eMethods. Calculation of smoking-attributable morbidity and variance estimates.
Rostron BL, Chang CM, Pechacek TF. Estimation of Cigarette Smoking–Attributable Morbidity in the United States. JAMA Intern Med. 2014;174(12):1922-1928. doi:10.1001/jamainternmed.2014.5219
Copyright 2014 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Cigarette smoking has been found to harm nearly every bodily organ and is a leading cause of preventable disease, but current estimates of smoking-attributable morbidity by condition for the United States are generally unavailable.
To estimate the burden of major medical conditions attributable to cigarette smoking in the United States.
Design, Setting, and Participants
The disease burden of smoking was estimated using population-attributable risk calculations, taking into account the uncertainty of estimates. Population estimates came from 2009 US Census Bureau data and smoking prevalence, disease prevalence, and disease relative risk estimates came from National Health Interview Survey data for surveyed adults from 2006 through 2012. National Health and Nutrition Examination Survey spirometry data obtained from medical examination of surveyed adults from 2007 through 2010 was used to adjust for underreporting of chronic obstructive pulmonary disease.
Smoking status was assessed from self-reported National Health Interview Survey data.
Main Outcomes and Measures
The number of adults 35 years and older who had had a major smoking-attributable disease by sex and condition and the total number of these conditions were estimated for the United States in 2009.
Using National Health Interview Survey data, we estimated that 6.9 million (95% CI, 6.5-7.4 million) US adults had had a combined 10.9 million (95% CI, 10.3-11.5 million) self-reported smoking-attributable medical conditions. Using chronic obstructive pulmonary disease prevalence estimates obtained from National Health and Nutrition Examination Survey self-reported and spirometry data, we estimated that US adults had had a combined 14.0 million (95% CI, 12.9-15.1 million) smoking-attributable conditions in 2009.
Conclusions and Relevance
We estimate that US adults have had approximately 14 million major medical conditions that were attributable to smoking. This figure is generally conservative owing to the existence of other diseases and medical events that were not included in these estimates. Cigarette smoking remains a leading cause of preventable disease in the United States, underscoring the need for continuing and vigorous smoking-prevention efforts.
Cigarette smoking has been found to harm nearly every organ and organ system in the body1 and is the leading cause of preventable death in the United States2 and of disease burden worldwide.3 The Centers for Disease Control and Prevention (CDC) periodically publishes estimates of smoking-attributable mortality and its economic costs,4,5 but the population disease burden of smoking has been much less studied, even though smoking is known to be a leading cause of many serious medical conditions.6 The CDC previously published estimates of smoking-attributable morbidity for the United States in 2000,7 finding that 8.6 million individuals had 12.7 million smoking-attributable conditions. Most of these conditions were chronic bronchitis and emphysema, often classified as chronic obstructive pulmonary disease (COPD), but these estimates and methods, to our knowledge, have not been subsequently updated or refined.
The nature and magnitude of smoking-attributable morbidity has changed in the intervening years, and additional medical conditions have been linked to smoking. The recent 50th-anniversary Report of the Surgeon General8 on the health effects of smoking concluded that previous estimates of the disease burden of smoking could be substantial underestimates given the absence of several major medical conditions caused by smoking. The report also noted that the burden of COPD due to smoking could be particularly underestimated. To address these issues, we present estimates of smoking-attributable morbidity for the United States in 2009. We used recent smoking and disease prevalence data obtained for participants in the National Health Interview Survey (NHIS) from 2006 through 2012. We also used NHIS data to estimate relative risks for disease conditions by smoking status, controlling for confounding risk factors. Numerous studies have demonstrated that COPD is often significantly underdiagnosed and underreported in self-reported national health survey data.9- 11 We, therefore, used recent National Health and Nutrition Examination Survey (NHANES) data to analyze the degree of underreporting of COPD and estimate COPD prevalence based on self-reported and spirometry data, thus producing more accurate estimates of smoking-attributable morbidity for this condition. Our analysis also included conditions such as diabetes mellitus and colorectal and stomach cancer, which have been linked to smoking by research in the past decade,1,8,12,13 thus producing estimates of smoking-attributable morbidity for these medical conditions for the first time.
This analysis is a quantitative assessment of smoking-attributable morbidity in the United States using publicly available data sources. Institutional review board approval was not required. We estimated smoking-attributable morbidity in the United States, centering the estimates around the year 2009 to allow for sufficient data to make accurate estimates. We estimated smoking-attributable morbidity by sex, age group, smoking status, and condition as the product of population count (N), smoking prevalence (Ps), disease prevalence among those who have never smoked (Pd |ns), and disease relative risk (RR) by smoking status, using the following formula:
SAMb = N × Ps ×Pd |ns × (RR − 1).
We obtained US population estimates for 2009 from the US Census Bureau.14 We obtained smoking prevalence estimates for current and former smokers for 2006 through 2012 from NHIS data.15 The NHIS is a health survey of the US civilian noninstitutionalized population that is conducted on an annual basis by the National Center for Health Statistics. The NHIS collects cigarette smoking information from approximately 35 000 adults each year. From 2006 through 2012, a total of 190 226 survey participants reported basic smoking status information. We defined current smokers as individuals who reported having smoked at least 100 cigarettes in their lives and reported currently smoking every day or some days, former smokers as individuals who had smoked at least 100 cigarettes and reported that they currently did not smoke at all, and never-smokers as individuals who had never smoked 100 cigarettes.
We also obtained estimates of lifetime smoking-related disease prevalence among never-smokers and disease relative risks by smoking status from 2006-2012 NHIS data. The NHIS participants were asked if a physician or other health professional had ever told them that they had chronic bronchitis during the past 12 months and during their lifetimes for all other smoking-related conditions. We estimated prevalence and relative risks for heart attack, stroke, lung cancer, other smoking-related cancer (bladder; cervix; colon and/or rectum; esophagus; kidney; larynx, windpipe, mouth, tongue, lip, throat, and/or pharynx; liver; pancreas; or stomach), COPD (reported as chronic bronchitis or emphysema), and diabetes. Heart attack, stroke, chronic bronchitis, emphysema, and cancers of the bladder, cervix, esophagus, kidney, larynx, mouth and/or pharynx, and pancreas have been previously identified as smoking-related conditions and were included in the previous CDC analysis of smoking-attributable morbidity.7 Stomach cancer was identified as a smoking-attributable condition in the 2004 Report of the Surgeon General,1 and colorectal and liver cancer and diabetes were identified as smoking-attributable conditions in the 2014 Report of the Surgeon General.8
We estimated disease relative risks by smoking status as prevalence ratios using Poisson regression models16 that also adjusted for age, race or ethnicity (non-Hispanic white, non-Hispanic black, non-Hispanic other race or multiracial, and Hispanic), educational attainment (less than high school graduate, high school graduate, and more than high school graduate), alcohol consumption (consumed fewer than 12 drinks in lifetime, consumed at least 12 drinks in lifetime but none in past year, consumed 1 to 2 drinks on average on days consuming alcohol in past year, and consumed 3 or more drinks on average on days consuming alcohol in past year), and body mass index (as a continuous variable). These covariates have been previously used to estimate relative risks by smoking status for use in estimating smoking-attributable mortality in the United States.17 For smoking-related cancers other than lung cancer, we estimated prevalence ratios for reporting having been diagnosed with any of these cancers to increase the precision of estimates. We also estimated prevalence ratios for each smoking-related cancer type and present the results for cancers for which there were at least 100 cases in the NHIS data. Information for all the covariates was available for 180 515 of the 190 226 survey participants, and these individuals were included in the logistic regression analyses.
We estimated the variance of SAMb as the product of 3 independent variables, treating N as a constant and Ps, Pd|ns, and RR as independent random variables. We approximated the variance of RR (Var[RR]) using the delta method as a Taylor series expansion as RR2Var (logRR). We summed variance estimates by age group and constructed 95% CIs for morbidity estimates by sex and condition (for information about estimate and variance calculations, see eMethods in the Supplement). In addition to estimating smoking-attributable morbidity by condition, we also estimated the number of people in the United States with at least 1 of these conditions that was attributable to smoking using the same general procedures and the disease prevalence ratios for having had any of the smoking-related conditions. We estimated smoking prevalence, disease prevalence, and prevalence ratios for heart attack, stroke, and having had any of the smoking-related conditions for persons aged 35 to 64 years and 65 years and older and disease prevalence ratios for all other conditions for persons 35 years and older. We conducted all analyses using R, version 2.15.2,18 and the Survey package, version 3.29,19 applying the appropriate NHIS sample weights and taking into account the NHIS complex survey design.
Previous studies have found that COPD is the leading cause of smoking-attributable morbidity in the United States7 and that COPD is substantially underreported in national health surveys.9 We therefore evaluated the accuracy of COPD reporting by comparing the prevalence of self-reported COPD, defined as chronic bronchitis or emphysema, with the prevalence of clinically defined COPD using NHANES data from 2007 through 2010.20 The NHANES is an examination-based survey of the US civilian noninstitutionalized population that is conducted by the National Center for Health Statistics and includes approximately 10 000 participants of all ages in each 2-year data cycle. Participants in the NHANES receive a physical examination that includes spirometric tests of respiratory function. These measurements include forced vital capacity (FVC), which is the maximum volume of air that an individual can exhale forcefully after a maximal inhalation, and forced expiratory volume in 1 second (FEV1), which is the volume of air that an individual exhales during the first second of a forced exhalation. Participants in the NHANES with an FEV1:FVC ratio of less than 0.7 are considered to have impaired respiratory function. These participants are then given a β2-adrenergic bronchodilator medication, if medically appropriate, to differentiate between restrictive conditions such as asthma, which tend to respond to bronchodilator treatment, and obstructive conditions such as COPD, which generally do not respond to such treatment. After treatment, participants completed the same spirometric tests as before.
We used NHANES spirometry data to examine the relationship between reporting a diagnosis of COPD and meeting a clinical standard for COPD. We identified a diagnosis of COPD based on survey participants 20 years and older who reported that a physician or other health professional had told them they had chronic bronchitis or emphysema. We used the clinical standard of COPD developed by the Global Initiative for Chronic Obstructive Lung Disorder (GOLD), an international effort sponsored by health organizations including the World Health Organization and the National Heart, Lung, and Blood Institute.21 GOLD defines COPD based on an FEV1:FVC ratio of less than 0.7 after bronchodilator use.21,22 We further classified NHANES participants who met this definition of COPD by the severity of their condition using the grades established by GOLD, which are based on the ratio of observed to predicted FEV1.22 We calculated predicted FEV1 for NHANES participants using formulas developed for this purpose with NHANES data.23
We also used these NHANES data to adjust for underreporting of COPD in the estimation of smoking-attributable morbidity. In addition to calculating estimates in the manner described previously, we also estimated the number of cases of COPD attributable to smoking using COPD prevalence estimates obtained from 2007-2010 NHANES data in place of NHIS data. In this case, COPD prevalence was defined as the proportion of NHANES participants 35 years and older who either reported having been diagnosed with chronic bronchitis or emphysema by a health professional or who met the GOLD definition for moderate to very severe COPD based on spirometry data among the survey participants who had both medical diagnosis and spirometry data. The threshold of moderate COPD was used to be conservative in the estimation of additional smoking-attributable cases of COPD. The resulting estimate of the number of cases of COPD attributable to smoking was used with the estimates for other conditions calculated in the manner described previously to obtain the total number of major medical conditions attributable to smoking in the United States.
Table 1 and Table 2 present US smoking and smoking-related disease prevalence estimates, respectively, from NHIS data. Current and former smokers were consistently more likely to have had at least 1 smoking-related condition and multiple smoking-related diseases compared with never-smokers. Approximately 47.5% of male and 44.9% of female current smokers 65 years and older reported having been diagnosed with at least 1 smoking-related condition, and 16.9% of the men and 14.3% of the women reported having been diagnosed with multiple conditions. Diabetes was the most prevalent condition, with 16 522 (11.8%; 95% CI, 11.5%-12.0%) NHIS participants aged 35 years and older reporting that they had been diagnosed with the condition.
Table 3 shows adjusted disease prevalence ratios by sex, age, and smoking status as well as the number of cases among NHIS participants by disease. Estimated prevalence ratios were higher for current and former smokers for most conditions. Prevalence ratios were particularly high for lung cancer, with prevalence ratios from 4.45 to 9.35, and COPD, with prevalence ratios from 2.02 to 4.00.
We also analyzed the accuracy of self-reporting of a medical diagnosis of COPD using data for NHANES participants who met a clinical standard for COPD. Only 42 of 384 participants (10.9%) who met the GOLD definition for having COPD, with a posttreatment FEV1:FVC ratio of less than 0.7, reported ever having been told by a health professional that they had chronic bronchitis or emphysema. Results were generally consistent among never-smokers (7 of 94 participants who met the GOLD definition reported having been diagnosed with chronic bronchitis or emphysema), former smokers (20 of 144 participants), and current smokers (15 of 146 participants). A similar proportion of NHANES participants who met the GOLD definition for having moderate to very severe COPD, with an FEV1:FVC ratio less than 0.7 and an observed FEV1 to predicted FEV1 ratio less than 0.8, reported never having received a diagnosis of COPD. Of these survey participants, 112 of 138 (81.2%) reported that they had never been told by a health professional that they had chronic bronchitis or emphysema.
We next estimated smoking-attributable morbidity for major medical conditions among US adults, first using NHIS medical diagnosis information for prevalence for all conditions. We estimated that 6.9 million (95% CI, 6.5-7.4 million) individuals reported a major smoking-attributable condition in 2009 based on self-reported NHIS lifetime disease prevalence information. These individuals reported an estimated 10.9 million (95% CI, 10.3-11.5 million) lifetime cases of smoking-attributable disease. Chronic obstructive pulmonary disease accounted for the largest number of these conditions, with 4.3 million (95% CI, 4.0-4.7 million) cases. Heart attacks represented another 2.3 million (95% CI, 2.0-2.5 million) conditions and diabetes 1.8 million (95% CI, 1.5-2.1 million).
We then estimated smoking-attributable morbidity for the United States using prevalence estimates for COPD obtained from NHANES data. Prevalence of COPD, as defined by either self-reported medical diagnosis or spirometry data, was 3.9% (95% CI, 2.8%-5.0%) for men and 7.5% (95% CI, 5.6%-9.3%) for women among NHANES never-smokers 35 years and older. The corresponding prevalence estimates for NHIS never-smokers based on self-reported medical diagnosis were 2.2% (95% CI, 2.0%-2.4%) for men and 4.4% (95% CI, 4.2%-4.6%) for women. Using COPD prevalence estimates from NHANES data, there were an estimated 14.0 million (95% CI, 12.9-15.1 million) lifetime major medical conditions attributable to smoking in the United States in 2009, as shown in Table 4. The largest cause of smoking-attributable morbidity in the United States was still COPD, with an estimated 7.5 million (95% CI, 6.5-8.5 million) cases attributable to smoking, but this number is 70% higher than the estimated cases based on self-reported prevalence data.
We have presented updated, nationally representative estimates of smoking-attributable morbidity for the United States that control for confounding risk factors and account for the statistical uncertainty of estimates. These estimates demonstrate that smoking accounts for millions of serious medical conditions in the United States that could be avoided in the absence of cigarette use. Our results also indicate that previous estimates may have substantially underestimated smoking-attributable morbidity in the United States given that our results support the notion that COPD prevalence is often substantially underreported by participants in national health surveys.
Our study offers several methodological and empirical contributions to the existing research. We estimated smoking-related disease prevalence and relative risks using nationally representative data from 2006 through 2012 from approximately 180 000 survey participants. Previous CDC estimates of smoking-attributable morbidity used disease prevalence and relative risk data from approximately 20 000 adults who participated in the NHANES III from 1988 to 1994.7 We also estimated relative risks adjusted for important confounding risk factors such as race or ethnicity, educational attainment, and alcohol consumption. We also calculated the full variance of our estimates of US smoking-attributable morbidity, which, to our knowledge, has not been generally done previously.
Our estimates are generally consistent with previous CDC estimates, although there may have been some changes in smoking-attributable morbidity in the United States over time. In terms of estimates by disease, our estimates for heart attack and stroke are similar to the CDC estimates. We estimated somewhat more cases of lung cancer and fewer cases of other cancers, even though we included some additional smoking-related cancers in our analysis. Our estimated prevalence ratios for diabetes are also consistent with a previous study12 that found an adjusted relative risk for diabetes of 1.44 (95% CI, 1.31-1.58) among active smokers in a meta-analysis of prospective cohort studies.
Our results are also consistent with a previous study9 that found that COPD was substantially underreported among persons with low lung function based on analysis of NHANES III spirometry data. This previous study found that 63% of NHANES participants with low lung function did not report a current or previous diagnosis of emphysema, chronic bronchitis, or asthma. The study defined low lung function as an FEV1:FVC of less than 0.7 and observed FEV1 to predicted FEV1 of less than 0.8, which are standards that are generally consistent with the GOLD definition for moderate to very severe COPD.22,24 It should be noted that other research organizations, such as the American Thoracic Society, European Respiratory Society, and National Institute for Clinical Excellence, have developed diagnostic criteria for COPD that differ somewhat from the GOLD definition and that these differences can affect prevalence estimates.25- 27 Researchers also continue to debate the best way to characterize COPD and its severity.28,29 One reason COPD may be underreported in health surveys compared with spirometry testing is the lack of a clinical diagnosis of COPD by a health professional. Clinical guidelines commonly advise physicians to use spirometry to diagnose chronic airway obstruction in patients who report symptoms such as wheezing, chronic cough, or physical limitation due to respiratory issues.30 These guidelines also advise physicians not to screen asymptomatic patients for COPD in this manner owing to the economic and health care costs associated with subsequent screening and treatment. As a result, individuals with slowly declining respiratory function or individuals who have become accustomed to some degree of chronic airway obstruction may not report these conditions to physicians and consequently would not be screened for or diagnosed with COPD.
Self-reported diagnosis of COPD in the NHANES and NHIS is further complicated by the fact that survey participants were asked if they had been told by a health professional that they had emphysema or chronic bronchitis and that the term COPD was not used in the surveys. Reporting of diagnoses of chronic airway obstruction in national health surveys could be improved if survey participants were asked if they have been diagnosed by a health professional with chronic bronchitis, emphysema, or COPD. The CDC Behavioral Risk Factor Surveillance System31 recently introduced this type of question and reported a lifetime prevalence for these conditions of 6.3% (95% CI, 6.2%-6.5%) among US adults 18 years and older in 2011, which is slightly higher than the equivalent NHIS prevalence for US adults from 2006 through 2012 of 5.3% (95% CI, 5.1%-5.4%).
Our study is subject to certain limitations. Disease prevalence and risks are calculated from self-reported information on medical diagnosis, and conditions may not be accurately diagnosed or reported. There also may be additional smoking-related conditions that are not included in our analysis. The International Agency for Research on Cancer13 has concluded, for example, that ovarian cancer, specifically mucinous tumors, is caused by smoking. The 2014 Report of the Surgeon General8 noted that estimates of the cardiovascular disease burden caused by smoking do not include important conditions such as history of cardiovascular surgical procedures, congestive heart failure, and peripheral arterial disease. The report also identified several other medical conditions as attributable to smoking, including pneumonia, rheumatoid arthritis, and macular degeneration, all of which would add to the total disease burden caused by smoking. Secondhand smoke exposure could also contribute to the total disease burden caused by smoking, although quantifying the morbidity and mortality burden from secondhand smoke exposure presents its own methodological challenges that go beyond the scope of this analysis.32
Our study confirms that cigarette smoking remains a major cause of preventable disease in the United States. Overall, we estimate that US adults in 2009 had had at least 14 million serious medical conditions that were attributable to cigarette smoking. This estimate corrects for underreporting and diagnosis of COPD given that evidence presented here and in previous studies suggests that COPD is substantially underreported in national health survey data. The resulting estimate indicates that the number of major smoking-attributable medical conditions in the United States is larger than has been previously reported, demonstrating the need for vigorous smoking prevention efforts. The disease burden of cigarette smoking in the United States remains immense, and updated estimates indicate that COPD may be substantially underreported in health survey data.
Accepted for Publication: August 2, 2014.
Corresponding Author: Brian L. Rostron, PhD, MPH, Center for Tobacco Products, US Food and Drug Administration, 10903 New Hampshire Ave, Building 75, Room 4404, Silver Spring, MD 20993 (firstname.lastname@example.org).
Published Online: October 13, 2014. doi:10.1001/jamainternmed.2014.5219.
Author Contributions: Dr Rostron had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: All authors.
Acquisition, analysis, or interpretation of data: Rostron, Chang.
Drafting of the manuscript: Rostron, Pechacek.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Rostron.
Administrative, technical, or material support: Chang, Pechacek.
Study supervision: Pechacek
Conflict of Interest Disclosures: None reported.
Disclaimer: The views and opinions expressed in this article are those of the authors only and do not necessarily represent the views, official policy, or position of the US Department of Health and Human Services or any of its affiliated institutions or agencies.
Additional Contributions: Catherine Corey, MSPH, and Priscilla Callahan-Lyon, MD, of the US Food and Drug Administration provided technical advice regarding chronic obstructive pulmonary disease classification and reporting, and Benjamin Apelberg, PhD, of the US Food and Drug Administration coordinated this project. None received financial compensation.