Cigarette Smoking and Mortality Risk:  Twenty-five–Year Follow-up of the Seven Countries Study FREE

DESPITE WIDESPREAD agreement that cigarette smoking is harmful for health,^1- 12 some observations in the Seven Countries Study have raised the question whether this is universal. Keys' monograph¹³ points out that cigarette smoking was much less predictive of coronary heart disease (CHD) and all-causes death in 10 years of follow-up in southern Europe and Japan than in the United States and northern Europe. A more recent analysis of 25-year follow-up data reported that number of cigarettes smoked per day was significantly related to death from stroke both in pooled data and in 3 cohorts, but was statistically significantly inversely related to death from stroke in 1 Greek cohort.¹⁴ Besides area of residence,¹³ it has been suggested that level of lifetime exposure to cigarette smoking, the manner of smoking, the type of cigarette smoked, and the typical CHD rate and serum cholesterol level or blood pressure level in a culture might modify the effects of smoking.^13,15- 16 On the other hand, chance may explain the few findings in which never smokers had a higher cause-specific death rate than smokers. First, some of the Seven Countries Study areas are small, and never smokers were a relative rarity in the early 1960s when the baseline data were collected. Second, no specific hypothesis exists that smoking is unrelated or inversely related to diseases such as cardiovascular disease or cancer in specific areas; therefore, statistical significance in country-specific analyses should be computed with adjustment for multiple comparisons. Third, data are often more variable in studies that pool data across diverse centers than in studies from a single center.

Thus, although most observations in the Seven Countries Study and elsewhere suggest that cigarette smoking is harmful for health, it remains important to assess whether this finding is universal. In this article, a comprehensive investigation of all-causes and cause-specific death in the 16 cohorts of the Seven Countries Study was undertaken with the hypothesis that smoking would be consistently associated with several causes of death. We specifically considered the possibility that deviations from this hypothesis might be explained by chance.

The Seven Countries Study is a longitudinal observational study of risk factors for coronary heart disease in 16 cohorts situated in 7 countries in Europe, the United States, and Japan. The baseline examination was performed between 1957 and 1964 on 12,763 men aged 40 to 59 years in 16 cohorts in 7 countries; that is, 1 cohort in the United States, 2 in Finland, 1 in the Netherlands, 3 in Italy, 2 in Croatia, and 3 in Serbia (former Yugoslavia), 2 in Greece, and 2 in Japan. Details on the samples are given elsewhere.¹⁷

Information about smoking habits was collected using a standardized questionnaire developed for the Seven Countries Study. The questions covered cigarette smoking, pipe smoking, cigar smoking, and former smoking. For this analysis, daily cigarette consumption at the entry examination was expressed as never smoker, current smoking of 1 to 4, 5 to 9, 10 to 19, 20 to 29, or 30 or more cigarettes per day or less than 1, 1 to 9, or 10 years since smoking cessation. In Japan, former smokers were not asked how many years ago they had stopped smoking, and all former Japanese smokers are treated as if they quit smoking 1 to 9 years before baseline. In the United States, information on cigar and pipe smoking was missing. In analyses stratifying on country and smoking status, we eliminated the most recent former smokers, because they were few in number and most likely to have returned to smoking after baseline; pooled the remaining former smokers of 1-year duration or longer; pooled the 2 categories of 1 to 4 and 5 to 9 cigarettes per day; and pooled the 3 categories of current smokers of 10 cigarettes per day or more.

The following data obtained on all men at entry examination have been considered as covariates: country of residence; age, rounded to the nearest birthday; body mass index (BMI, a measure of weight in kilograms divided by the square of the height in meters); serum cholesterol level measured in milligrams per deciliter (nonfasting measurement using the Abel-Kendall method as modified by Anderson and Keys¹⁸); and systolic blood pressure, measured with a mercury sphygmomanometer. The clinical cardiovascular diseases at baseline were assessed by physical examination and history and included angina pectoris, myocardial infarction, CHD manifested as arrhythmias or failure without obvious cause, but likely coronary, rheumatic heart disease, heart disease of unknown origin, heart disease of other cause, peripheral arteriosclerotic disease including medical history of claudication, and cerebral arteriosclerotic disease including medical history of stroke.

The study was fully explained to all participants. Appropriate institutional review boards approved the project with informed, but unwritten, consent, as was the custom in the early 1960s.

Collection of data on vital status and causes of death over the next 25 years was based on review of death certificates and collection of medical information from hospitals, clinical records, interviews of physicians and relatives of the deceased, or any other witnesses of fatal events. After the 10-year follow-up in the US Railroad, the 15-year follow-up in Finland, and irregularly in other areas, only review and coding of official death certificates were performed. All these data enabled central reviewers (H.B. and A.M.) to assign the final underlying cause of death, following standard criteria and using the World Health Organization's International Classification of Diseases, Eighth Revision.¹⁹ In case of coexisting causes of death, violence, cancer in advanced stages, and CHD took precedence over stroke and other causes in the hierarchy that assigned underlying cause of death. After 25 years the vital status of all but 56 men was known.^14,20- 21

Death rates (per 1000) from all-causes and cause-specific death in each cohort were computed with adjustment for age using linear regression models. Age and cohort-adjusted death rates of participants according to 9 categories of smoking habit were also modeled, and relative risks vs never smokers computed.

Within each cohort, we computed age-adjusted absolute and relative risks for all-causes and cause-specific mortality for smokers of 10 cigarettes per day or more vs never smokers. Because there was no specific hypothesis that smoking would be differentially related to any given cause of death in any cohort, we applied the Bonferroni rule when comparing smoking-mortality relationships across the 16 cohorts, by dividing nominal statistical significance levels by 16 (equivalent to requiring a nominal P<.003125). The Bonferroni rule corrects for the fact that, with no a priori hypothesis, large differences were noted only because, for each cause of death, they were the most extreme inverse findings among 16 cohort-specific comparisons of heavy vs never smokers. We also examined consistency of findings by noting in this table the number of instances of 16 in which death rate in heavy smokers was greater than the corresponding death rate in never smokers. Each of the 16 differences in death rate between never smokers and smokers of 10 cigarettes per day or more is an estimate of the association of smoking and mortality, and is approximately normally distributed. Therefore, the probability that a difference is higher than 0 is .5.

Finally, hazard ratios compared with never smokers for former, light, and heavy smokers, in one model, and for former smokers and a linear term for cigarettes per day, in another model, were computed using the Cox proportional hazards model, with all-causes and cause-specific mortality in 25 years as the end points and country of residence plus 5 risk factors (age, BMI, serum cholesterol level, systolic blood pressure, and the presence of clinical cardiovascular diseases at baseline) as covariates. These models omitted those who quit smoking within the past year. A possible interaction effect of smoking with BMI, serum cholesterol level, or systolic blood pressure was tested by adding interaction terms to the model and testing (by means of an F statistic) whether the explained variance was significantly increased. All analyses were performed using the SAS statistical software.²²

Table 1 shows the size of the cohorts and the all-causes death rates per 1000 in 25 years. Cumulative death rates through average age 75 years ranged from about 30% in Belgrade, Serbia, and Crete, Greece, to about 60% in Slavonia, Croatia, East Finland, and Zrenjanin, Serbia. Men examined at entry represented, on average, 90.4% of those invited to participate.¹⁷ Large differences in CHD death rates were found among these countries, with the Northern European (Finland and the Netherlands) and Northern American (United States) samples highest, compared with the Southern European (Italy, Croatia, Serbia, and Greece) and Japanese cohorts. The same tendency was seen for deaths from lung cancer. On the other hand, rates of death from stroke in Northern Europe and the United States were lower than in Southern Europe and Japan; cancer other than lung was highest in Japan; and the highest rates of "other" diseases were found in Croatia and Japan (data not shown).

Table 2 gives age and cohort-adjusted all-cause and cause-specific death rates of participants according to detailed cigarette smoking status. Rates of death from total, CHD, lung cancer, other cancer, and chronic obstructive pulmonary disease (COPD) were much higher in smokers than in never smokers. Dose response with number of cigarettes smoked per day among current smokers was clearest in total death and death from lung cancer, but present with variation also in the cases of CHD, other cancer, and death from COPD. In smokers, there were smaller increases in death rates attributed to other arterial disease and infection. There were no clear differences in rates across current smoking categories for death due to other heart disease, stroke, accidents, and other diseases. The smokers of 30 cigarettes per day or more had a 21% elevation in 25-year total death rate compared with never smokers (57.7% vs 36.3%). Clear dose-response relationships were found across the 3 former smoker categories for all-causes, other heart disease, lung cancer, and COPD mortality. The 625 men who had quit smoking 10 years or more before baseline had a total death rate of 2 percentage points higher than that for never smokers, while recent quitters had a death rate equivalent to that of current smokers of 5 to 9 cigarettes per day, 9 percentage points greater than the rate in never smokers.

Because there were in some cases large differences in total mortality between cohorts within the same country (Table 1), we analyzed smoking data separately for each of the 16 cohorts. Figure 1 shows higher age-adjusted death rates in smokers of 10 cigarettes per day or more than in never smokers in all 16 cohorts for all causes and other cancers, in 15 cohorts for lung cancer and COPD, in 14 cohorts for infection, in 13 cohorts for other arterial disease, in 12 cohorts for CHD, in 11 cohorts for other deaths, in 9 cohorts for stroke and other heart disease, and in 7 cohorts for accidents. (Note that scales for death rate differences vary in Figure 1, to highlight differences between cohorts for each cause of death.) Absolute excess risk of all-cause death in heavy smokers compared with never smokers ranged from a low of 6.2% in Ushibuka, Japan, to a high of 30.4% in Zrenjanin. Notable exceptions to expected excess risk associated with heavy smoking were infections in Crevalcore, Italy, stroke in Corfu, Greece, and CHD death in Zutphen, Holland. The first 2 did not achieve even nominal statistical significance. We note that stroke was also inversely associated with cigarette smoking in Crete, the other Greek cohort. The P value for the inverse association of stroke with heavy smoking in the combined Greek cohorts was .02. The inverse association of death from CHD with smoking in Zutphen was based on only 64 never smokers but achieved nominal statistical significance (P = .03). After the Bonferroni correction for multiple comparisons, both nominally statistically significant comparisons became nonsignificant: corrected P = .14 for stroke in Greece (based on 8-country comparisons, treating Croatia and Serbia as separate countries) and P = .36 for CHD in the Netherlands (based on 16 cohort comparisons).

As a summary, we computed adjusted hazard ratios and significance probabilities for 3 categories of former and current smokers vs never smokers, adjusting for area of residence and other covariates using the Cox proportional hazards model. In Table 3, hazard ratios per pack of cigarettes smoked per day in descending order were those for lung cancer, other arterial disease, COPD, total death, and CHD. Associations were small and not statistically significant for infections, accidents, and other heart disease. Former smoking (excluding recent quitters) was not associated with statistically significantly increased risk for any cause of death.

Under the hypothesis that risk associated with smoking was greatest when the men also had high levels of other cardiovascular disease risk factors, we tested for interactions of all-cause mortality and smoking with cholesterol level, systolic blood pressure, and BMI. This hypothesis was not borne out. None of these interaction coefficients achieved statistical significance while the general pattern persisted of greatly increased all-cause death risk in heavy smokers compared with never smokers (data not shown).

Because changes in smoking habit may influence risk, we categorized smoking changes, comparing the greater smoking habit reported at the 5- and 10-year examinations (year 10 only in Japan, year 5 only in the US Railroad) to the habit reported at baseline. Of 10,494 men who survived through the first 10 years of study, 567 started smoking and 568 quit smoking for more than 1 year during follow-up. Mortality patterns for never smokers, former smokers, and continuing smokers during the 15 years of subsequent follow-up were qualitatively similar to those reported herein for 25 years of follow-up in baseline never smokers, former smokers, and current smokers (data not shown).

We found consistently elevated death rates in smokers across all countries for several causes. For all causes, risk increased in dose-response increments starting from never smokers to 1 to 4, 5 to 9, 10 to 19, 20 to 29, and more than 30 cigarettes per day; former smokers of 10 years' duration or more had only slightly elevated risk, increasing to the level of 1 to 4 cigarettes per day for those who quit smoking 1 to 9 years before baseline, and to the level of 5 to 9 cigarettes per day for baseline recent quitters, who are most likely to have resumed smoking. The following causes of death were apparently unrelated to smoking: other heart disease, infection, and accidents.

Total mortality, which is not subject to diagnostic classification bias, was positively related to smoking in all 16 cohorts. As with any other dependent variable observed in samples, the relative death rate between smoking categories is expected to vary between cohorts by chance alone. Under a null hypothesis of no difference, the chance that total mortality would be greater in heavy smokers than in never smokers in any given cohort is 0.5; under the approximation of equal precision in each cohort, the chance that all 16 differences are in the same direction is 0.00003. Another way to think about among-cohort variation is to consider the confidence interval around the 16.9% mortality difference in pooled countries (mean age- and cohort-adjusted total mortality rate in smokers of ≥10 cigarettes per day, 53.3%; in never smokers, 36.4%). Considering the actual numbers of never smokers and smokers of 10 cigarettes per day or more, and assuming a binomial distribution, the 95% confidence interval around 16.9% includes the observed difference in all but 4 cohorts: the observed total death rate difference was slightly below the lower confidence bound in Belgrade; and in East Finland, the US Railroad, and Zrenjanin, the observed differences were slightly above the upper confidence bound. The observed differences in the other 12 cohorts fall within their respective confidence intervals; this pattern is suggestive of a small variance component associated with country, analogous to that often seen in meta-analyses of several studies.²³ Although the Seven Countries Study used a common protocol, small differences may have existed between countries in such factors as the participants' understanding of the questions asked and in diagnostic customs pertaining to death certification.

For cause-specific mortality within countries, in which heavy smokers had lower risk than never smokers, notable was the risk of stroke in Greece, which occurred in both Corfu and Crete, and achieved nominal statistical significance when the cohorts were pooled. However, this association became nonsignificant after adjustment for multiple comparisons.

AS OBSERVED in this study, it is well known that cigarette smoking is associated with increased risk of CHD,⁵ lung cancer,^6- 8 other cancers,^24- 25 and COPD.^1,25 Besides lung cancer, our data show increased risk with smoking for the pool of other cancers. It is established that smoking is causally associated with cancers in 6 other sites (oral cavity, esophagus, larynx, bladder, pancreas, and kidney) in men.²⁴

In the Seven Countries Study, pooled analyses of cigarette smoking and stroke suggest a weakly increased risk in heavier smokers. Relationships between cigarette smoking and stroke are controversial.^{9- 12,26- 27} This inconsistency, both in the literature and in our study, may be in part due to differences in the association of cigarette smoking with the subtype of stroke. Shinton and Beevers²⁶ reported that considerable difference was seen in relative risks among the subtypes: cerebral infarction, 1.9; cerebral hemorrhage, 1.0; and subarachnoid hemorrhage, 2.9. Kawachi et al²⁸ reported similar findings in women aged 30 to 55 years, followed up for 12 years. We were not able to differentiate type of stroke in this study because most deaths occurred before diagnoses were routinely based on computed tomography.²⁹ Another possible factor is that stroke tends to occur at older ages (in this study, average age at death was 69 years for stroke, compared with 67 years for CHD and 66 years for lung cancer), and men may have quit smoking years before having their stroke, for example because of other disease. Several years of nonsmoking would tend to attenuate the association of smoking and stroke death. Several studies^30- 32 report much stronger relative risk of smoking for stroke in younger ages. Associations of smoking with death from stroke may also be attenuated, compared with the situation with incidence of stroke, because many survivors of stroke die of nonstroke causes. Highly variable results are expected in individual cohorts when there is a small observable excess risk for death from stroke and variability in circumstances of death certification. Despite the positive association seen in the pooled participant experience, only 9 of the 16 cohorts showed a greater risk of stroke among heavier smokers.

The findings for infectious disease death also vary, depending on the mode of statistical analysis. It is reasonable that smoking might be positively associated with infection, for example, because cigarette smoking may diminish antioxidant reserve.³³ A nonstatistically significant 50% to 60% excess risk was seen for all smokers in the proportional hazards models, but it was not graded with number of cigarettes smoked. The pooled difference between smokers of 10 cigarettes per day or more and never smokers was significant in linear regression, however; and in 14 of the 16 cohorts, smokers of 10 cigarettes per day or more had a higher infectious disease death rate than never smokers (P = .004 under the approximation of equal precision in each cohort). On closer inspection, there is substantial heterogeneity between cohorts: Slavonia has a death rate difference much larger than any of the other cohorts, while Crevalcore and Montegiorgio, Italy, have differences much lower and several cohorts have differences close to zero. Infectious disease as the underlying cause of death is a highly variable phenomenon, as most deaths with infectious diseases are coded to other causes. The association of smoking with infectious disease should be further studied in other settings.

Our results also demonstrated the favorable effects of quitting smoking: after 10 or more years of smoking cessation, risk of total death is attenuated to almost the same level as that of never smokers. Several articles^34- 39 deal with smoking cessation and decreased risks. Kawachi et al^34- 35 suggested that health benefits of cessation were clearly present regardless of the age of smoking initiation, the age of smoking cessation, and daily number of cigarettes smoked among current smokers. They also concluded that former smokers had a 24% reduction in risk for cardiovascular disease mortality within 2 years of quitting compared with continuing smokers, and that the excess risk for total mortality and both cardiovascular disease and total cancer mortality among former smokers approached the level of that for never smokers after 10 to 14 years of abstinence. However, Cook et al³⁶ suggested that the major benefit of giving up smoking may lie in accumulating fewer total pack-years of smoking. One example of an effect of smoking that endures for long after smoking cessation is elevated fibrinogen concentration. Dobson et al³⁷ found that fibrinogen concentrations were higher in current smokers and former smokers up to 2 years after quitting than in nonsmokers and after that time were similar to levels in nonsmokers.

Some have suggested that the variation between countries in the association of smoking with cause-specific mortality is based on the manner of smoking,¹³ the type of cigarette smoked,^15- 16 or the relation of smoking and health to other factors such as serum cholesterol level. Although there may be small differences between countries, we suggest that these variations are more likely explained by chance. Specifically, in the context of a large study such as the Seven Countries Study, it is preferable to make interpretations in light of the entire body of evidence, rather than emphasizing discordant subgroup (cohort) observations that were not hypothesized a priori.

The attributable risk for total death, pooled across countries, is striking. If none of the participants had ever smoked, the experience of the never smokers suggests that 64% of men aged 40 to 59 years in the early 1960s would still be alive after 25 years, at ages 65 to 84 years. On the other hand, if everyone had been a pack-a-day smoker, only about 47% would have been alive after 25 years. Attributable risk, given the actual smoking patterns and death rates of these men, was 23%.

In conclusion, this investigation was motivated by the observation that smoking was inversely related to selected causes of death in some areas of the Seven Countries Study, particularly stroke in the 2 Greek cohorts. This observation raised the question whether, under certain circumstances, an inverse association between smoking and mortality risk might be present. A rigorous examination of the data by cause of death and by geographic area finds overwhelming evidence that smoking is related to adverse outcomes in every Seven Countries Study area. It strongly suggests that the few observations in which smoking is inversely related to mortality attributed to CHD, lung cancer, and COPD are chance aberrations, related to small numbers, failure to account for multiple comparisons, differences in death certification between countries, and imprecision in data collected in multiple research settings. It was concluded that the few exceptions in which smoking was not found related to worse outcomes was possibly attributable to chance. Therefore, this investigation supports the view that cigarette smoking is bad for your health, independent of country of residence or the type of cigarette smoked.

Despite the decline in prevalence of cigarette smoking in the United States and some other developed countries,^40- 41 mortality from several smoking-related causes continues to increase.^25,42 Smoking prevalence has been increasing in teenagers in the United States⁴³ and in whole populations in developing countries.⁴⁴ This investigation in a wide variety of cultures strongly supports continuing efforts toward smoking cessation and the prevention of smoking initiation in youth.

Accepted for publication August 23, 1998.

Reprints: David R. Jacobs, Jr, PhD, Division of Epidemiology, School of Public Health, University of Minnesota, 1300 S Second St, Suite 300, Minneapolis, MN 55454 (e-mail: jacobs@epivax.epi.umn.edu).