Figure 1. Flow diagram of the systematic literature search. HR indicates hazard ratio; OR, odds ratio; and RR, relative risk.
Figure 2. Results of meta-analyses of 15 studies on current smoking and all-cause mortality.
Figure 3. Results of meta-analyses of 14 studies on former smoking and all-cause mortality.
Gellert C, Schöttker B, Brenner H. Smoking and all-cause mortality in older people:
systematic review and meta-analysis. Arch Intern Med.. doi:10.1001/archinternmed.2012.1397
Factors adjusted for in the 17 included studies
eTable 2.. Studies reporting associations of amount of smoking with all-cause
eTable 3. Former smokers stratified by length of time since smoking cessation
eTable 4. Absolute mortality rates per 100,000 person-years
Forest plot of meta-analyses for current smokers compared to never
eFigure 2. Forest plot of meta-analyses for male and female current smokers
compared to never smokers
eFigure 3. Forest plot of meta-analyses for current smokers in age groups 60-69
years, 70-79 years and =80 years compared to never smokers
eFigure 4. Forest plot of meta-analyses for former smokers compared to never
eFigure 5. Forest plot of meta-analyses for male and female former smokers
compared to never smokers
eFigure 6. Forest plot of meta-analyses for former smokers in age groups 60-69
years, 70-79 years and =80 years compared to never smokers
Gellert C, Schöttker B, Brenner H. Smoking and All-Cause Mortality in Older PeopleSystematic Review and Meta-analysis. Arch Intern Med. 2012;172(11):837-844. doi:10.1001/archinternmed.2012.1397
Author Affiliations: Division of Clinical Epidemiology and Aging Research, German Cancer Research Center (DKFZ), Heidelberg, Germany.
Background Smoking is an established risk factor of premature death. However, most pertinent studies primarily relied on middle-aged adults. We performed a systematic review and meta-analysis of the empirical evidence on the association of smoking with all-cause mortality in people 60 years and older.
Methods A systematic literature search was conducted in multiple databases including MEDLINE, EMBASE, and ISI Web of Knowledge and complemented by cross-referencing to identify cohort studies published before July 2011. Core items of identified studies were independently extracted by 2 reviewers, and results were summarized by standard methods of meta-analysis.
Results We identified 17 studies from 7 countries. Current smoking was associated with increased all-cause mortality in all studies. Relative mortality (RM) compared with never smokers ranged from 1.2 to 3.4 across studies and was 1.83 (95% CI, 1.65-2.03) in the meta-analysis. A decrease of RM of current smokers with increasing age was observed, but mortality remained increased up to the highest ages. Furthermore, a dose-response relationship of the amount of smoked cigarettes and premature death was observed. Former smokers likewise had an increased mortality (meta-analysis: RM, 1.34; 95% CI, 1.28-1.40), but excess mortality compared with never smokers clearly decreased with duration of cessation. Benefits of smoking cessation were evident in all age groups, including subjects 80 years and older.
Conclusions Smoking remains a strong risk factor for premature mortality also at older age. Smoking cessation is beneficial at any age.
It is well established that smoking is hazardous to health.1- 3 Smoking is one of the major risk factors for multiple chronic diseases, such as cardiovascular diseases4- 6 and cancer,7- 9 as well as for mortality from the leading causes of death and consequently also for all-cause mortality. Smoking is one of the 10 leading risk factors for death, and according to estimates of the World Health Organization, it is responsible for 12% of male deaths and 6% of female deaths in the world.2 In the 21st century, a billion deaths due to smoking are expected if no changes in smoking behavior are achieved.7
However, as for most other risk factors, epidemiological evidence mostly relies on studies conducted among middle-aged adults, and specific evidence for the impact of smoking at older age is still sparse. Furthermore, evaluation of the impact of smoking and smoking cessation at old age may be particularly challenging owing to a number of methodological issues, such as attenuated relative risks in the presence of strongly increased absolute levels of mortality among both smokers and nonsmokers and the “depletion of susceptibles” effect.10,11
In the present article, we provide a thorough review and meta-analysis of studies assessing the impact of smoking on all-cause mortality in people 60 years and older, paying particular attention to the strength of the association by age, the impact of smoking cessation at older age, and factors that might specifically affect results of epidemiological studies on the impact of smoking in an older population.
A protocol was developed based on widely recommended methods for systematic reviews of observational studies.12,13 A systematic literature search was carried out to identify cohort studies published before July 2011 that report on the association of smoking and all-cause mortality in individuals 60 years and older. The electronic databases MEDLINE, EMBASE, and ISI Web of Knowledge were searched for relevant articles by the following search strategy: (Smoking OR Tobacco OR Cigarette) AND (Aged OR Old OR Elderly) AND (Mortality OR Death OR Dead) AND (Cohort OR Longitudinal). No language restrictions were used.
To be included, studies had to report smoking status and examine the outcome of interest (all-cause mortality) in people 60 years and older in the general population in a longitudinal cohort study design. We excluded studies that did not publish separate results for older people or did not estimate a relative-effect measure (hazard ratio, odds ratio, or relative risk) for the comparison of current or former smokers with never smokers. Furthermore, studies that did not reflect random general population samples were excluded.
Study selection and extraction of study characteristics from included studies were independently performed by 2 reviewers (C.G. and B.S.). Any disagreement was resolved by review and discussion among the authors. Cross-referencing in finally included publications was used to verify the completeness of the literature search. Study quality was evaluated using established protocols.13,14 Good-quality studies were deemed to have the following features: (1) random recruitment of participants or representative population, (2) detailed ascertainment of smoking variables in face-to-face interview, (3) reporting on completeness of registry-based mortality follow-up, and (4) adjustment (or stratification) for important covariates (age, sex, alcohol, and body mass index). Associations reported in the studies included relative mortality rates and hazard ratios (typically derived from Cox proportional hazards models) and are referred to as “relative mortality” (RM) in this report.
The “Comprehensive Meta-analysis” software (Biostat) was used for the conduction of all meta-analyses. In a conservative approach, the random-effects estimates, which allow for variation of true effects across studies, were taken as “main results.”15 Random-effects estimates were derived using the DerSimonian-Laird method.16,17 Heterogeneity was assessed by the I2 and the Q statistics. To explore heterogeneity, we performed subgroup analyses according to study population (age, sex, and region of study conduction), according to characteristics of study design (sample size and follow-up period) and study quality score. The funnel plot, Begg and Mazumdar rank correlation test, and Egger test of the intercept were used to assess indications of publication bias.18
The process of the systematic literature search is displayed in a flow diagram in Figure 1. In brief, the search in the electronic databases identified 8802 articles. After exclusion of duplicates, title and abstract selection, and full-text selection, 17 studies met the inclusion criteria of this review.19- 35 Cross-referencing did not identify any additional articles.
A description of the baseline characteristics of the included studies is given in Table 1. The studies were published between 1987 and 2011 and were conducted in 17 different cohorts. Seven studies were conducted in the United States,22,23,26,27,29,31,33 3 in China,20,25,28 2 in Australia19,30 and 2 in Japan.24,35 The remaining studies were from England,21 Spain,34 and France.32 The follow-up time ranged from 3 to 50 years, and the size of the study populations ranged from 863 to 877 243 participants. Fourteen studies provided data on all-cause mortality according to smoking for both sexes,19,22- 32,34,35 whereas 2 studies20,21 reported results only for men and 1 study only for women.33
Study quality scores (range, 0-4) averaged 2.35, with a proportion of high-quality studies (quality score ≥3) of 35%.
An overview of factors adjusted for in the 17 studies is given in eTable 1. Age and sex were adjusted for in all studies that were not restricted to a single age group or a single sex. Systolic blood pressure, alcohol consumption, and physical activity were controlled for in 9,20,23- 25,27,28,30,33,35 9,20,25- 28,30,31,34,35 and 6 studies,20,23,26,28,33,34 respectively. Other factors were adjusted for to a heterogeneous degree in a minority of studies only.
Estimated RM for current smokers compared with never smokers is given in Table 2, stratified by sex and age groups. For currently smoking men, the point estimates of RM compared with never smokers ranged from 1.3 to 3.4, with an outlier of 0.5 in men 90 years or older in the study of Paganini-Hill et al,29 likely owing to a very small sample size (numbers or confidence intervals not provided). For women, the point estimates of RM ranged from 1.2 to 2.5. Three studies reported combined mortality ratios for currently smoking men and women. Those point estimates ranged from 1.4 to 3.0.
Results of meta-analyses are given in Table 3, Figure 2 and eFigures1, 2, and 3. In a meta-analysis of 15 studies reporting mortality of current smokers compared with never smokers, an RM of 1.83 (95% CI, 1.65-2.03) was found for both sexes and all age groups combined. In this analysis, age- and sex-specific estimates were first combined to a study-specific summary random-effects estimate for those studies that reported RM by age or sex only. In sex-specific meta-analyses, very similar results were obtained for men and women. Meta-analyses of age-specific estimates of RM yielded summary estimates of 1.94 (95% CI, 1.57-2.40), 1.86 (95% CI, 1.55-2.22), and 1.66 (95% CI, 1.30-2.12) for age groups 60 to 69 years, 70 to 79 years, and 80 years or older, respectively. Population-specific meta-analyses indicated lower RM for Asian populations in comparison to European, United States, and Australian populations. There were no indications of publication bias. Despite modest variation of RM estimates across studies, heterogeneity was ascertained in almost all meta-analyses. These high values were mainly driven by the study of Taylor et al,31 the largest study by far, which showed relatively high estimates of RM. After excluding this study, no indications of heterogeneity persisted.
Estimates of RM for former smokers compared with never smokers are also given inTable 2. The point estimates of RM ranged from 1.1 to 2.2 for male former smokers and from 0.8 to 2.1 for female former smokers. Combined estimates for both sexes are in a comparable range (1.0-1.9).
Results of meta-analyses are given in Table 3, Figure 3, and eFigures 4, 5, and 6. A meta-analysis of 14 studies reporting hazard ratios for all-cause mortality for former smokers compared with never smokers showed an RM of 1.34 (95% CI, 1.28-1.40) for both sexes and all age groups combined. In this analysis, age- and sex-specific estimates were first combined to a study-specific summary random-effects estimate for those studies that reported RM by age or sex only. Sex-specific meta-analyses yielded very similar results for men and women. A decrease of RM was observed with age in age-specific meta-analyses, but an increased mortality of former smokers compared with never smokers persisted up to the highest ages. Population-specific meta-analyses showed very similar results for all populations. There were no indications of publication bias. Heterogeneity was found for meta-analyses of age groups 70 to 79 years and 80 years or older and for studies from US populations. After exclusion of the very large study by Taylor et al,31 no indications of heterogeneity persisted.
No major variation in summary estimates of RM was seen in subgroup meta-analyses by duration of follow-up, sample size, and study quality (Table 3). There were no indications of publication bias. Heterogeneity was found for almost all meta-analyses.
eTable 2 provides information on 10 studies reporting on the association of the amount of smoking and all-cause mortality. Seven of the included studies reported on the average number of cigarettes smoked per day in current smokers, and another 2 studies reported on numbers of cigarettes smoked per day among ever smokers (current and former smokers combined). With few exceptions, a clear dose-response relation of increasing mortality with increasing number of cigarettes was observed. In 3 studies, the impact of the amount of smoking on mortality was also investigated using the concept of pack-years, a measure for the amount of smoking over the full life-span.36 Again, a clear dose-response with RM was observed.
To determine the benefit of smoking cessation at older age, mortality was evaluated in relation to age at smoking cessation and the number of years since smoking cessation of former smokers in 5 studies in which such information was available (eTable 3). With few exceptions, a clear dose-response relationship of decreasing RM with time since cessation was observed consistently.
In eTable 4 absolute mortality rates are presented for studies that either provided them directly or had valid information for calculation of those. A steep rise of absolute mortality rates with increasing age can be seen. Current smokers show highest absolute mortality rates in all studies. In studies with age-specific mortality rates, mortality differences between current smokers and never smokers varied by age groups, but the ranking of differences by age groups was not consistent across studies.
To our knowledge, this is the first systematic review and meta-analysis on the impact of smoking on all-cause mortality focusing on older people. Summarizing the results from 17 cohort studies, we observed an 83% increased mortality for current smokers and a 34% increased mortality for former smokers compared with never smokers. Relative mortality of former smokers decreased with time since cessation. A dose-response relationship of the amount of currently smoked cigarettes and premature death was consistently observed. Current smoking was significantly associated with increased mortality even in the oldest age groups, for both sexes and people from different geographical regions.
Smoking is an established risk factor for premature mortality.3,37,38 However, most reviews and studies on smoking and mortality used broad age ranges, focused on middle-aged adults,37,39,40 included only subjects with certain diseases,41 or investigated disease-specific incidence or mortality.3,42- 44 For some causes of death, such as cancers of mouth, pharynx, and larynx, an up-to 10-fold increased mortality was reported for current smokers compared with never smokers.3
In this review and meta-analysis on the association of smoking and all-cause mortality at older age, current and former smokers showed an approximately 2-fold and 1.3-fold risk for mortality, respectively. Relative mortality for both current and former smokers slightly decreased with increasing age. One plausible explanation can be the “depletion of susceptibles” effect.10,11 Smokers who are still alive at oldest age might be less likely to die from smoking because they showed a tolerance for harmful smoking effects in the past, while smokers who were more susceptible to harmful smoking effects have died already at younger age and dropped out of the population at risk. A second explanation for the decrease in smoking-related RM risk at older age may be the steep rise of absolute mortality above age 70 years among both smokers and nonsmokers that attenuates the magnitude of relative-effect estimates even in case of an increasing mortality difference on the absolute scale. The finding of a notable excessive mortality up to the oldest ages for both current and former smokers that persists despite these potential reasons for attenuation underlines the strength of smoking as a key risk factor for premature mortality also at older age.
Conversely, smoking cessation is an established preventive factor for premature mortality.45 However, most studies have been carried out in middle-aged populations.22,38 This review and meta-analysis demonstrates that the relative risk for death notably decreases with time since smoking cessation even at older age. However, it has to be noted that results are based on former smokers at baseline only, who already have survived some time after cessation. There is a lack of results on baseline current smokers who quit smoking during follow-up compared with those who continued to smoke. Also, analyses are based on former smokers compared with never smokers and not compared with current smokers. Furthermore, some former smokers might have quit smoking owing to ill health, and their mortality risk could be higher than the mortality risk of those who continued smoking. Nevertheless, these results strongly suggest that smoking cessation is effective for mortality reduction also at older age, a suggestion that should be corroborated by intervention studies, ideally with interventions specifically designed and developed for this target group. Although older smokers have been included in successful smoking cessation studies,45,46 they were typically a minority in the study samples, and there is a lack of specific data on efficacy of smoking cessation programs among older smokers. Perspectives for successful smoking cessation appear to be particularly good after diagnoses of major smoking-related diseases, such as myocardial infarction, stroke, or cancer,47 when people are personally confronted with the harmful effects of smoking. Preferably, however, smoking cessation should be achieved prior to manifestation of such serious diseases. If smoking cessation cannot be achieved, smoking less may attenuate the mortality risk. Dose-response relationships of the amount of daily smoked cigarettes with total mortality were remarkably constant across all studies.
In the interpretation of the results, several limitations should be kept in mind. Although we searched 3 databases, ie, Ovid MEDLINE, EMBASE, and ISI Web of Knowledge, and performed extensive checks for completeness by cross-referencing, it cannot be guaranteed that all relevant studies were found. In particular, we did not seek additional published or unpublished reports from experts. Other limitations are that the included studies had varying follow-up periods and age ranges. Furthermore, the age ranges did not always completely match with those chosen for the age-specific meta-analyses. Each study was adjusted for a different set of covariates, which might have contributed to heterogeneity in meta-analyses. In addition, important confounders were not always fully controlled for, which might have resulted in some overestimation of effects due to residual confounding. Furthermore, disparities in smoked products (eg, pipes, cigars, cigarillos) and the resulting impact on the outcome of interest could not be differentiated. Another limitation is the lack of information on duration of smoking or age at initiation in most of the included studies. Furthermore, we had to exclude 18 studies with approximately 100 000 participants that did not publish separate results for our target age group. Finally, given the observational nature of cohort studies, causal conclusions should be drawn with due caution. Nevertheless, causality of the associations is strongly supported by a number of important criteria, including the strength of the associations, consistency of results across studies, biological plausibility, and clear dose-response patterns.
Notwithstanding their limitations, the results presented in this systematic review demonstrate the need for effective smoking cessation programs because the hazardous effects of smoking persist even in oldest age. Even older people who smoked for a lifetime without negative health consequences should be encouraged and supported to quit smoking. Because of demographic changes and the need to work longer for sufficient retirement pensions (up to age 67 years in many developed countries), the individual and public health burden of smoking-related morbidity and mortality at older age will further increase substantially unless major progress is made in reducing the prevalence of smoking at all ages, including old age. Future research should include meta-analyses on the impact of smoking on cause-specific mortalities at older age, as well as evaluation of smoking cessation interventions specifically designed for older people and the impact of smoking cessation during follow-up on health-related outcomes.
In conclusion, smoking is a strong risk factor for premature mortality at older age. A dose-response relationship of the number of currently smoked cigarettes with mortality was observed in all age groups, even though the number of studies reporting such data are still rather limited. The longer the time since smoking cessation, the lower the RM of older former smokers; this fact calls for effective smoking cessation programs that are likely to have major preventive effects even for smokers aged 60 years and older.
Correspondence: Hermann Brenner, MD, MPH, Division of Clinical Epidemiology and Aging Research, German Cancer Research Center, PO Box 10 19 49, Heidelberg 69009, Germany (firstname.lastname@example.org).
Accepted for Publication: March 12, 2012.
Author Contributions: Ms Gellert and Dr Schottker had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of data analysis. Study concept and design: Gellert, Schöttker, and Brenner. Acquisition of data: Gellert and Schöttker. Analysis and interpretation of data: Gellert, Schöttker, and Brenner. Drafting of the manuscript: Gellert. Critical revision of the manuscript for important intellectual content: Schöttker and Brenner. Statistical analysis: Gellert. Obtained funding: Brenner. Administrative, technical, and material support: Schöttker. Study supervision: Schöttker and Brenner.
Financial Disclosure: None reported.
Funding/Support: This analysis was conducted in the context of the CHANCES project funded in the FP7 Framework Programme of DG-RESEARCH in the European Commission. The project is coordinated by the Hellenic Health Foundation, Athens, Greece.