an attempt to drive reductions in global tobacco use and second-hand smoke exposure 3 . However, as of 2020, an estimated 1.18 billion individuals globally were current smokers and 7 million deaths and 177 million disability-adjusted life-years were attributed to smoking, reflecting a persistent public health challenge 4 . Quantifying the relationship between smoking and various important health outcomes—in particular, highlighting any significant dose–response relationships—is crucial to understanding the attributable health risk experienced by these individuals and informing responsive public policy.

Existing literature on the relationship between smoking and specific health outcomes is prolific, including meta-analyses, cohort studies and case–control studies analyzing the risk of outcomes such as lung cancer 5 , 6 , 7 , chronic obstructive pulmonary disease (COPD) 8 , 9 , 10 and ischemic heart disease 11 , 12 , 13 , 14 due to smoking. There are few if any attempts, however, to systematically and comprehensively evaluate the landscape of evidence on smoking risk across a diverse range of health outcomes, with most current research focusing on risk or attributable burden of smoking for a specific condition 7 , 15 , thereby missing the opportunity to provide a comprehensive picture of the health risk experienced by smokers. Furthermore, although evidence surrounding specific health outcomes, such as lung cancer, has generated widespread consensus, findings about the attributable risk of other outcomes are much more heterogeneous and inconclusive 16 , 17 , 18 . These studies also vary in their risk definitions, with many comparing dichotomous exposure measures of ever smokers versus nonsmokers 19 , 20 . Others examine the distinct risks of current smokers and former smokers compared with never smokers 21 , 22 , 23 . Among the studies that do analyze dose–response relationships, there is large variation in the units and dose categories used in reporting their findings (for example, the use of pack-years or cigarettes per day) 24 , 25 , which complicates the comparability and consolidation of evidence. This, in turn, can obscure data that could inform personal health choices, public health practices and policy measures. Guidance on the health risks of smoking, such as the Surgeon General’s Reports on smoking 26 , 27 , is often based on experts’ evaluation of heterogenous evidence, which, although extremely useful and well suited to carefully consider nuances in the evidence, is fundamentally subjective.

The present study, as part of the Global Burden of Diseases, Risk Factors, and Injuries Study (GBD) 2020, re-estimated the continuous dose–response relationships (the mean risk functions and associated uncertainty estimates) between current smoking and 36 health outcomes (Supplementary Table 1 ) by identifying input studies using a systematic review approach and employing a meta-analytic method 28 . The 36 health outcomes that were selected based on existing evidence of a relationship included 16 cancers (lung cancer, esophageal cancer, stomach cancer, leukemia, liver cancer, laryngeal cancer, breast cancer, cervical cancer, colorectal cancer, lip and oral cavity cancer, nasopharyngeal cancer, other pharynx cancer (excluding nasopharynx cancer), pancreatic cancer, bladder cancer, kidney cancer and prostate cancer), 5 cardiovascular diseases (CVDs: ischemic heart disease, stroke, atrial fibrillation and flutter, aortic aneurysm and peripheral artery disease) and 15 other diseases (COPD, lower respiratory tract infections, tuberculosis, asthma, type 2 diabetes, Alzheimer’s disease and related dementias, Parkinson’s disease, multiple sclerosis, cataracts, gallbladder diseases, low back pain, peptic ulcer disease, rheumatoid arthritis, macular degeneration and fractures). Definitions of the outcomes are described in Supplementary Table 1 . We conducted a separate systematic review for each risk–outcome pair with the exception of cancers, which were done together in a single systematic review. This approach allowed us to systematically identify all relevant studies indexed in PubMed up to 31 May 2022, and we extracted relevant data on risk of smoking, including study characteristics, following a pre-specified template (Supplementary Table 2 ). The meta-analytic tool overcomes many of the limitations of traditional meta-analyses by incorporating between-study heterogeneity into the uncertainty of risk estimates, accounting for small numbers of studies, relaxing the assumption of log(linearity) applied to the risk functions, handling differences in exposure ranges between comparison groups, and systematically testing and adjusting for bias due to study designs and characteristics. We then estimated the burden-of-proof risk function (BPRF) for each risk–outcome pair, as proposed by Zheng et al. 29 ; the BPRF is a conservative risk function defined as the 5th quantile curve (for harmful risks) that reflects the smallest harmful effect at each level of exposure consistent with the available evidence. Given all available data for each outcome, the risk of smoking is at least as harmful as the BPRF indicates.

We used the BPRF for each risk–outcome pair to calculate risk–outcome scores (ROSs) and categorize the strength of evidence for the association between smoking and each health outcome using a star rating from 1 to 5. The interpretation of the star ratings is as follows: 1 star (*) indicates no evidence of association; 2 stars (**) correspond to a 0–15% increase in risk across average range of exposures for harmful risks; 3 stars (***) represent a 15–50% increase in risk; 4 stars (****) refer to >50–85% increase in risk; and 5 stars (*****) equal >85% increase in risk. The thresholds for each star rating were developed in consultation with collaborators and other stakeholders.

The increasing disease burden attributable to current smoking, particularly in low- and middle-income countries 4 , demonstrates the relevance of the present study, which quantifies the strength of the evidence using an objective, quantitative, comprehensive and comparative framework. Findings from the present study can be used to support policy makers in making informed smoking recommendations and regulations focusing on the associations for which the evidence is strongest (that is, the 4- and 5-star associations). However, associations with a lower star rating cannot be ignored, especially when the outcome has high prevalence or severity. A summary of the main findings, limitations and policy implications of the study is presented in Table 1 .

We evaluated the mean risk functions and the BPRFs for 36 health outcomes that are associated with current smoking 30 (Table 2 ). Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines 31 for each of our systematic reviews, we identified studies reporting relative risk (RR) of incidence or mortality from each of the 36 selected outcomes for smokers compared with nonsmokers. We reviewed 21,108 records, which were identified to have been published between 1 May 2018 and 31 May 2022; this represents the most recent time period since the last systematic review of the available evidence for the GBD at the time of publication. The meta-analyses reported in the present study for each of the 36 health outcomes are based on evidence from a total of 793 studies published between 1970 and 2022 (Extended Data Fig. 1 – 5 and Supplementary Information 1.5 show the PRISMA diagrams for each outcome). Only prospective cohort and case–control studies were included for estimating dose–response risk curves, but cross-sectional studies were also included for estimating the age pattern of smoking risk on cardiovascular and circulatory disease (CVD) outcomes. Details on each, including the study’s design, data sources, number of participants, length of follow-up, confounders adjusted for in the input data and bias covariates included in the dose–response risk model, can be found in Supplementary Information 2 and 3 . The theoretical minimum risk exposure level used for current smoking was never smoking or zero 30 .

Five-star associations

When the most conservative interpretation of the evidence, that is, the BPRF, suggests that the average exposure (15th–85th percentiles of exposure) of smoking increases the risk of a health outcome by >85% (that is, ROS > 0.62), smoking and that outcome are categorized as a 5-star pair. Among the 36 outcomes, there are 5 that have a 5-star association with current smoking: laryngeal cancer (375% increase in risk based on the BPRF, 1.56 ROS), aortic aneurysm (150%, 0.92), peripheral artery disease (137%, 0.86), lung cancer (107%, 0.73) and other pharynx cancer (excluding nasopharynx cancer) (92%, 0.65).

Results for all 5-star risk–outcome pairs are available in Table 2 and Supplementary Information 4.1 . In the present study, we provide detailed results for one example 5-star association: current smoking and lung cancer. We extracted 371 observations from 25 prospective cohort studies and 53 case–control studies across 25 locations (Supplementary Table 3 ) 5 , 6 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 , 64 , 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 , 90 , 91 , 92 , 93 , 94 , 95 , 96 , 97 , 98 , 99 , 100 , 101 , 102 , 103 , 104 , 105 , 106 , 107 . Exposure ranged from 1 pack-year to >112 pack-years, with the 85th percentile of exposure being 50.88 pack-years (Fig. 1a ).

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on the x axis and the estimated s.d. that includes reported s.d. and between-study heterogeneity on the y axis.

We found a very strong and significant harmful relationship between pack-years of current smoking and the RR of lung cancer (Fig. 1b ). The mean RR of lung cancer at 20 pack-years of smoking was 5.11 (95% uncertainty interval (UI) inclusive of between-study heterogeneity = 1.84–14.99). At 50.88 pack-years (85th percentile of exposure), the mean RR of lung cancer was 13.42 (2.63–74.59). See Table 2 for mean RRs at other exposure levels. The BPRF, which represents the most conservative interpretation of the evidence (Fig. 1a ), suggests that smoking in the 15th–85th percentiles of exposure increases the risk of lung cancer by an average of 107%, yielding an ROS of 0.73.

The relationship between pack-years of current smoking and RR of lung cancer is nonlinear, with diminishing impact of further pack-years of smoking, particularly for middle-to-high exposure levels (Fig. 1b ). To reduce the effect of bias, we adjusted observations that did not account for more than five confounders, including age and sex, because they were the significant bias covariates identified by the bias covariate selection algorithm 29 (Supplementary Table 7 ). The reported RRs across studies were very heterogeneous. Our meta-analytic method, which accounts for the reported uncertainty in both the data and between-study heterogeneity, fit the data and covered the estimated residuals well (Fig. 1c ). After trimming 10% of outliers, we still detected publication bias in the results for lung cancer. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 5-star pairs.

Four-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of a health outcome by 50–85% (that is, ROS > 0.41–0.62), smoking is categorized as having a 4-star association with that outcome. We identified three outcomes with a 4-star association with smoking: COPD (72% increase in risk based on the BPRF, 0.54 ROS), lower respiratory tract infection (54%, 0.43) and pancreatic cancer (52%, 0.42).

In the present study, we provide detailed results for one example 4-star association: current smoking and COPD. We extracted 51 observations from 11 prospective cohort studies and 4 case–control studies across 36 locations (Supplementary Table 3 ) 6 , 8 , 9 , 10 , 78 , 108 , 109 , 110 , 111 , 112 , 113 , 114 , 115 , 116 . Exposure ranged from 1 pack-year to 100 pack-years, with the 85th percentile of exposure in the exposed group being 49.75 pack-years.

We found a strong and significant harmful relationship between pack-years of current smoking and RR of COPD (Fig. 2b ). The mean RR of COPD at 20 pack-years was 3.17 (1.60–6.55; Table 2 reports RRs at other exposure levels). At the 85th percentile of exposure, the mean RR of COPD was 6.01 (2.08–18.58). The BPRF suggests that average smoking exposure raises the risk of COPD by an average of 72%, yielding an ROS of 0.54. The results for the other health outcomes that have an association with smoking rated as 4 stars are shown in Table 2 and Supplementary Information 4.2 .

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on th e x axis and the estimated s.d. that includes the reported s.d. and between-study heterogeneity on the y axis.

The relationship between smoking and COPD is nonlinear, with diminishing impact of further pack-years of current smoking on risk of COPD, particularly for middle-to-high exposure levels (Fig. 2a ). To reduce the effect of bias, we adjusted observations that did not account for age and sex and/or were generated for individuals aged >65 years 116 , because they were the two significant bias covariates identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was large heterogeneity in the reported RRs across studies, and our meta-analytic method fit the data and covered the estimated residuals well (Fig. 2b ). Although we trimmed 10% of outliers, publication bias was still detected in the results for COPD. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for reported RR data and alternative exposures across studies for the remaining health outcomes that have a 4-star association with smoking.

Three-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of a health outcome by 15–50% (or, when protective, decreases the risk of an outcome by 13–34%; that is, ROS >0.14–0.41), the association between smoking and that outcome is categorized as having a 3-star rating. We identified 15 outcomes with a 3-star association: bladder cancer (40% increase in risk, 0.34 ROS); tuberculosis (31%, 0.27); esophageal cancer (29%, 0.26); cervical cancer, multiple sclerosis and rheumatoid arthritis (each 23–24%, 0.21); lower back pain (22%, 0.20); ischemic heart disease (20%, 0.19); peptic ulcer and macular degeneration (each 19–20%, 0.18); Parkinson's disease (protective risk, 15% decrease in risk, 0.16); and stomach cancer, stroke, type 2 diabetes and cataracts (each 15–17%, 0.14–0.16).

We present the findings on smoking and type 2 diabetes as an example of a 3-star risk association. We extracted 102 observations from 24 prospective cohort studies and 4 case–control studies across 15 locations (Supplementary Table 3 ) 117 , 118 , 119 , 120 , 121 , 122 , 123 , 124 , 125 , 126 , 127 , 128 , 129 , 130 , 131 , 132 , 133 , 134 , 135 , 136 , 137 , 138 , 139 , 140 , 141 , 142 , 143 , 144 . The exposure ranged from 1 cigarette to 60 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 26.25 cigarettes smoked per day.

We found a moderate and significant harmful relationship between cigarettes smoked per day and the RR of type 2 diabetes (Fig. 3b ). The mean RR of type 2 diabetes at 20 cigarettes smoked per day was 1.49 (1.18–1.90; see Table 2 for other exposure levels). At the 85th percentile of exposure, the mean RR of type 2 diabetes was 1.54 (1.20–2.01). The BPRF suggests that average smoking exposure raises the risk of type 2 diabetes by an average of 16%, yielding an ROS of 0.15. See Table 2 and Supplementary Information 4.3 for results for the additional health outcomes with an association with smoking rated as 3 stars.

a , The log(RR) function. b , RR function. c , A modified funnel plot showing the residuals (relative to 0) on the x axis and the estimated s.d. that includes the reported s.d. and between-study heterogeneity on the y axis.

The relationship between smoking and type 2 diabetes is nonlinear, particularly for high exposure levels where the mean risk curve becomes flat (Fig. 3a ). We adjusted observations that were generated in subpopulations, because it was the only significant bias covariate identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was moderate heterogeneity in the observed RR data across studies and our meta-analytic method fit the data and covered the estimated residuals extremely well (Fig. 3b,c ). After trimming 10% of outliers, we still detected publication bias in the results for type 2 diabetes. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 3-star pairs.

Two-star associations

When the BPRF suggests that the average exposure of smoking increases the risk of an outcome by 0–15% (that is, ROS 0.0–0.14), the association between smoking and that outcome is categorized as a 2-star rating. We identified six 2-star outcomes: nasopharyngeal cancer (14% increase in risk, 0.13 ROS); Alzheimer’s and other dementia (10%, 0.09); gallbladder diseases and atrial fibrillation and flutter (each 6%, 0.06); lip and oral cavity cancer (5%, 0.05); and breast cancer (4%, 0.04).

We present the findings on smoking and breast cancer as an example of a 2-star association. We extracted 93 observations from 14 prospective cohort studies and 9 case–control studies across 14 locations (Supplementary Table 3 ) 84 , 87 , 145 , 146 , 147 , 148 , 149 , 150 , 151 , 152 , 153 , 154 , 155 , 156 , 157 , 158 , 159 , 160 , 161 , 162 , 163 , 164 , 165 . The exposure ranged from 1 cigarette to >76 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 34.10 cigarettes smoked per day.

We found a weak but significant relationship between pack-years of current smoking and RR of breast cancer (Extended Data Fig. 6 ). The mean RR of breast cancer at 20 pack-years was 1.17 (1.04–1.31; Table 2 reports other exposure levels). The BPRF suggests that average smoking exposure raises the risk of breast cancer by an average of 4%, yielding an ROS of 0.04. See Table 2 and Supplementary Information 4.4 for results on the additional health outcomes for which the association with smoking has been categorized as 2 stars.

The relationship between smoking and breast cancer is nonlinear, particularly for high exposure levels where the mean risk curve becomes flat (Extended Data Fig. 6a ). To reduce the effect of bias, we adjusted observations that were generated in subpopulations, because it was the only significant bias covariate identified by the bias covariate selection algorithm (Supplementary Table 7 ). There was heterogeneity in the reported RRs across studies, but our meta-analytic method fit the data and covered the estimated residuals (Extended Data Fig. 6b ). After trimming 10% of outliers, we did not detect publication bias in the results for breast cancer. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 2-star pairs.

One-star associations

When average exposure to smoking does not significantly increase (or decrease) the risk of an outcome, once between-study heterogeneity and other sources of uncertainty are accounted for (that is, ROS < 0), the association between smoking and that outcome is categorized as 1 star, indicating that there is not sufficient evidence for the effect of smoking on the outcome to reject the null (that is, there may be no association). There were seven outcomes with an association with smoking that rated as 1 star: colorectal and kidney cancer (each –0.01 ROS); leukemia (−0.04); fractures (−0.05); prostate cancer (−0.06); liver cancer (−0.32); and asthma (−0.64).

We use smoking and prostate cancer as examples of a 1-star association. We extracted 78 observations from 21 prospective cohort studies and 1 nested case–control study across 15 locations (Supplementary Table 3 ) 157 , 160 , 166 , 167 , 168 , 169 , 170 , 171 , 172 , 173 , 174 , 175 , 176 , 177 , 178 , 179 , 180 , 181 , 182 , 183 , 184 , 185 . The exposure among the exposed group ranged from 1 cigarette to 90 cigarettes smoked per day, with the 85th percentile of exposure in the exposed group being 29.73 cigarettes smoked per day.

Based on our conservative interpretation of the data, we did not find a significant relationship between cigarettes smoked per day and the RR of prostate cancer (Fig. 4B ). The exposure-averaged BPRF for prostate cancer was 0.94, which was opposite null from the full range of mean RRs, such as 1.16 (0.89–1.53) at 20 cigarettes smoked per day. The corresponding ROS was −0.06, which is consistent with no evidence of an association between smoking and increased risk of prostate cancer. See Table 2 and Supplementary Information 4.5 for results for the additional outcomes that have a 1-star association with smoking.

The relationship between smoking and prostate cancer is nonlinear, particularly for middle-to-high exposure levels where the mean risk curve becomes flat (Fig. 4a ). We did not adjust for any bias covariate because no significant bias covariates were selected by the algorithm (Supplementary Table 7 ). The RRs reported across studies were very heterogeneous, but our meta-analytic method fit the data and covered the estimated residuals well (Fig. 4b,c ). The ROS associated with the BPRF is −0.05, suggesting that the most conservative interpretation of all evidence, after accounting for between-study heterogeneity, indicates an inconclusive relationship between smoking exposure and the risk of prostate cancer. After trimming 10% of outliers, we still detected publication bias in the results for prostate cancer, which warrants further studies using sample populations. See Supplementary Tables 4 and 7 for study bias characteristics and selected bias covariates, Supplementary Fig. 5 for results without 10% trimming and Supplementary Table 8 for observed RR data and alternative exposures across studies for the remaining 1-star pairs.

Age-specific dose–response risk for CVD outcomes

We produced age-specific dose–response risk curves for the five selected CVD outcomes ( Methods ). The ROS associated with each smoking–CVD pair was calculated based on the reference risk curve estimated using all risk data regardless of age information. Estimation of the BPRF, calculation of the associated ROS and star rating of the smoking–CVD pairs follow the same rules as the other non-CVD smoking–outcome pairs (Table 1 and Supplementary Figs. 2 – 4 ). Once we had estimated the reference dose–response risk curve for each CVD outcome, we determined the age group of the reference risk curve. The reference age group is 55–59 years for all CVD outcomes, except for peripheral artery disease, the reference age group for which is 60–64 years. We then estimated the age pattern of smoking on all CVD outcomes (Supplementary Fig. 2 ) and calculated age attenuation factors of the risk for each age group by comparing the risk of each age group with that of the reference age group, using the estimated age pattern (Supplementary Fig. 3 ). Last, we applied the draws of age attenuation factors of each age group to the dose–response risk curve for the reference age group to produce the age group-specific dose–response risk curves for each CVD outcome (Supplementary Fig. 4 ).

Using our burden-of-proof meta-analytic methods, we re-estimated the dose–response risk of smoking on 36 health outcomes that had previously been demonstrated to be associated with smoking 30 , 186 . Using these methods, which account for both the reported uncertainty of the data and the between-study heterogeneity, we found that 29 of the 36 smoking–outcome pairs are supported by evidence that suggests a significant dose–response relationship between smoking and the given outcome (28 with a harmful association and 1 with a protective association). Conversely, after accounting for between-study heterogeneity, the available evidence of smoking risk on seven outcomes (that is, colon and rectum cancer, kidney cancer, leukemia, prostate cancer, fractures, liver cancer and asthma) was insufficient to reject the null or draw definitive conclusions on their relationship to smoking. Among the 29 outcomes that have evidence supporting a significant relationship to smoking, 8 had strong-to-very-strong evidence of a relationship, meaning that, given all the available data on smoking risk, we estimate that average exposure to smoking increases the risk of those outcomes by >50% (4- and 5-star outcomes). The currently available evidence for the remaining 21 outcomes with a significant association with current smoking was weak to moderate, indicating that smoking increases the risk of those outcomes by at least >0–50% (2- and 3-star associations).

Even under our conservative interpretation of the data, smoking is irrefutably harmful to human health, with the greatest increases in risk occurring for laryngeal cancer, aortic aneurysm, peripheral artery disease, lung cancer and other pharynx cancer (excluding nasopharynx cancer), which collectively represent large causes of death and ill-health. The magnitude of and evidence for the associations between smoking and its leading health outcomes are among the highest currently analyzed in the burden-of-proof framework 29 . The star ratings assigned to each smoking–outcome pair offer policy makers a way of categorizing and comparing the evidence for a relationship between smoking and its potential health outcomes ( https://vizhub.healthdata.org/burden-of-proof ). We found that, for seven outcomes in our analysis, there was insufficient or inconsistent evidence to demonstrate a significant association with smoking. This is a key finding because it demonstrates the need for more high-quality data for these particular outcomes; availability of more data should improve the strength of evidence for whether or not there is an association between smoking and these health outcomes.

Our systematic review approach and meta-analytic methods have numerous benefits over existing systematic reviews and meta-analyses on the same topic that use traditional random effects models. First, our approach relaxes the log(linear) assumption, using a spline ensemble to estimate the risk 29 . Second, our approach allows variable reference groups and exposure ranges, allowing for more accurate estimates regardless of whether or not the underlying relative risk is log(linear). Furthermore, it can detect outliers in the data automatically. Finally, it quantifies uncertainty due to between-study heterogeneity while accounting for small numbers of studies, minimizing the risk that conclusions will be drawn based on spurious findings.

We believe that the results for the association between smoking and each of the 36 health outcomes generated by the present study, including the mean risk function, BPRF, ROS, average excess risk and star rating, could be useful to a range of stakeholders. Policy makers can formulate their decisions on smoking control priorities and resource allocation based on the magnitude of the effect and the consistency of the evidence relating smoking to each of the 36 outcomes, as represented by the ROS and star rating for each smoking–outcome association 187 . Physicians and public health practitioners can use the estimates of average increased risk and the star rating to educate patients and the general public about the risk of smoking and to promote smoking cessation 188 . Researchers can use the estimated mean risk function or BPRF to obtain the risk of an outcome at a given smoking exposure level, as well as uncertainty surrounding that estimate of risk. The results can also be used in the estimation of risk-attributable burden, that is, the deaths and disability-adjusted life-years due to each outcome that are attributable to smoking 30 , 186 . For the general public, these results could help them to better understand the risk of smoking and manage their health 189 .

Although our meta-analysis was comprehensive and carefully conducted, there are limitations to acknowledge. First, the bias covariates used, although carefully extracted and evaluated, were based on observable study characteristics and thus may not fully capture unobserved characteristics such as study quality or context, which might be major sources of bias. Second, if multiple risk estimates with different adjustment levels were reported in a given study, we included only the fully adjusted risk estimate and modeled the adjustment level according to the number of covariates adjusted for (rather than which covariates were adjusted for) and whether a standard adjustment for age and sex had been applied. This approach limited our ability to make full use of all available risk estimates in the literature. Third, although we evaluated the potential for publication bias in the data, we did not test for other forms of bias such as when studies are more consistent with each other than expected by chance 29 . Fourth, our analysis assumes that the relationships between smoking and health outcomes are similar across geographical regions and over time. We do not have sufficient evidence to quantify how the relationships may have evolved over time because the composition of smoking products has also changed over time. Perhaps some of the heterogeneity of the effect sizes in published studies reflects this; however, this cannot be discerned with the currently available information.

In the future, we plan to include crude and partially adjusted risk estimates in our analyses to fully incorporate all available risk estimates, to model the adjusted covariates in a more comprehensive way by mapping the adjusted covariates across all studies comprehensively and systematically, and to develop methods to evaluate additional forms of potential bias. We plan to update our results on a regular basis to provide timely and up-to-date evidence to stakeholders.

To conclude, we have re-estimated the dose–response risk of smoking on 36 health outcomes while synthesizing all the available evidence up to 31 May 2022. We found that, even after factoring in the heterogeneity between studies and other sources of uncertainty, smoking has a strong-to-very-strong association with a range of health outcomes and confirmed that smoking is irrefutably highly harmful to human health. We found that, due to small numbers of studies, inconsistency in the data, small effect sizes or a combination of these reasons, seven outcomes for which some previous research had found an association with smoking did not—under our meta-analytic framework and conservative approach to interpreting the data—have evidence of an association. Our estimates of the evidence for risk of smoking on 36 selected health outcomes have the potential to inform the many stakeholders of smoking control, including policy makers, researchers, public health professionals, physicians, smokers and the general public.

For the present study, we used a meta-analytic tool, MR-BRT (metaregression—Bayesian, regularized, trimmed), to estimate the dose–response risk curves of the risk of a health outcome across the range of current smoking levels along with uncertainty estimates 28 . Compared with traditional meta-analysis using linear mixed effect models, MR-BRT relaxes the assumption of a log(linear) relationship between exposure and risk, incorporates between-study heterogeneity into the uncertainty of risk estimates, handles estimates reported across different exposure categories, automatically identifies and trims outliers, and systematically tests and adjusts for bias due to study designs and characteristics. The meta-analytic methods employed by the present study followed the six main steps proposed by Zheng et al. 28 , 29 , namely: (1) enacting a systematic review approach and data extraction following a pre-specified and standardized protocol; (2) estimating the shape of the relationship between exposure and RR; (3) evaluating and adjusting for systematic bias as a function of study characteristics and risk estimation; (4) quantifying between-study heterogeneity while adjusting for within-study correlation and the number of studies; (5) evaluating potential publication or reporting biases; and (6) estimating the mean risk function and the BPRF, calculating the ROS and categorizing smoking–outcome pairs using a star-rating scheme from 1 to 5.

The estimates for our primary indicators of this work—mean RRs across a range of exposures, BRPFs, ROSs and star ratings for each risk–outcome pair—are not specific to or disaggregated by specific populations. We did not estimate RRs separately for different locations, sexes (although the RR of prostate cancer was estimated only for males and of cervical and breast cancer only for females) or age groups (although this analysis was applied to disease endpoints in adults aged ≥30 years only and, as detailed below, age-specific estimates were produced for the five CVD outcomes).

The present study complies with the PRISMA guidelines 190 (Supplementary Tables 9 and 10 and Supplementary Information 1.5 ) and Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations 191 (Supplementary Table 11 ). The study was approved by the University of Washington Institutional Review Board (study no. 9060). The systematic review approach was not registered.

Selecting health outcomes

In the present study, current smoking is defined as the current use of any smoked tobacco product on a daily or occasional basis. Health outcomes were initially selected using the World Cancer Research Fund criteria for convincing or probable evidence as described in Murray et al. 186 . The 36 health outcomes that were selected based on existing evidence of a relationship included 16 cancers (lung cancer, esophageal cancer, stomach cancer, leukemia, liver cancer, laryngeal cancer, breast cancer, cervical cancer, colorectal cancer, lip and oral cavity cancer, nasopharyngeal cancer, other pharynx cancer (excluding nasopharynx cancer), pancreatic cancer, bladder cancer, kidney cancer and prostate cancer), 5 CVDs (ischemic heart disease, stroke, atrial fibrillation and flutter, aortic aneurysm and peripheral artery disease) and 15 other diseases (COPD, lower respiratory tract infections, tuberculosis, asthma, type 2 diabetes, Alzheimer’s disease and related dementias, Parkinson’s disease, multiple sclerosis, cataracts, gallbladder diseases, low back pain, peptic ulcer disease, rheumatoid arthritis, macular degeneration and fracture). Definitions of the outcomes are described in Supplementary Table 1 .

Step 1: systematic review approach to literature search and data extraction

Informed by the systematic review approach we took for the GBD 2019 (ref. 30 ), for the present study we identified input studies in the literature using a systematic review approach for all 36 smoking–outcome pairs using updated search strings to identify all relevant studies indexed in PubMed up to 31 May 2022 and extracted data on smoking risk estimates. Briefly, the studies that were extracted represented several types of study design (for example, cohort and case–control studies), measured exposure in several different ways and varied in their choice of reference categories (where some compared current smokers with never smokers, whereas others compared current smokers with nonsmokers or former smokers). All these study characteristics were catalogued systematically and taken into consideration during the modeling part of the analysis.

In addition, for CVD outcomes, we also estimated the age pattern of risk associated with smoking. We applied a systematic review of literature approach for smoking risk for the five CVD outcomes. We developed a search string to search for studies reporting any association between binary smoking status (that is, current, former and ever smokers) and the five CVD outcomes from 1 January 1970 to 31 May 2022, and included only studies reporting age-specific risk (RR, odds ratio (OR), hazard ratio (HR)) of smoking status. The inclusion criteria and results of the systematic review approach are reported in accordance with PRISMA guidelines 31 . Details for each outcome on the search string used in the systematic review approach, refined inclusion and exclusion criteria, data extraction template and PRISMA diagram are given in Supplementary Information 1 . Title and/or abstract screening, full text screening and data extraction were conducted by 14 members of the research team and extracted data underwent manual quality assurance by the research team to verify accuracy.

Selecting exposure categories

Cumulative exposure in pack-years was the measure of exposure used for COPD and all cancer outcomes except for prostate cancer, to reflect the risk of both duration and intensity of current smoking on these outcomes. For prostate cancer, CVDs and all the other outcomes except for fractures, we used cigarette-equivalents smoked per day as the exposure for current smoking, because smoking intensity is generally thought to be more important than duration for these outcomes. For fractures, we used binary exposure, because there were few studies examining intensity or duration of smoking on fractures. The smoking–outcome pairs and the corresponding exposures are summarized in Supplementary Table 4 and are congruent with the GBD 2019 (refs. 30 , 186 ).

Steps 2–5: modeling dose–response RR of smoking on the selected health outcomes

Of the six steps proposed by Zheng et al. 29 , steps 2–5 cover the process of modeling dose–response risk curves. In step 2, we estimated the shape (or the ‘signal’) of the dose–response risk curves, integrating over different exposure ranges. To relax the log(linear) assumption usually applied to continuous dose–response risk and make the estimates robust to the placement of spline knots, we used an ensemble spline approach to fit the functional form of the dose–response relationship. The final ensemble model was a weighted combination of 50 models with random knot placement, with the weight of each model proportional to measures of model fit and total variation. To avoid the influence of extreme data and reduce publication bias, we trimmed 10% of data for each outcome as outliers. We also applied a monotonicity constraint to ensure that the mean risk curves were nondecreasing (or nonincreasing in the case of Parkinson’s disease).

In step 3, following the GRADE approach 192 , 193 , we quantified risk of bias across six domains, namely, representativeness of the study population, exposure, outcome, reverse causation, control for confounding and selection bias. Details about the bias covariates are provided in Supplementary Table 4 . We systematically tested for the effect of bias covariates using metaregression, selected significant bias covariates using the Lasso approach 194 , 195 and adjusted for the selected bias covariates in the final risk curve.

In step 4, we quantified between-study heterogeneity accounting for within-study correlation, uncertainty of the heterogeneity, as well as small number of studies. Specifically, we used a random intercept in the mixed-effects model to account for the within-study correlation and used a study-specific random slope with respect to the ‘signal’ to capture between-study heterogeneity. As between-study heterogeneity can be underestimated or even zero when the number of studies is small 196 , 197 , we used Fisher’s information matrix to estimate the uncertainty of the heterogeneity 198 and incorporated that uncertainty into the final results.

In step 5, in addition to generating funnel plots and visually inspecting for asymmetry (Figs. 1c , 2c , 3c and 4c and Extended Data Fig. 6c ) to identify potential publication bias, we also statistically tested for potential publication or reporting bias using Egger’s regression 199 . We flagged potential publication bias in the data but did not correct for it, which is in line with the general literature 10 , 200 , 201 . Full details about the modeling process have been published elsewhere 29 and model specifications for each outcome are in Supplementary Table 6 .

Step 6: estimating the mean risk function and the BPRF

In the final step, step 6, the metaregression model inclusive of the selected bias covariates from step 3 (for example, the highest adjustment level) was used to predict the mean risk function and its 95% UI, which incorporated the uncertainty of the mean effect, between-study heterogeneity and the uncertainty in the heterogeneity estimate accounting for small numbers of studies. Specifically, 1,000 draws were created for each 0.1 level of doses from 0 pack-years to 100 pack-years or cigarette-equivalents smoked per day using the Bayesian metaregression model. The mean of the 1,000 draws was used to estimate the mean risk at each exposure level, and the 25th and 95th draws were used to estimate the 95% UIs for the mean risk at each exposure level.

The BPRF 29 is a conservative estimate of risk function consistent with the available evidence, correcting for both between-study heterogeneity and systemic biases related to study characteristics. The BPRF is defined as either the 5th (if harmful) or 95th (if protective) quantile curve closest to the line of log(RR) of 0, which defines the null (Figs. 1a , 2b , 3a and 4a ). The BPRF represents the smallest harmful (or protective) effect of smoking on the corresponding outcome at each level of exposure that is consistent with the available evidence. A BPRF opposite null from the mean risk function indicates that insufficient evidence is available to reject null, that is, that there may not be an association between risk and outcome. Likewise, the further the BPRF is from null on the same side of null as the mean risk function, the higher the magnitude and evidence for the relationship. The BPRF can be interpreted as indicating that, even accounting for between-study heterogeneity and its uncertainty, the log(RR) across the studied smoking range is at least as high as the BPRF (or at least as low as the BPRF for a protective risk).

To quantify the strength of the evidence, we calculated the ROS for each smoking–outcome association as the signed value of the log(BPRF) averaged between the 15th and 85th percentiles of observed exposure levels for each outcome. The ROS is a single summary of the effect of smoking on the outcome, with higher positive ROSs corresponding to stronger and more consistent evidence and a higher average effect size of smoking and a negative ROS, suggesting that, based on the available evidence, there is no significant effect of smoking on the outcome after accounting for between-study heterogeneity.

For ease of communication, we further classified each smoking–outcome association into a star rating from 1 to 5. Briefly, 1-star associations have an ROS <0, indicating that there is insufficient evidence to find a significant association between smoking and the selected outcome. We divided the positive ROSs into ranges 0.0–0.14 (2-star), >0.14–0.41 (3-star), >0.41–0.62 (4-star) and >0.62 (5-star). These categories correspond to excess risk ranges for harmful risks of 0–15%, >15–50%, >50–85% and >85%. For protective risks, the ranges of exposure-averaged decreases in risk by star rating are 0–13% (2 stars), >13–34% (3 stars), >34–46% (4 stars) and >46% (5 stars).

Among the 36 smoking–outcome pairs analyzed, smoking fracture was the only binary risk–outcome pair, which was due to limited data on the dose–response risk of smoking on fracture 202 . The estimation of binary risk was simplified because the RR was merely a comparison between current smokers and nonsmokers or never smokers. The concept of ROS for continuous risk can naturally extend to binary risk because the BPRF is still defined as the 5th percentile of the effect size accounting for data uncertainty and between-study heterogeneity. However, binary ROSs must be divided by 2 to make them comparable with continuous ROSs, which were calculated by averaging the risk over the range between the 15th and the 85th percentiles of observed exposure levels. Full details about estimating mean risk functions, BPRFs and ROSs for both continuous and binary risk–outcome pairs can be found elsewhere 29 .

Estimating the age-specific risk function for CVD outcomes

For non-CVD outcomes, we assumed that the risk function was the same for all ages and all sexes, except for breast, cervical and prostate cancer, which were assumed to apply only to females or males, respectively. As the risk of smoking on CVD outcomes is known to attenuate with increasing age 203 , 204 , 205 , 206 , we adopted a four-step approach for GBD 2020 to produce age-specific dose–response risk curves for CVD outcomes.

First, we estimated the reference dose–response risk of smoking for each CVD outcome using dose-specific RR data for each outcome regardless of the age group information. This step was identical to that implemented for the other non-CVD outcomes. Once we had generated the reference curve, we determined the age group associated with it by calculating the weighted mean age across all dose-specific RR data (weighted by the reciprocal of the s.e.m. of each datum). For example, if the weighted mean age of all dose-specific RR data was 56.5, we estimated the age group associated with the reference risk curve to be aged 55–59 years. For cohort studies, the age range associated with the RR estimate was calculated as a mean age at baseline plus the mean/median years of follow-up (if only the maximum years of follow-up were reported, we would halve this value and add it to the mean age at baseline). For case–control studies, the age range associated with the OR estimate was simply the reported mean age at baseline (if mean age was not reported, we used the midpoint of the age range instead).

In the third step, we extracted age group-specific RR data and relevant bias covariates from the studies identified in our systematic review approach of age-specific smoking risk on CVD outcomes, and used MR-BRT to model the age pattern of excess risk (that is, RR-1) of smoking on CVD outcomes with age group-specific excess RR data for all CVD outcomes. We modeled the age pattern of smoking risk on CVDs following the same steps we implemented for modeling dose–response risk curves. In the final model, we included a spline on age, random slope on age by study and the bias covariate encoding exposure definition (that is, current, former and ever smokers), which was picked by the variable selection algorithm 28 , 29 . When predicting the age pattern of the excess risk of smoking on CVD outcomes using the fitted model, we did not include between-study heterogeneity to reduce uncertainty in the prediction.

In the fourth step, we calculated the age attenuation factors of excess risk compared with the reference age group for each CVD outcome as the ratio of the estimated excess risk for each age group to the excess risk for the reference age group. We performed the calculation at the draw level to obtain 1,000 draws of the age attenuation factors for each age group. Once we had estimated the age attenuation factors, we carried out the last step, which consisted of adjusting the risk curve for the reference age group from step 1 using equation (1) to produce the age group-specific risk curves for each CVD outcome:

We implemented the age adjustment at the draw level so that the uncertainty of the age attenuation factors could be naturally incorporated into the final adjusted age-specific RR curves. A PRISMA diagram detailing the systematic review approach, a description of the studies included and the full details about the methods are in Supplementary Information 1.5 and 5.2 .

Estimating the theoretical minimum risk exposure level

The theoretical minimum risk exposure level for smoking was 0, that is, no individuals in the population are current or former smokers.

Model validation

The validity of the meta-analytic tool has been extensively evaluated by Zheng and colleagues using simulation experiments 28 , 29 . For the present study, we conducted two additional sensitivity analyses to examine how the shape of the risk curves was impacted by applying a monotonicity constraint and trimming 10% of data. We present the results of these sensitivity analyses in Supplementary Information 6 . In addition to the sensitivity analyses, the dose–response risk estimates were also validated by plotting the mean risk function along with its 95% UI against both the extracted dose-specific RR data from the studies included and our previous dose–response risk estimates from the GBD 2019 (ref. 30 ). The mean risk functions along with the 95% UIs were validated based on data fit and the level, shape and plausibility of the dose–response risk curves. All curves were validated by all authors and reviewed by an external expert panel, comprising professors with relevant experience from universities including Johns Hopkins University, Karolinska Institute and University of Barcelona; senior scientists working in relevant departments at the WHO and the Center for Disease Control and Prevention (CDC) and directors of nongovernmental organizations such as the Campaign for Tobacco-Free Kids.

Statistical analysis

Analyses were carried out using R v.3.6.3, Python v.3.8 and Stata v.16.

Statistics and reproducibility

The study was a secondary analysis of existing data involving systematic reviews and meta-analyses. No statistical method was used to predetermine sample size. As the study did not involve primary data collection, randomization and blinding, data exclusions were not relevant to the present study, and, as such, no data were excluded and we performed no randomization or blinding. We have made our data and code available to foster reproducibility.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability

The findings from the present study are supported by data available in the published literature. Data sources and citations for each risk–outcome pair can be downloaded using the ‘download’ button on each risk curve page currently available at https://vizhub.healthdata.org/burden-of-proof . Study characteristics and citations for all input data used in the analyses are also provided in Supplementary Table 3 , and Supplementary Table 2 provides a template of the data collection form.

Code availability

All code used for these analyses is publicly available online ( https://github.com/ihmeuw-msca/burden-of-proof ).

Doll, R. & Hill, A. B. Smoking and carcinoma of the lung. Br. Med. J. 2 , 739–748 (1950).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Di Cicco, M. E., Ragazzo, V. & Jacinto, T. Mortality in relation to smoking: the British Doctors Study. Breathe 12 , 275–276 (2016).

Article   PubMed   PubMed Central   Google Scholar  

World Health Organization. WHO Framework Convention on Tobacco Control 36 (WHO, 2003).

Dai, X., Gakidou, E. & Lopez, A. D. Evolution of the global smoking epidemic over the past half century: strengthening the evidence base for policy action. Tob. Control 31 , 129–137 (2022).

Article   PubMed   Google Scholar  

Dikshit, R. P. & Kanhere, S. Tobacco habits and risk of lung, oropharyngeal and oral cavity cancer: a population-based case-control study in Bhopal, India. Int. J. Epidemiol. 29 , 609–614 (2000).

Article   CAS   PubMed   Google Scholar  

Liaw, K. M. & Chen, C. J. Mortality attributable to cigarette smoking in Taiwan: a 12-year follow-up study. Tob. Control 7 , 141–148 (1998).

Gandini, S. et al. Tobacco smoking and cancer: a meta-analysis. Int. J. Cancer 122 , 155–164 (2008).

Deng, X., Yuan, C. & Chang, D. Interactions between single nucleotide polymorphism of SERPINA1 gene and smoking in association with COPD: a case–control study. Int. J. Chron. Obstruct. Pulmon. Dis. 12 , 259–265 (2017).

Leem, A. Y., Park, B., Kim, Y. S., Jung, J. Y. & Won, S. Incidence and risk of chronic obstructive pulmonary disease in a Korean community-based cohort. Int. J. Chron. Obstruct. Pulmon. Dis. 13 , 509–517 (2018).

Forey, B. A., Thornton, A. J. & Lee, P. N. Systematic review with meta-analysis of the epidemiological evidence relating smoking to COPD, chronic bronchitis and emphysema. BMC Pulmon. Med. 11 , 36 (2011).

Article   Google Scholar  

Tan, J. et al. Smoking, blood pressure, and cardiovascular disease mortality in a large cohort of Chinese men with 15 years follow-up. Int. J. Environ. Res. Public Health 15 , E1026 (2018).

Doll, R., Peto, R., Boreham, J. & Sutherland, I. Mortality in relation to smoking: 50 years’ observations on male British doctors. Br. Med. J. 328 , 1519 (2004).

Huxley, R. R. & Woodward, M. Cigarette smoking as a risk factor for coronary heart disease in women compared with men: a systematic review and meta-analysis of prospective cohort studies. Lancet 378 , 1297–1305 (2011).

Hbejan, K. Smoking effect on ischemic heart disease in young patients. Heart Views 12 , 1–6 (2011).

Chao, H. et al. A meta-analysis of active smoking and risk of meningioma. Tob. Induc. Dis. 19 , 34 (2021).

Shi, H., Shao, X. & Hong, Y. Association between cigarette smoking and the susceptibility of acute myeloid leukemia: a systematic review and meta-analysis. Eur. Rev. Med Pharm. Sci. 23 , 10049–10057 (2019).

CAS   Google Scholar  

Macacu, A., Autier, P., Boniol, M. & Boyle, P. Active and passive smoking and risk of breast cancer: a meta-analysis. Breast Cancer Res. Treat. 154 , 213–224 (2015).

Pujades-Rodriguez, M. et al. Heterogeneous associations between smoking and a wide range of initial presentations of cardiovascular disease in 1 937 360 people in England: lifetime risks and implications for risk prediction. Int. J. Epidemiol. 44 , 129–141 (2015).

Kanazir, M. et al. Risk factors for hepatocellular carcinoma: a case-control study in Belgrade (Serbia). Tumori 96 , 911–917 (2010).

Pytynia, K. B. et al. Matched-pair analysis of survival of never smokers and ever smokers with squamous cell carcinoma of the head and neck. J. Clin. Oncol. 22 , 3981–3988 (2004).

Barengo, N. C., Antikainen, R., Harald, K. & Jousilahti, P. Smoking and cancer, cardiovascular and total mortality among older adults: the Finrisk Study. Prev. Med. Rep. 14 , 100875 (2019).

Guo, Y. et al. Modifiable risk factors for cognitive impairment in Parkinson’s disease: a systematic review and meta-analysis of prospective cohort studies. Mov. Disord. 34 , 876–883 (2019).

Aune, D., Vatten, L. J. & Boffetta, P. Tobacco smoking and the risk of gallbladder disease. Eur. J. Epidemiol. 31 , 643–653 (2016).

Qin, L., Deng, H.-Y., Chen, S.-J. & Wei, W. Relationship between cigarette smoking and risk of chronic myeloid leukaemia: a meta-analysis of epidemiological studies. Hematology 22 , 193–200 (2017).

Petrick, J. L. et al. Tobacco, alcohol use and risk of hepatocellular carcinoma and intrahepatic cholangiocarcinoma: the Liver Cancer Pooling Project. Br. J. Cancer 118 , 1005–1012 (2018).

United States Department of Health, Education and Welfare. Smoking and Health. Report of the Advisory Committee on Smoking and Health to the Surgeon General of the United States Public Health Service https://www.cdc.gov/tobacco/data_statistics/sgr/index.htm (US DHEW, 1964).

United States Public Health Service Office of the Surgeon General & National Center for Chronic Disease Prevention and Health Promotion (US) Office on Smoking and Health. Smoking Cessation: A Report of the Surgeon General . (US Department of Health and Human Services, 2020).

Zheng, P., Barber, R., Sorensen, R. J. D., Murray, C. J. L. & Aravkin, A. Y. Trimmed constrained mixed effects models: formulations and algorithms. J. Comput. Graph Stat. 30 , 544–556 (2021).

Zheng, P. et al. The Burden of Proof studies: assessing the evidence of risk. Nat. Med. in press (2022).

Reitsma, M. B. et al. Spatial, temporal, and demographic patterns in prevalence of smoking tobacco use and attributable disease burden in 204 countries and territories, 1990–2019: a systematic analysis from the Global Burden of Disease Study 2019. Lancet 397 , 2337–2360 (2021).

Moher, D., Liberati, A., Tetzlaff, J. & Altman, D. G. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Br. Med. J. 339 , b2535 (2009).

Liu, Z. Y., He, X. Z. & Chapman, R. S. Smoking and other risk factors for lung cancer in Xuanwei, China. Int. J. Epidemiol. 20 , 26–31 (1991).

Brownson, R. C., Reif, J. S., Keefe, T. J., Ferguson, S. W. & Pritzl, J. A. Risk factors for adenocarcinoma of the lung. Am. J. Epidemiol. 125 , 25–34 (1987).

Marugame, T. et al. Lung cancer death rates by smoking status: comparison of the Three-Prefecture Cohort study in Japan to the Cancer Prevention Study II in the USA. Cancer Sci. 96 , 120–126 (2005).

Dosemeci, M., Gokmen, I., Unsal, M., Hayes, R. B. & Blair, A. Tobacco, alcohol use, and risks of laryngeal and lung cancer by subsite and histologic type in Turkey. Cancer Causes Control 8 , 729–737 (1997).

Freedman, N. D. et al. Impact of changing US cigarette smoking patterns on incident cancer: risks of 20 smoking-related cancers among the women and men of the NIH-AARP cohort. Int. J. Epidemiol. 45 , 846–856 (2016).

Bae, J.-M. et al. Lung cancer incidence by smoking status in Korean men: 16 years of observations in the Seoul Male Cancer Cohort study. J. Korean Med. Sci. 28 , 636–637 (2013).

Everatt, R., Kuzmickienė, I., Virvičiūtė, D. & Tamošiūnas, A. Cigarette smoking, educational level and total and site-specific cancer: a cohort study in men in Lithuania. Eur. J. Cancer Prev. 23 , 579–586 (2014).

Nordlund, L. A., Carstensen, J. M. & Pershagen, G. Are male and female smokers at equal risk of smoking-related cancer: evidence from a Swedish prospective study. Scand. J. Public Health 27 , 56–62 (1999).

Siemiatycki, J., Krewski, D., Franco, E. & Kaiserman, M. Associations between cigarette smoking and each of 21 types of cancer: a multi-site case–control study. Int. J. Epidemiol. 24 , 504–514 (1995).

Chyou, P. H., Nomura, A. M. & Stemmermann, G. N. A prospective study of the attributable risk of cancer due to cigarette smoking. Am. J. Public Health 82 , 37–40 (1992).

Potter, J. D., Sellers, T. A., Folsom, A. R. & McGovern, P. G. Alcohol, beer, and lung cancer in postmenopausal women. The Iowa Women’s Health Study. Ann. Epidemiol. 2 , 587–595 (1992).

Chyou, P. H., Nomura, A. M., Stemmermann, G. N. & Kato, I. Lung cancer: a prospective study of smoking, occupation, and nutrient intake. Arch. Environ. Health 48 , 69–72 (1993).

Pesch, B. et al. Cigarette smoking and lung cancer–relative risk estimates for the major histological types from a pooled analysis of case–control studies. Int. J. Cancer 131 , 1210–1219 (2012).

Jöckel, K. H. et al. Occupational and environmental hazards associated with lung cancer. Int. J. Epidemiol. 21 , 202–213 (1992).

Jöckel, K. H., Ahrens, W., Jahn, I., Pohlabeln, H. & Bolm-Audorff, U. Occupational risk factors for lung cancer: a case-control study in West Germany. Int. J. Epidemiol. 27 , 549–560 (1998).

Lei, Y. X., Cai, W. C., Chen, Y. Z. & Du, Y. X. Some lifestyle factors in human lung cancer: a case-control study of 792 lung cancer cases. Lung Cancer 14 , S121–S136 (1996).

Pawlega, J., Rachtan, J. & Dyba, T. Evaluation of certain risk factors for lung cancer in Cracow (Poland)—a case–control study. Acta Oncol. 36 , 471–476 (1997).

Mao, Y. et al. Socioeconomic status and lung cancer risk in Canada. Int. J. Epidemiol. 30 , 809–817 (2001).

Barbone, F., Bovenzi, M., Cavallieri, F. & Stanta, G. Cigarette smoking and histologic type of lung cancer in men. Chest 112 , 1474–1479 (1997).

Matos, E., Vilensky, M., Boffetta, P. & Kogevinas, M. Lung cancer and smoking: a case–control study in Buenos Aires, Argentina. Lung Cancer 21 , 155–163 (1998).

Simonato, L. et al. Lung cancer and cigarette smoking in Europe: an update of risk estimates and an assessment of inter-country heterogeneity. Int. J. Cancer 91 , 876–887 (2001).

Risch, H. A. et al. Are female smokers at higher risk for lung cancer than male smokers? A case–control analysis by histologic type. Am. J. Epidemiol. 138 , 281–293 (1993).

Sankaranarayanan, R. et al. A case–control study of diet and lung cancer in Kerala, south India. Int. J. Cancer 58 , 644–649 (1994).

Band, P. R. et al. Identification of occupational cancer risks in British Columbia. Part I: methodology, descriptive results, and analysis of cancer risks, by cigarette smoking categories of 15,463 incident cancer cases. J. Occup. Environ. Med. 41 , 224–232 (1999).

Becher, H., Jöckel, K. H., Timm, J., Wichmann, H. E. & Drescher, K. Smoking cessation and nonsmoking intervals: effect of different smoking patterns on lung cancer risk. Cancer Causes Control 2 , 381–387 (1991).

Brockmöller, J., Kerb, R., Drakoulis, N., Nitz, M. & Roots, I. Genotype and phenotype of glutathione S-transferase class mu isoenzymes mu and psi in lung cancer patients and controls. Cancer Res. 53 , 1004–1011 (1993).

PubMed   Google Scholar  

Vena, J. E., Byers, T. E., Cookfair, D. & Swanson, M. Occupation and lung cancer risk. An analysis by histologic subtypes. Cancer 56 , 910–917 (1985).

Cascorbi, I. et al. Homozygous rapid arylamine N -acetyltransferase (NAT2) genotype as a susceptibility factor for lung cancer. Cancer Res. 56 , 3961–3966 (1996).

CAS   PubMed   Google Scholar  

Chiazze, L., Watkins, D. K. & Fryar, C. A case–control study of malignant and non-malignant respiratory disease among employees of a fiberglass manufacturing facility. Br. J. Ind. Med 49 , 326–331 (1992).

CAS   PubMed   PubMed Central   Google Scholar  

Ando, M. et al. Attributable and absolute risk of lung cancer death by smoking status: findings from the Japan Collaborative Cohort Study. Int. J. Cancer 105 , 249–254 (2003).

De Matteis, S. et al. Are women who smoke at higher risk for lung cancer than men who smoke? Am. J. Epidemiol. 177 , 601–612 (2013).

He, Y. et al. Changes in smoking behavior and subsequent mortality risk during a 35-year follow-up of a cohort in Xi’an, China. Am. J. Epidemiol. 179 , 1060–1070 (2014).

Nishino, Y. et al. Cancer incidence profiles in the Miyagi Cohort Study. J. Epidemiol. 14 , S7–S11 (2004).

Papadopoulos, A. et al. Cigarette smoking and lung cancer in women: results of the French ICARE case–control study. Lung Cancer 74 , 369–377 (2011).

Shimazu, T. et al. Alcohol and risk of lung cancer among Japanese men: data from a large-scale population-based cohort study, the JPHC study. Cancer Causes Control 19 , 1095–1102 (2008).

Tindle, H. A. et al. Lifetime smoking history and risk of lung cancer: results from the Framingham Heart Study. J. Natl Cancer Inst. 110 , 1201–1207 (2018).

PubMed   PubMed Central   Google Scholar  

Yong, L. C. et al. Intake of vitamins E, C, and A and risk of lung cancer. The NHANES I epidemiologic followup study. First National Health and Nutrition Examination Survey. Am. J. Epidemiol. 146 , 231–243 (1997).

Hansen, M. S. et al. Sex differences in risk of smoking-associated lung cancer: results from a cohort of 600,000 Norwegians. Am. J. Epidemiol. 187 , 971–981 (2018).

Boffetta, P. et al. Tobacco smoking as a risk factor of bronchioloalveolar carcinoma of the lung: pooled analysis of seven case-control studies in the International Lung Cancer Consortium (ILCCO). Cancer Causes Control 22 , 73–79 (2011).

Yun, Y. D. et al. Hazard ratio of smoking on lung cancer in Korea according to histological type and gender. Lung 194 , 281–289 (2016).

Suzuki, I. et al. Risk factors for lung cancer in Rio de Janeiro, Brazil: a case–control study. Lung Cancer 11 , 179–190 (1994).

De Stefani, E., Deneo-Pellegrini, H., Carzoglio, J. C., Ronco, A. & Mendilaharsu, M. Dietary nitrosodimethylamine and the risk of lung cancer: a case–control study from Uruguay. Cancer Epidemiol. Biomark. Prev. 5 , 679–682 (1996).

Google Scholar  

Kreuzer, M. et al. Risk factors for lung cancer in young adults. Am. J. Epidemiol. 147 , 1028–1037 (1998).

Armadans-Gil, L., Vaqué-Rafart, J., Rosselló, J., Olona, M. & Alseda, M. Cigarette smoking and male lung cancer risk with special regard to type of tobacco. Int. J. Epidemiol. 28 , 614–619 (1999).

Kubík, A. K., Zatloukal, P., Tomásek, L. & Petruzelka, L. Lung cancer risk among Czech women: a case–control study. Prev. Med. 34 , 436–444 (2002).

Rachtan, J. Smoking, passive smoking and lung cancer cell types among women in Poland. Lung Cancer 35 , 129–136 (2002).

Thun, M. J. et al. 50-year trends in smoking-related mortality in the United States. N. Engl. J. Med. 368 , 351–364 (2013).

Zatloukal, P., Kubík, A., Pauk, N., Tomásek, L. & Petruzelka, L. Adenocarcinoma of the lung among women: risk associated with smoking, prior lung disease, diet and menstrual and pregnancy history. Lung Cancer 41 , 283–293 (2003).

Hansen, M. S., Licaj, I., Braaten, T., Lund, E. & Gram, I. T. The fraction of lung cancer attributable to smoking in the Norwegian Women and Cancer (NOWAC) Study. Br. J. Cancer 124 , 658–662 (2021).

Zhang, P. et al. Association of smoking and polygenic risk with the incidence of lung cancer: a prospective cohort study. Br. J. Cancer 126 , 1637–1646 (2022).

Weber, M. F. et al. Cancer incidence and cancer death in relation to tobacco smoking in a population-based Australian cohort study. Int. J. Cancer 149 , 1076–1088 (2021).

Guo, L.-W. et al. A risk prediction model for selecting high-risk population for computed tomography lung cancer screening in China. Lung Cancer 163 , 27–34 (2022).

Mezzoiuso, A. G., Odone, A., Signorelli, C. & Russo, A. G. Association between smoking and cancers among women: results from the FRiCaM multisite cohort study. J. Cancer 12 , 3136–3144 (2021).

Hawrysz, I., Wadolowska, L., Slowinska, M. A., Czerwinska, A. & Golota, J. J. Adherence to prudent and mediterranean dietary patterns is inversely associated with lung cancer in moderate but not heavy male Polish smokers: a case–control study. Nutrients 12 , E3788 (2020).

Huang, C.-C., Lai, C.-Y., Tsai, C.-H., Wang, J.-Y. & Wong, R.-H. Combined effects of cigarette smoking, DNA methyltransferase 3B genetic polymorphism, and DNA damage on lung cancer. BMC Cancer 21 , 1066 (2021).

Viner, B., Barberio, A. M., Haig, T. R., Friedenreich, C. M. & Brenner, D. R. The individual and combined effects of alcohol consumption and cigarette smoking on site-specific cancer risk in a prospective cohort of 26,607 adults: results from Alberta’s Tomorrow Project. Cancer Causes Control 30 , 1313–1326 (2019).

Park, E. Y., Lim, M. K., Park, E., Oh, J.-K. & Lee, D.-H. Relationship between urinary 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol and lung cancer risk in the general population: a community-based prospective cohort study. Front. Oncol. 11 , 611674 (2021).

De Stefani, E., Deneo-Pellegrini, H., Mendilaharsu, M., Carzoglio, J. C. & Ronco, A. Dietary fat and lung cancer: a case–control study in Uruguay. Cancer Causes Control 8 , 913–921 (1997).

Wünsch-Filho, V., Moncau, J. E., Mirabelli, D. & Boffetta, P. Occupational risk factors of lung cancer in São Paulo, Brazil. Scand. J. Work Environ. Health 24 , 118–124 (1998).

Hu, J. et al. A case-control study of diet and lung cancer in northeast China. Int. J. Cancer 71 , 924–931 (1997).

Jia, G., Wen, W., Massion, P. P., Shu, X.-O. & Zheng, W. Incorporating both genetic and tobacco smoking data to identify high-risk smokers for lung cancer screening. Carcinogenesis 42 , 874–879 (2021).

Rusmaully, J. et al. Risk of lung cancer among women in relation to lifetime history of tobacco smoking: a population-based case–control study in France (the WELCA study). BMC Cancer 21 , 711 (2021).

Jin, K. et al. Tobacco smoking modifies the association between hormonal factors and lung cancer occurrence among post-menopausal Chinese women. Transl. Oncol. 12 , 819–827 (2019).

Tse, L. A., Wang, F., Wong, M. C.-S., Au, J. S.-K. & Yu, I. T.-S. Risk assessment and prediction for lung cancer among Hong Kong Chinese men. BMC Cancer 22 , 585 (2022).

Huang, C.-C. et al. Joint effects of cigarette smoking and green tea consumption with miR-29b and DNMT3b mRNA expression in the development of lung cancer. Genes 13 , 836 (2022).

Hosseini, M. et al. Environmental risk factors for lung cancer in Iran: a case–control study. Int. J. Epidemiol. 38 , 989–996 (2009).

Naghibzadeh-Tahami, A. et al. Is opium use associated with an increased risk of lung cancer? A case–control study. BMC Cancer 20 , 807 (2020).

Shimatani, K., Ito, H., Matsuo, K., Tajima, K. & Takezaki, T. Cumulative cigarette tar exposure and lung cancer risk among Japanese smokers. Jpn J. Clin. Oncol. 50 , 1009–1017 (2020).

Lai, C.-Y. et al. Genetic polymorphism of catechol- O -methyltransferase modulates the association of green tea consumption and lung cancer. Eur. J. Cancer Prev. 28 , 316–322 (2019).

Schwartz, A. G. et al. Hormone use, reproductive history, and risk of lung cancer: the Women’s Health Initiative studies. J. Thorac. Oncol. 10 , 1004–1013 (2015).

Kreuzer, M., Gerken, M., Heinrich, J., Kreienbrock, L. & Wichmann, H.-E. Hormonal factors and risk of lung cancer among women? Int. J. Epidemiol. 32 , 263–271 (2003).

Sreeja, L. et al. Possible risk modification by CYP1A1, GSTM1 and GSTT1 gene polymorphisms in lung cancer susceptibility in a South Indian population. J. Hum. Genet. 50 , 618–627 (2005).

Siemiatycki, J. et al. Are the apparent effects of cigarette smoking on lung and bladder cancers due to uncontrolled confounding by occupational exposures? Epidemiology 5 , 57–65 (1994).

Chan-Yeung, M. et al. Risk factors associated with lung cancer in Hong Kong. Lung Cancer 40 , 131–140 (2003).

Lawania, S., Singh, N., Behera, D. & Sharma, S. Xeroderma pigmentosum complementation group D polymorphism toward lung cancer susceptibility survival and response in patients treated with platinum chemotherapy. Future Oncol. 13 , 2645–2665 (2017).

De Stefani, E. et al. Mate drinking and risk of lung cancer in males: a case-control study from Uruguay. Cancer Epidemiol. Biomark. Prev. 5 , 515–519 (1996).

Pérez-Padilla, R. et al. Exposure to biomass smoke and chronic airway disease in Mexican women. A case-control study. Am. J. Respir. Crit. Care Med. 154 , 701–706 (1996).

Zhang, X.-R. et al. Glucosamine use, smoking and risk of incident chronic obstructive pulmonary disease: a large prospective cohort study. Br. J. Nutr . https://doi.org/10.1017/S000711452100372X (2021).

Johannessen, A., Omenaas, E., Bakke, P. & Gulsvik, A. Incidence of GOLD-defined chronic obstructive pulmonary disease in a general adult population. Int. J. Tuberc. Lung Dis. 9 , 926–932 (2005).

Fox, J. Life-style and mortality: a large-scale census-based cohort study in Japan. J. Epidemiol. Community Health 45 , 173 (1991).

Article   PubMed Central   Google Scholar  

Thomson, B. et al. Low-intensity daily smoking and cause-specific mortality in Mexico: prospective study of 150 000 adults. Int. J. Epidemiol. 50 , 955–964 (2021).

van Durme, Y. M. T. A. et al. Prevalence, incidence, and lifetime risk for the development of COPD in the elderly: the Rotterdam study. Chest 135 , 368–377 (2009).

Li, L. et al. SERPINE2 rs16865421 polymorphism is associated with a lower risk of chronic obstructive pulmonary disease in the Uygur population: a case–control study. J. Gene Med. 21 , e3106 (2019).

Ganbold, C. et al. The cumulative effect of gene-gene interactions between GSTM1 , CHRNA3 , CHRNA5 and SOD3 gene polymorphisms combined with smoking on COPD risk. Int. J. Chron. Obstruct. Pulmon. Dis. 16 , 2857–2868 (2021).

Omori, H. et al. Twelve-year cumulative incidence of airflow obstruction among Japanese males. Intern. Med. 50 , 1537–1544 (2011).

Manson, J. E., Ajani, U. A., Liu, S., Nathan, D. M. & Hennekens, C. H. A prospective study of cigarette smoking and the incidence of diabetes mellitus among US male physicians. Am. J. Med. 109 , 538–542 (2000).

Lv, J. et al. Adherence to a healthy lifestyle and the risk of type 2 diabetes in Chinese adults. Int. J. Epidemiol. 46 , 1410–1420 (2017).

Waki, K. et al. Alcohol consumption and other risk factors for self-reported diabetes among middle-aged Japanese: a population-based prospective study in the JPHC study cohort I. Diabet. Med. 22 , 323–331 (2005).

Meisinger, C., Döring, A., Thorand, B. & Löwel, H. Association of cigarette smoking and tar and nicotine intake with development of type 2 diabetes mellitus in men and women from the general population: the MONICA/KORA Augsburg Cohort Study. Diabetologia 49 , 1770–1776 (2006).

Huh, Y. et al. Association of smoking status with the risk of type 2 diabetes among young adults: a nationwide cohort study in South Korea. Nicotine Tob. Res. 24 , 1234–1240 (2022).

Sawada, S. S., Lee, I.-M., Muto, T., Matuszaki, K. & Blair, S. N. Cardiorespiratory fitness and the incidence of type 2 diabetes: prospective study of Japanese men. Diabetes Care 26 , 2918–2922 (2003).

Will, J. C., Galuska, D. A., Ford, E. S., Mokdad, A. & Calle, E. E. Cigarette smoking and diabetes mellitus: evidence of a positive association from a large prospective cohort study. Int. J. Epidemiol. 30 , 540–546 (2001).

Nakanishi, N., Nakamura, K., Matsuo, Y., Suzuki, K. & Tatara, K. Cigarette smoking and risk for impaired fasting glucose and type 2 diabetes in middle-aged Japanese men. Ann. Intern. Med. 133 , 183–191 (2000).

Sairenchi, T. et al. Cigarette smoking and risk of type 2 diabetes mellitus among middle-aged and elderly Japanese men and women. Am. J. Epidemiol. 160 , 158–162 (2004).

Hou, X. et al. Cigarette smoking is associated with a lower prevalence of newly diagnosed diabetes screened by OGTT than non-smoking in Chinese men with normal weight. PLoS ONE 11 , e0149234 (2016).

Hu, F. B. et al. Diet, lifestyle, and the risk of type 2 diabetes mellitus in women. N. Engl. J. Med. 345 , 790–797 (2001).

Teratani, T. et al. Dose-response relationship between tobacco or alcohol consumption and the development of diabetes mellitus in Japanese male workers. Drug Alcohol Depend. 125 , 276–282 (2012).

Kawakami, N., Takatsuka, N., Shimizu, H. & Ishibashi, H. Effects of smoking on the incidence of non-insulin-dependent diabetes mellitus. Replication and extension in a Japanese cohort of male employees. Am. J. Epidemiol. 145 , 103–109 (1997).

Patja, K. et al. Effects of smoking, obesity and physical activity on the risk of type 2 diabetes in middle-aged Finnish men and women. J. Intern. Med. 258 , 356–362 (2005).

White, W. B. et al. High-intensity cigarette smoking is associated with incident diabetes mellitus in Black adults: the Jackson Heart Study. J. Am. Heart Assoc. 7 , e007413 (2018).

Uchimoto, S. et al. Impact of cigarette smoking on the incidence of Type 2 diabetes mellitus in middle-aged Japanese men: the Osaka Health Survey. Diabet. Med . 16 , 951–955 (1999).

Rimm, E. B., Chan, J., Stampfer, M. J., Colditz, G. A. & Willett, W. C. Prospective study of cigarette smoking, alcohol use, and the risk of diabetes in men. Br. Med. J. 310 , 555–559 (1995).

Article   CAS   Google Scholar  

Hilawe, E. H. et al. Smoking and diabetes: is the association mediated by adiponectin, leptin, or C-reactive protein? J. Epidemiol. 25 , 99–109 (2015).

InterAct, Consortium et al. Smoking and long-term risk of type 2 diabetes: the EPIC-InterAct study in European populations. Diabetes Care 37 , 3164–3171 (2014).

Jee, S. H., Foong, A. W., Hur, N. W. & Samet, J. M. Smoking and risk for diabetes incidence and mortality in Korean men and women. Diabetes Care 33 , 2567–2572 (2010).

Rasouli, B. et al. Smoking and the risk of LADA: results from a Swedish population-based case-control study. Diabetes Care 39 , 794–800 (2016).

Wannamethee, S. G., Shaper, A. G. & Perry, I. J., British Regional Heart Study. Smoking as a modifiable risk factor for type 2 diabetes in middle-aged men. Diabetes Care 24 , 1590–1595 (2001).

Radzeviciene, L. & Ostrauskas, R. Smoking habits and type 2 diabetes mellitus in women. Women Health 58 , 884–897 (2018).

Carlsson, S., Midthjell, K. & Grill, V., Nord-Trøndelag Study. Smoking is associated with an increased risk of type 2 diabetes but a decreased risk of autoimmune diabetes in adults: an 11-year follow-up of incidence of diabetes in the Nord-Trøndelag study. Diabetologia 47 , 1953–1956 (2004).

Akter, S. et al. Smoking, smoking cessation, and the risk of type 2 diabetes among Japanese adults: Japan Epidemiology Collaboration on Occupational Health Study. PLoS ONE 10 , e0132166 (2015).

Pirie, K. et al. The 21st century hazards of smoking and benefits of stopping: a prospective study of one million women in the UK. Lancet 381 , 133–141 (2013).

Park, C.-H. et al. [The effect of smoking status upon occurrence of impaired fasting glucose or type 2 diabetes in Korean men]. J. Prev. Med. Public Health 41 , 249–254 (2008).

Doi, Y. et al. Two risk score models for predicting incident Type 2 diabetes in Japan. Diabet. Med. 29 , 107–114 (2012).

van den Brandt, P. A. A possible dual effect of cigarette smoking on the risk of postmenopausal breast cancer. Eur. J. Epidemiol. 32 , 683–690 (2017).

Dossus, L. et al. Active and passive cigarette smoking and breast cancer risk: results from the EPIC cohort. Int. J. Cancer 134 , 1871–1888 (2014).

Kawai, M., Malone, K. E., Tang, M.-T. C. & Li, C. I. Active smoking and the risk of estrogen receptor-positive and triple-negative breast cancer among women ages 20 to 44 years. Cancer 120 , 1026–1034 (2014).

Reynolds, P. et al. Active smoking, household passive smoking, and breast cancer: evidence from the California Teachers Study. J. Natl Cancer Inst. 96 , 29–37 (2004).

Ellingjord-Dale, M. et al. Alcohol, physical activity, smoking, and breast cancer subtypes in a large, nested case-control study from the Norwegian Breast Cancer Screening Program. Cancer Epidemiol. Biomark. Prev. 26 , 1736–1744 (2017).

Arthur, R. et al. Association between lifestyle, menstrual/reproductive history, and histological factors and risk of breast cancer in women biopsied for benign breast disease. Breast Cancer Res. Treat. 165 , 623–631 (2017).

Luo, J. et al. Association of active and passive smoking with risk of breast cancer among postmenopausal women: a prospective cohort study. Br. Med. J. 342 , d1016 (2011).

White, A. J., D’Aloisio, A. A., Nichols, H. B., DeRoo, L. A. & Sandler, D. P. Breast cancer and exposure to tobacco smoke during potential windows of susceptibility. Cancer Causes Control 28 , 667–675 (2017).

Gram, I. T. et al. Breast cancer risk among women who start smoking as teenagers. Cancer Epidemiol. Biomark. Prev. 14 , 61–66 (2005).

Gammon, M. D. et al. Cigarette smoking and breast cancer risk among young women (United States). Cancer Causes Control 9 , 583–590 (1998).

Magnusson, C., Wedrén, S. & Rosenberg, L. U. Cigarette smoking and breast cancer risk: a population-based study in Sweden. Br. J. Cancer 97 , 1287–1290 (2007).

Chu, S. Y. et al. Cigarette smoking and the risk of breast cancer. Am. J. Epidemiol. 131 , 244–253 (1990).

Lemogne, C. et al. Depression and the risk of cancer: a 15-year follow-up study of the GAZEL cohort. Am. J. Epidemiol. 178 , 1712–1720 (2013).

Morabia, A., Bernstein, M., Héritier, S. & Khatchatrian, N. Relation of breast cancer with passive and active exposure to tobacco smoke. Am. J. Epidemiol. 143 , 918–928 (1996).

Conlon, M. S. C., Johnson, K. C., Bewick, M. A., Lafrenie, R. M. & Donner, A. Smoking (active and passive), N -acetyltransferase 2, and risk of breast cancer. Cancer Epidemiol. 34 , 142–149 (2010).

Ozasa, K., Japan Collaborative Cohort Study for Evaluation of Cancer. Smoking and mortality in the Japan Collaborative Cohort Study for Evaluation of Cancer (JACC). Asian Pac. J. Cancer Prev. 8 , 89–96 (2007).

Jones, M. E., Schoemaker, M. J., Wright, L. B., Ashworth, A. & Swerdlow, A. J. Smoking and risk of breast cancer in the Generations Study cohort. Breast Cancer Res. 19 , 118 (2017).

Bjerkaas, E. et al. Smoking duration before first childbirth: an emerging risk factor for breast cancer? Results from 302,865 Norwegian women. Cancer Causes Control 24 , 1347–1356 (2013).

Gram, I. T., Little, M. A., Lund, E. & Braaten, T. The fraction of breast cancer attributable to smoking: the Norwegian women and cancer study 1991–2012. Br. J. Cancer 115 , 616–623 (2016).

Li, C. I., Malone, K. E. & Daling, J. R. The relationship between various measures of cigarette smoking and risk of breast cancer among older women 65–79 years of age (United States). Cancer Causes Control 16 , 975–985 (2005).

Xue, F., Willett, W. C., Rosner, B. A., Hankinson, S. E. & Michels, K. B. Cigarette smoking and the incidence of breast cancer. Arch. Intern. Med. 171 , 125–133 (2011).

Parker, A. S., Cerhan, J. R., Putnam, S. D., Cantor, K. P. & Lynch, C. F. A cohort study of farming and risk of prostate cancer in Iowa. Epidemiology 10 , 452–455 (1999).

Sawada, N. et al. Alcohol and smoking and subsequent risk of prostate cancer in Japanese men: the Japan Public Health Center-based prospective study. Int. J. Cancer 134 , 971–978 (2014).

Hiatt, R. A., Armstrong, M. A., Klatsky, A. L. & Sidney, S. Alcohol consumption, smoking, and other risk factors and prostate cancer in a large health plan cohort in California (United States). Cancer Causes Control 5 , 66–72 (1994).

Cerhan, J. R. et al. Association of smoking, body mass, and physical activity with risk of prostate cancer in the Iowa 65+ Rural Health Study (United States). Cancer Causes Control 8 , 229–238 (1997).

Watters, J. L., Park, Y., Hollenbeck, A., Schatzkin, A. & Albanes, D. Cigarette smoking and prostate cancer in a prospective US cohort study. Cancer Epidemiol. Biomark. Prev. 18 , 2427–2435 (2009).

Butler, L. M., Wang, R., Wong, A. S., Koh, W.-P. & Yu, M. C. Cigarette smoking and risk of prostate cancer among Singapore Chinese. Cancer Causes Control 20 , 1967–1974 (2009).

Lotufo, P. A., Lee, I. M., Ajani, U. A., Hennekens, C. H. & Manson, J. E. Cigarette smoking and risk of prostate cancer in the physicians’ health study (United States). Int. J. Cancer 87 , 141–144 (2000).

Hsing, A. W. et al. Diet, tobacco use, and fatal prostate cancer: results from the Lutheran Brotherhood Cohort Study. Cancer Res. 50 , 6836–6840 (1990).

Veierød, M. B., Laake, P. & Thelle, D. S. Dietary fat intake and risk of prostate cancer: a prospective study of 25,708 Norwegian men. Int. J. Cancer 73 , 634–638 (1997).

Meyer, J., Rohrmann, S., Bopp, M. & Faeh, D. & Swiss National Cohort Study Group. Impact of smoking and excess body weight on overall and site-specific cancer mortality risk. Cancer Epidemiol. Biomark. Prev . 24 , 1516–1522 (2015).

Putnam, S. D. et al. Lifestyle and anthropometric risk factors for prostate cancer in a cohort of Iowa men. Ann. Epidemiol. 10 , 361–369 (2000).

Taghizadeh, N., Vonk, J. M. & Boezen, H. M. Lifetime smoking history and cause-specific mortality in a cohort study with 43 years of follow-up. PLoS ONE 11 , e0153310 (2016).

Park, S.-Y. et al. Racial/ethnic differences in lifestyle-related factors and prostate cancer risk: the Multiethnic Cohort Study. Cancer Causes Control 26 , 1507–1515 (2015).

Nomura, A. M., Lee, J., Stemmermann, G. N. & Combs, G. F. Serum selenium and subsequent risk of prostate cancer. Cancer Epidemiol. Biomark. Prev. 9 , 883–887 (2000).

Rodriguez, C., Tatham, L. M., Thun, M. J., Calle, E. E. & Heath, C. W. Smoking and fatal prostate cancer in a large cohort of adult men. Am. J. Epidemiol. 145 , 466–475 (1997).

Rohrmann, S. et al. Smoking and risk of fatal prostate cancer in a prospective U.S. study. Urology 69 , 721–725 (2007).

Giovannucci, E. et al. Smoking and risk of total and fatal prostate cancer in United States health professionals. Cancer Epidemiol. Biomark. Prev. 8 , 277–282 (1999).

Rohrmann, S. et al. Smoking and the risk of prostate cancer in the European Prospective Investigation into Cancer and Nutrition. Br. J. Cancer 108 , 708–714 (2013).

Lund Nilsen, T. I., Johnsen, R. & Vatten, L. J. Socio-economic and lifestyle factors associated with the risk of prostate cancer. Br. J. Cancer 82 , 1358–1363 (2000).

Hsing, A. W., McLaughlin, J. K., Hrubec, Z., Blot, W. J. & Fraumeni, J. F. Tobacco use and prostate cancer: 26-year follow-up of US veterans. Am. J. Epidemiol. 133 , 437–441 (1991).

Murray, C. J. L. et al. Global burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 396 , 1223–1249 (2020).

Bero, L. A. & Jadad, A. R. How consumers and policymakers can use systematic reviews for decision making. Ann. Intern. Med. 127 , 37–42 (1997).

Centers for Disease Control and Prevention (CDC). Cigarette smoking among adults and trends in smoking cessation—United States, 2008. MMWR Morb. Mortal. Wkly Rep. 58 , 1227–1232 (2009).

Prochaska, J. O. & Goldstein, M. G. Process of smoking cessation: implications for clinicians. Clin. Chest Med. 12 , 727–735 (1991).

Page, M. J. et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Br. Med. J. 372 , n71 (2021).

Stevens, G. A. et al. Guidelines for Accurate and Transparent Health Estimates Reporting: the GATHER statement. Lancet 388 , e19–e23 (2016).

BMJ Best Practice. What is GRADE? https://bestpractice.bmj.com/info/us/toolkit/learn-ebm/what-is-grade (BMJ, 2021).

The GRADE Working Group. GRADE handbook . https://gdt.gradepro.org/app/handbook/handbook.html (The GRADE Working Group, 2013).

Efron, B., Hastie, T., Johnstone, I. & Tibshirani, R. Least angle regression. Ann. Stat. 32 , 407–499 (2004).

Tibshirani, R. Regression shrinkage and selection via the lasso. J. R. Stat. Soc. Ser. B Stat. Methodol. 58 , 267–288 (1996).

von Hippel, P. T. The heterogeneity statistic I2 can be biased in small meta-analyses. BMC Med. Res. Methodol. 15 , 35 (2015).

Kontopantelis, E., Springate, D. A. & Reeves, D. A re-analysis of the Cochrane Library data: the dangers of unobserved heterogeneity in meta-analyses. PLoS ONE 8 , e69930 (2013).

Biggerstaff, B. J. & Tweedie, R. L. Incorporating variability in estimates of heterogeneity in the random effects model in meta-analysis. Stat. Med. 16 , 753–768 (1997).

Egger, M., Smith, G. D., Schneider, M. & Minder, C. Bias in meta-analysis detected by a simple, graphical test. Br. Med. J. 315 , 629–634 (1997).

Lee, P. N., Forey, B. A. & Coombs, K. J. Systematic review with meta-analysis of the epidemiological evidence in the 1900s relating smoking to lung cancer. BMC Cancer 12 , 385 (2012).

Rücker, G., Carpenter, J. R. & Schwarzer, G. Detecting and adjusting for small-study effects in meta-analysis. Biometr. J. 53 , 351–368 (2011).

Wu, Z.-J., Zhao, P., Liu, B. & Yuan, Z.-C. Effect of cigarette smoking on risk of hip fracture in men: a meta-analysis of 14 prospective cohort studies. PLoS ONE 11 , e0168990 (2016).

Thun, M. J. et al. in Cigarette Smoking Behaviour in the United States: changes in cigarette-related disease risks and their implication for prevention and control (eds Burns, D.M. et al.) Tobacco Control Monograph No. 8 Ch. 4 (National Cancer Institute, 1997).

Tolstrup, J. S. et al. Smoking and risk of coronary heart disease in younger, middle-aged, and older adults. Am. J. Public Health 104 , 96–102 (2014).

Jonas, M. A., Oates, J. A., Ockene, J. K. & Hennekens, C. H. Statement on smoking and cardiovascular disease for health care professionals. American Heart Association. Circulation 86 , 1664–1669 (1992).

Khan, S. S. et al. Cigarette smoking and competing risks for fatal and nonfatal cardiovascular disease subtypes across the life course. J. Am. Heart Assoc. 10 , e021751 (2021).

Download references

Acknowledgements

Research reported in this publication was supported by the Bill & Melinda Gates Foundation and Bloomberg Philanthropies. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funders. The study funders had no role in study design, data collection, data analysis, data interpretation, writing of the final report or the decision to publish.

We thank the Tobacco Metrics Team Advisory Group for their valuable input and review of the work. The members of the Advisory Group are: P. Allebeck, R. Chandora, J. Drope, M. Eriksen, E. Fernández, H. Gouda, R. Kennedy, D. McGoldrick, L. Pan, K. Schotte, E. Sebrie, J. Soriano, M. Tynan and K. Welding.

Author information

Authors and affiliations.

Institute for Health Metrics and Evaluation, University of Washington, Seattle, WA, USA

Xiaochen Dai, Gabriela F. Gil, Marissa B. Reitsma, Noah S. Ahmad, Jason A. Anderson, Catherine Bisignano, Sinclair Carr, Rachel Feldman, Simon I. Hay, Jiawei He, Vincent Iannucci, Hilary R. Lawlor, Matthew J. Malloy, Laurie B. Marczak, Susan A. McLaughlin, Larissa Morikawa, Erin C. Mullany, Sneha I. Nicholson, Erin M. O’Connell, Chukwuma Okereke, Reed J. D. Sorensen, Joanna Whisnant, Aleksandr Y. Aravkin, Peng Zheng, Christopher J. L. Murray & Emmanuela Gakidou

Department of Health Metrics Sciences, School of Medicine, University of Washington, Seattle, WA, USA

Xiaochen Dai, Simon I. Hay, Jiawei He, Peng Zheng, Christopher J. L. Murray & Emmanuela Gakidou

Department of Applied Mathematics, University of Washington, Seattle, WA, USA

Aleksandr Y. Aravkin

You can also search for this author in PubMed   Google Scholar

Contributions

X.D., S.I.H., S.A.M., E.C.M., E.M.O., C.J.L.M. and E.G. managed the estimation or publications process. X.D. and G.F.G. wrote the first draft of the manuscript. X.D. and P.Z. had primary responsibility for applying analytical methods to produce estimates. X.D., G.F.G., N.S.A., J.A.A., S.C., R.F., V.I., M.J.M., L.M., S.I.N., C.O., M.B.R. and J.W. had primary responsibility for seeking, cataloguing, extracting or cleaning data, and for designing or coding figures and tables. X.D., G.F.G., M.B.R., N.S.A., H.R.L., C.O. and J.W. provided data or critical feedback on data sources. X.D., J.H., R.J.D.S., A.Y.A., P.Z., C.J.L.M. and E.G. developed methods or computational machinery. X.D., G.F.G., M.B.R., S.I.H., J.H., R.J.D.S., A.Y.A., P.Z., C.J.L.M. and E.G. provided critical feedback on methods or results. X.D., G.F.G., M.B.R., C.B., S.I.H., L.B.M., S.A.M., A.Y.A. and E.G. drafted the work or revised it critically for important intellectual content. X.D., S.I.H., L.B.M., E.C.M., E.M.O. and E.G. managed the overall research enterprise.

Corresponding author

Correspondence to Xiaochen Dai .

Ethics declarations

Competing interests.

The authors declare no competing interests.

Peer review

Peer review information.

Nature Medicine thanks Frederic Sitas and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Jennifer Sargent and Ming Yang, in collaboration with the Nature Medicine team.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended data fig. 1 prisma 2020 flow diagram for an updated systematic review of the smoking and tracheal, bronchus, and lung cancer risk-outcome pair..

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and lung cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Extended Data Fig. 2 PRISMA 2020 flow diagram for an updated systematic review of the Smoking and Chronic obstructive pulmonary disease risk-outcome pair.

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and chronic obstructive pulmonary disease conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Extended Data Fig. 3 PRISMA 2020 flow diagram for an updated systematic review of the Smoking and Diabetes mellitus type 2 risk- outcome pair.

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and type 2 diabetes conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Extended Data Fig. 4 PRISMA 2020 flow diagram for an updated systematic review of the Smoking and Breast cancer risk-outcome pair.

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and breast cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Extended Data Fig. 5 PRISMA 2020 flow diagram for an updated systematic review of the Smoking and Prostate cancer risk-outcome pair.

The PRISMA flow diagram of an updated systematic review on the relationship between smoking and prostate cancer conducted on PubMed to update historical review from previous cycles of the Global Burden of Disease Study. Template is from: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/ .

Extended Data Fig. 6 Smoking and Breast Cancer.

a , log-relative risk function. b , relative risk function. c , A modified funnel plot showing the residuals (relative to 0) on the x-axis and the estimated standard deviation (SD) that includes reported SD and between-study heterogeneity on the y-axis.

Supplementary information

Supplementary information.

Supplementary Information 1: Data source identification and assessment. Supplementary Information 2: Data inputs. Supplementary Information 3: Study quality and bias assessment. Supplementary Information 4: The dose–response RR curves and their 95% UIs for all smoking–outcome pairs. Supplementary Information 5: Supplementary methods. Supplementary Information 6: Sensitivity analysis. Supplementary Information 7: Binary smoking–outcome pair. Supplementary Information 8: Risk curve details. Supplementary Information 9: GATHER and PRISMA checklists.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Dai, X., Gil, G.F., Reitsma, M.B. et al. Health effects associated with smoking: a Burden of Proof study. Nat Med 28 , 2045–2055 (2022). https://doi.org/10.1038/s41591-022-01978-x

Download citation

Received : 11 April 2022

Accepted : 28 July 2022

Published : 10 October 2022

Issue Date : October 2022

DOI : https://doi.org/10.1038/s41591-022-01978-x

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

This article is cited by

Estimating effects of whole grain consumption on type 2 diabetes, colorectal cancer and cardiovascular disease: a burden of proof study.

Nutrition Journal (2024)

Effects of passive smoking on cortical spreading depolarization in male and female mice

• Tsubasa Takizawa
• Keiko Ihara
• Jin Nakahara

The Journal of Headache and Pain (2024)

Association of cigarette smoking, smoking cessation with the risk of cardiometabolic multimorbidity in the UK Biobank

BMC Public Health (2024)

Health effects associated with chewing tobacco: a Burden of Proof study

• Gabriela F. Gil
• Jason A. Anderson
• Emmanuela Gakidou

Nature Communications (2024)

Western diets and chronic diseases

• Timon E. Adolph
• Herbert Tilg

Nature Medicine (2024)

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: Cancer newsletter — what matters in cancer research, free to your inbox weekly.