Cigarette smoking has been identified as a leading cause of mortality in developed countries1 and mortality differences among developed countries.2 There are numerous methods to estimate smoking-attributable mortality,3 but published estimates often differ in the number of smoking-attributable deaths.4,5 The Centers for Disease Control and Prevention (CDC)6 periodically publishes estimates of smoking-attributable mortality in the United States based on smoking prevalence and relative mortality risks of smoking for selected causes. These estimates, however, are not produced for specific ages, and the CDC's methodology uses smoking prevalence and relative risks for broad age groups such as 35–64 and 65 and over. The CDC's method also does not control for potential confounding factors. Smoking-attributable deaths in the United States are estimated in this study by age, controlling for confounding factors.
This estimate of smoking-attributable mortality in the United States is based on hazard ratios for smokers, death rates for never-smokers, and smoking prevalence figures. Hazard ratios (HRs) by smoking status were estimated from National Health Interview Survey, which is a nationally representative health survey of the US civilian, noninstitutionalized population conducted by the National Center for Health Statistics (NCHS).7 Approximately 35,000 households and 87,500 persons are included in the survey each year. Additional information for one adult and, if applicable, one child in each family in every household is also collected. Survey participants are followed for mortality by the periodic matching of their records to the National Death Index. The National Death Index is maintained by NCHS and includes death certificate information for all US decedents since 1979.8
Hazard ratios by smoking status were calculated for those responding to the National Health Interview Survey Sample Adult Core questionnaire from 1997 to 2004, based on mortality follow-up through the end of 2006.9 A total of 241,098 survey participants were eligible for mortality follow-up, and 17,039 of these participants were identified as deceased through linkage with the National Death Index. Hazard ratios were calculated for heavy, moderate, and light current cigarette smokers and former smokers compared with never-smokers. Those who reported in the baseline interview that they smoked 2 or more packs of cigarettes a day were defined as heavy smokers, those who reported one to 2 packs a day were moderate smokers, and those smoking less than one pack a day were light smokers. Those who reported having smoked fewer than 100 cigarettes in their lives were classified as never-smokers. Hazard ratios were calculated by smoking status, with and without controls for confounding risk factors. This methodology is similar to that employed by Rogers et al4 who estimated US smoking-attributable mortality based on data from a 1990 National Health Interview Survey supplement and mortality follow-up. Rogers et al calculated relative mortality risks using detailed smoking status, a discrete-time hazards model, and controls for various potential confounding factors. They then used these risks along with NHIS smoking prevalence figures to estimate smoking-attributable mortality in the United States in 2000. Unlike the Rogers et al study, the present study estimates relative mortality risks by smoking status for specific age groups, a modification made possible in part by the much larger number of deaths analyzed here (over 17,000, compared with 3097 for Rogers et al).
The other mortality risk factors included as covariates in our analysis were race, Hispanic origin, marital status, family income, educational attainment, alcohol consumption, and body mass index. All of these covariates were treated as categorical variables. Other covariates were also considered for inclusion in the analysis. Participation in moderate or vigorous physical activity was tested in the model as a covariate, but was excluded from the final analysis. Inability to engage in physical activity may be the result of smoking-induced morbidity and a predictor of smoking-attributable mortality rather than a true confounding factor.
Hazard ratios for 10-year age groups were calculated using a Cox proportional hazards model. All analyses were conducted with R version 2.11.110 and the Survey package, using the appropriate National Health Interview Survey person-weights and taking into account the NHIS's complex survey design. The proportional-hazard assumption was evaluated for each covariate through χ2 tests for nonzero slope of the scaled Schoenfeld residuals. The covariates were found generally to satisfy the assumption. The consistency of the hazard ratios by smoking status was also evaluated by estimating the hazard ratios with follow-up restricted through the end of the following 2 calendar years. Hazard ratios by smoking status from this shorter follow-up period were very similar to those found with the longer follow-up period (eAppendix [http://links.lww.com/EDE/A466]). These results indicate that the hazard ratios for smoking were generally consistent during the follow-up period and that the depletion of susceptible individuals did not appreciably affect the hazard ratios over time, as has been observed in some other studies.11
All-cause death rates for never-smokers were used to calculate absolute mortality risks for smokers. Death rates for US never-smokers from 2002 through 2006 were estimated by sex and age based on mortality follow-up in these years for persons responding to the Adult Questionnaire from the 1997–2004 National Health Interview Survey cohorts. Each death rate was then multiplied by an adjustment factor by sex and age to ensure that the overall 2002–2006 death rates for NHIS participants by sex and age equaled the means of the annual death rates for the US population during this period.12 Adjusted death rates for current and former smokers for 2002–2006 were estimated by multiplying the resulting never-smoker death rates by the appropriate hazard ratios estimated previously. These adjusted death rates for smokers were then used with population estimates by smoking status to estimate the average number of annual smoking-attributable deaths in the United States during the period. Smoking prevalence by sex-age group in the United States for 2002–2006 was estimated from NHIS survey data. The resulting proportions were multiplied by US population counts for the period13 to produce the average number of people by smoking status in each sex-age group. The average number of people with each smoking status was multiplied by the difference between the adjusted death rate for that smoking group and the corresponding death rate for never smokers to produce the average number of annual smoking-attributable deaths (SAD) for that group. This calculation can be represented for sex x, age group a, and smoking status s as:
where Prev is smoking prevalence, Pop is population count, and m x,a,s and m x,a,n are death rates for the smoking group and never-smokers, respectively. The total number of average annual smoking-attributable deaths was then calculated as the sum of these estimates by sex-age group.
Tables 1 and 2 present hazard ratios by sex and age group for smoking status, estimated with and without additional covariates to control for other mortality risk factors. Table 3 shows the estimated average annual death rates for never-smokers in the United States from 2002 through 2006. Results from the survival analysis shown in Tables 1 and 2 agree with most expectations. Hazard ratios generally increase with smoking intensity and with age from 45 to 74 years, before declining somewhat at older ages. Hazard ratios for current smokers generally show a modest decrease after adjustment for confounding factors, which is similar to the results found by Rogers et al.4 Relative hazards are higher for blacks than for other ethnic groups, and are higher for single, widowed, and divorced persons than married persons. Relative hazards decline with increasing educational attainment and family income. Relative hazards are slightly lower for people who are somewhat overweight at older ages, compared with those who are underweight or normal weight, as has been found previously.14
The Figure shows the estimated age distribution of annual smoking-attributable deaths in the United States for 2002–2006. The age distribution of deaths was similar for men and women, although a greater proportion of smoking-attributable deaths occurred at ages 75 and over for women than for men. The Figure also shows that most smoking-attributable deaths of persons aged 75 and over occurred among former smokers. Overall, in the United States between 2002 and 2006, an estimated 291,000 men and 220,000 women aged 35 and over died each year due to smoking. Confidence intervals (CIs) for estimates of smoking-attributable deaths by sex and age group were prepared using a bootstrap technique; they are presented and explained in the eAppendix (http://links.lww.com/EDE/A466). The 95% confidence interval for total smoking-attributable deaths for men was 247,000 to 357,000 and for women it was 201,000 to 267,000. These estimates are generally consistent with, although somewhat higher than, published estimates from the CDC, particularly for women. The CDC6 estimated that smoking was responsible for 270,000 average annual deaths for males and 174,000 average annual deaths for females in the period 2000–2004 in the United States.
Estimates of smoking-attributable deaths among current smokers declined with use of hazard ratios adjusted for confounding factors, particularly for men (153,000 smoking-attributable deaths among male current smokers based on unadjusted hazard ratios, compared with 123,000 deaths based on adjusted hazard ratios). Estimates of smoking-attributable deaths among former smokers actually increased, however, with the use of adjusted hazard ratios, particularly for women (107,000 deaths among female former smokers based on unadjusted hazard ratios, compared with 137,000 deaths based on adjusted hazard ratios). The overall effect of adjustment was to lower the estimates of smoking-attributable deaths for men by 8000 deaths and increase estimates for women by 23,000 deaths.
These estimates of total smoking-attributable mortality for the United States offer certain substantive advantages over other methods. Estimates are provided for specific age groups, and adjusted for the confounding effects of other mortality risk factors. The overall estimates presented here are somewhat higher than those published by the CDC. These differences may result in part from some of the features of the methodology used here. For example, the CDC uses relative risks that were calculated from the American Cancer Society Cancer Prevention Study II (CPS-II). These data were collected from 1982 through 1988, and relative risks could have changed over time due to changes in smoking duration and intensity among smokers.6 Such changes in smoking behavior may particularly affect relative risks for women, given that smoking duration and intensity increased among US female smokers during much of the second half of the 20th century.15 The present estimates generally avoid this problem by using more recent National Health Interview Survey data. Another cause of difference in the estimates may be the effect of controlling for confounding factors. As noted in the Results section, estimates of smoking-attributable deaths among women increased with adjustment for confounding factors, due to higher adjusted hazard ratios for former smokers. This result is similar to those of Malarcher et al16 and Thun et al,17 who found increases in estimates of smoking-attributable deaths for women when controlling for confounding factors.
Our method is also similar to one used previously by Rogers et al,4 but our results are not. Those researchers estimated that there were 189,000 smoking-attributable deaths for men and 133,000 smoking-attributable deaths for women aged 35 and over in the United States in 2000. Their method did not consider relative mortality risks by smoking status for particular age groups, even though results presented here indicate that these risks vary substantially by age. Rogers et al also included as covariates in their model risks factors such as being underweight or not engaging in physical activity that could themselves be the result of smoking-attributable morbidity. These risk factors might thus be viewed more properly as intermediate variables on the causal pathway between smoking and mortality, rather than true confounders. Inclusion of these covariates in a survival analysis could therefore inadvertently control for a certain portion of the mortality risk that should be attributed to smoking, and may explain why their estimates were lower than those produced by the CDC or presented here.
1. Ezzati M, Lopez AD, Rodgers A, et al. Selected major risk factors and the global and regional burden of disease. Lancet
2. Staetsky L. Diverging trends in female old-age mortality: a reappraisal. Demogr Res
3. Perez-Rios M, Montes A. Methodologies used to estimate tobacco-attributable mortality: a review. BMC Public Health
4. Rogers RG, Hummer RA, Krueger PM, et al. Mortality attributable to cigarette smoking in the United States. Popul Dev Rev
5. Preston SH, Glei DA, Wilmoth JR. A new method for estimating smoking-attributable mortality in high-income countries. Int J Epidemiol
6. Centers for Disease Control and Prevention. Smoking-attributable mortality, years of potential life lost, and productivity losses—United States, 2000–2004. MMWR Morb Mortal Wkly Rep
7. Pleis JR, Lethbridge-Cejku M. Summary health statistics for US adults: National Health Interview Survey, 2005. Vital Health Stat
8. MacMahon B. The national death index. Am J Pub Health
10. R Development Core Team. R: A Language and Environment for Statistical Computing, Version 2.10.0. Vienna, Austria: R Development Core Team; 2010.
11. Hernan MA. The hazards of hazard ratios. Epidemiology
12. Heron M, Hoyert DL, Murphy SL, et al. Deaths: Final data for 2006. National Vital Statistics Report. Hyattsville, MD: National Center for Health Statistics; 2009.
14. Flegal KM, Graubarad BI, Williamson DF, et al. Excess deaths associated with underweight, overweight, and obesity. JAMA
15. Forey B, Hamling J, Lee P, et al. International Smoking Statistics
. London: Oxford University Press; 2002.
16. Malarcher AM, Schulman J, Epstein LA, et al. Methodological issues in estimating smoking-attributable mortality in the United States. Am J Epidemiol
17. Thun MJ, Apicella LF, Henley SJ. Smoking vs other risk factors as the cause of smoking-attributable deaths. JAMA