Increased alcohol consumption is linked with cigarette smoking [27] and with stroke [28, 29]. However, only five prospective studies in men have controlled for alcohol consumption [1, 4, 6, 7, 16], and none has evaluated the potential modification of the effect of smoking by the level of alcohol intake. The largest of these five studies followed 7895 men [7]. In this study, we examined the relation of cigarette smoking with stroke in a cohort of 22 071 male physicians participating in the Physician's Health Study.

Methods

Top Methods Results Discussion Author & Article Info References

Study Population

The Physicians' Health Study [30, 31] is a randomized, double-blind, placebo-controlled trial examining the effect of low-dose aspirin on cardiovascular disease and the effect of ß-carotene on cancer and cardiovascular disease. Briefly, 22 071 U.S. male physicians (40 to 84 years of age at entry in 1982) were randomly assigned to aspirin alone, ß-carotene alone, aspirin and ß-carotene, or placebo alone, using a 2 x 2 factorial design. All participants enrolled in the trial were free from self-reported previous myocardial infarction, stroke, and transient ischemic attack.

The participants completed a mailed baseline questionnaire that included information about cigarette smoking status (never, past only, or current) and included the number of cigarettes smoked per day if they were currently smoking. Other information collected at baseline included age, history of angina pectoris, history of coronary revascularization procedures (coronary artery bypass grafting or percutaneous transluminal coronary angioplasty), history of diabetes mellitus, history of hypertension, history of high cholesterol levels, height, weight, history of alcohol use, and frequency of vigorous exercise.

Every 6 months during the first year and then annually, participants completed brief questionnaires inquiring about their compliance with the assigned treatment and about the occurrence of any relevant events, including stroke. On 15 January 1988, the aspirin component of the study was terminated, mainly because of a statistically extreme 44% decrease in the risk for a first myocardial infarction among participants in the aspirin group. By that date, participants had been followed for an average of 60.2 months. This report includes all available data as of 14 October 1992. By this date, participants had been followed for an average of 9.7 years (range, 8.5 to 11.1 years). The completeness of follow-up for stroke-associated morbidity and mortality was 100%.

Ascertainment and Definition of Stroke End Points

Nonfatal strokes were reported on the semiannual or annual questionnaires. Deaths were usually reported by the families or postal authorities, and persistent nonresponders to the questionnaires were telephoned. Strokes were considered confirmed only after medical records and all available information were reviewed by the End Points Committee (which consisted of two internists, a cardiologist, and a neurologist, who did not have knowledge of treatment assignment). Unconfirmed events were not used in any of our analyses. A definite case of nonfatal stroke was defined as a typical neurologic deficit that was sudden or rapid in onset, lasted more than 24 hours, and was attributable to a cerebrovascular event. Strokes were classified according to the probable cause (ischemic or hemorrhagic) on the basis of medical records, computed tomographic scanning, and the judgment of the neurologist. Computed tomographic scans were available for more than 95% of all confirmed cases of stroke. Every case of hemorrhagic stroke was confirmed by either a computed tomographic scan or by lumbar puncture. "Unknown" strokes were those without an imaging study or other clear documentation of stroke subtype. Definite cases of fatal stroke were documented by convincing evidence of a cerebrovascular mechanism from all available sources, including death certificates, hospital records, and—for death outside the hospital—observers' impressions. Only first cases of stroke were counted.

Statistical Analyses

Smoking status was defined as never smoked, formerly smoked, currently smoking less than 20 cigarettes per day, or currently smoking 20 or more cigarettes per day. Follow-up began with questionnaire completion in 1982 to 1983 and ended with the diagnosis of nonfatal or fatal stroke, or other fatal event, or 14 October 1992, whichever came first. Five participants contributed no follow-up time to the analyses because, after randomization, they reported strokes that occurred before randomization.

To compute relative risks adjusted for age and other factors, we used the SAS PHREG procedure [32] to do proportional-hazards regression [33]. The proportional-hazards assumption was confirmed using "log-log" plots [34]. Initial analyses to rule out effect modification by age were done using interaction terms. We then adjusted for age and treatment assignment and then further adjusted for all other covariates, including self-reported hypertension, angina, history of coronary revascularization, diabetes mellitus, high cholesterol level, alcohol consumption, vigorous exercise frequency, and baseline obesity status. These covariates are independent risk indicators for stroke, and all have some association with smoking (except treatment assignment). Adjustment for treatment assignment has been routine in all analyses for the Physicians' Health Study.

Analyses restricted to specific levels of alcohol intake were done to assess the interactive effects of alcohol and smoking. To determine the independent effect of alcohol, we did analyses restricted to never-smokers (49.5% of the cohort). Using interaction terms, we also tested for a treatment-assignment interaction for smoking with aspirin (the ß-carotene component of the trial is still ongoing and is blinded) and an interaction for smoking with hypertension.

We tested for linear trend in relative risk across smoking categories using an ordinal smoking variable, with values ranging from 0 (never smoking) to 3 (currently smoking 20 or more cigarettes per day). We calculated 95% CIs for each relative risk, and all P values are two-tailed [35], with those less than 0.05 declared statistically significant. The population attributable risk percentage for current cigarette smoking in relation to total stroke was calculated as the difference between the stroke incidence rate in the total population and the rate in current nonsmokers divided by the incidence rate in the total population (x 100%) [36].

Results

Top Methods Results Discussion Author & Article Info References

Baseline Distribution of Smoking and Other Exposure Variables

In 1982, 11.0% of the physicians in this cohort were current smokers (7.1% of the participants currently smoked 20 or more cigarettes per day and 3.9% currently smoked less than 20 cigarettes per day), 39.3% formerly smoked, and 49.5% had never smoked. Information on cigarette smoking habit was missing for 67 (0.3%) physicians. With the exception of alcohol consumption (17.6% of never-smokers reported daily drinking compared with 37.5% of current heavy smokers), age and most of the other variables had similar distributions among the four groups Appendix Table 1.

Total Stroke and Subtypes

During 207 579 person-years of follow-up, 312 nonfatal and 28 fatal strokes occurred (275 ischemic, 56 hemorrhagic, and 9 of "unknown cause"). Physicians currently smoking 20 or more cigarettes per day had relative risks (adjusted for age and treatment assignment) for total nonfatal and fatal stroke of 2.52 and 1.24, respectively, and in multivariate analyses controlling for other risk factors, these relative risks were 2.71 and 1.46, respectively (Table 1). Ischemic stroke was more strongly related to current smoking than was hemorrhagic stroke. For total nonfatal stroke and ischemic stroke, we found linear increases in risk across the four cigarette smoking categories (P for trend, < 0.0001).

Aspirin treatment assignment was associated with an increased risk for hemorrhagic stroke. In multivariate models, the relative risks associated with aspirin assignment were as follows: total stroke, 1.10 (95% CI, 0.88 to 1.39); ischemic stroke, 0.99 (CI, 0.76 to 1.27); and hemorrhagic stroke, 1.91 (CI, 1.08 to 3.38).

Testing for Interaction

We did not find any significant statistical interaction between aspirin treatment and smoking or between hypertensive history and smoking. Data on the effect of smoking within categories of alcohol consumption Figure 1 were too sparse to permit firm conclusions. In multivariate analyses restricted to never-smokers, we found that the relative risks for total stroke associated with consuming alcohol rarely or never, monthly, weekly, and daily were 1.00 (referent); 0.95 (CI, 0.56 to 1.61); 0.75 (CI, 0.47 to 1.22); and 1.14 (CI, 0.63 to 2.05). Analyses excluding 333 physicians with a history of angina pectoris, of coronary artery bypass grafting, or of percutaneous transluminal coronary angioplasty at baseline showed essentially the same results as those shown in Table 1 and Figure 1.

View larger version (15K): [in this window] [in a new window]   Figure 1. Relative risk for total stroke associated with cigarette smoking by level of alcohol consumption. No strokes were noted among physicians currently smoking fewer than 20 cigarettes per day and rarely or never consuming alcohol ("rarely/never" denotes consuming alcohol less than monthly). Relative risks were adjusted for age (1-year categories), treatment assignment (aspirin alone, ß-carotene alone, aspirin and ß-carotene, or placebo), and history of angina or coronary revascularization.

Discussion

Top Methods Results Discussion Author & Article Info References

These findings provide further evidence that current cigarette smoking is an independent risk factor for ischemic and hemorrhagic stroke in men. We observed a linear increase in risk for stroke across the four cigarette smoking categories; current smoking of 20 or more cigarettes per day was associated with 2.52 to 2.71 times the risk for nonfatal stroke compared with never smoking. Cigarette smoking can increase the risk for stroke through several mechanisms. Smoking is associated with increased levels of fibrinogen and other clotting factors and has been shown to increase platelet aggregability [37-39]. Smoking may also decrease the level of high-density lipoprotein cholesterol [40] and may increase the hematocrit [41]. These changes are associated with an increased risk for stroke [42, 43]. Additionally, experimental evidence indicates that smoking can cause direct endothelial damage, an important component in the atherosclerotic process [44]. Using data from a wide variety of patients, 29 previous studies (13 prospective and 16 retrospective) have found a statistically significant relation between cigarette smoking and stroke [45]. These findings and ours, as well as several plausible biological mechanisms, argue in favor of a causal relation between smoking and stroke.

The large sample size makes this the largest study of smoking and stroke in men that controlled for potentially confounding stroke risk factors, including alcohol intake. Although the report by Iso and colleagues [2] from the Multiple Risk Factor Intervention Trial included information on 350 977 men, only a univariate analysis of smoking and stroke risk was done [2]. Similarly, several large prospective studies [12, 13, 25] in men either did not control for possible confounding risk factors or controlled only for age. Although coronary heart disease is an important risk factor for stroke [43], only one other prospective study [3] of smoking and stroke in men controlled for possible confounding by preexisting coronary heart disease. Two other prospective studies [6, 7] in men controlled for left ventricular hypertrophy, another marker of impaired cardiac function.

All cerebrovascular events during the study period were thoroughly investigated, and only those end points that met the study definition of stroke were used in the analyses reported here. The quality of end-point confirmation has varied widely among previous studies of smoking and stroke in men. Moreover, the events qualifying as end points have varied; for example, the study by Wolf and colleagues [3] included transient ischemic events in some of their analyses, and some studies [4, 15] excluded all cases of subarachnoid hemorrhage from analysis.

Recently, Kawachi and colleagues [26] reported relative risks of 2.58 (CI, 2.08 to 3.19) for total stroke, and Colditz and colleagues [46] reported risks of 2.25 for ischemic stroke (CI, 1.72 to 2.95) among nurses currently smoking. However, the strongest smoking-associated increase in risk among female nurses was for hemorrhagic stroke: Current smoking was associated with a relative risk of 4.96 (CI, 3.13 to 7.87). Male physicians who were current smokers had larger relative risks for ischemic stroke than for hemorrhagic stroke. However, the number of hemorrhagic events in the physicians was much smaller than in the nurses (54 compared with 161), giving our study a low power to detect an effect of smoking on the risk for hemorrhagic stroke.

In the Nurses' Health Study [47], moderate alcohol intake was associated with a decrease in risk for ischemic stroke but an increase in risk for subarachnoid hemorrhage. In another study [29] of men and women, light drinking was associated with a decreased risk for total stroke (ischemic and hemorrhagic combined). Our findings for the combined effects of alcohol and smoking are based on too few events to be conclusive.

One limitation of our study was that the Physicians' Health Study was not designed to investigate the relation between smoking and stroke. Thus, the data collected for cigarette smoking and alcohol use were not detailed. For example, at baseline we did not collect data on duration of smoking and duration since quitting. Using baseline data may have resulted in misclassifying former smokers (who quit during follow-up) as current smokers; this would tend to bias results for current smokers toward null findings. Additionally, if a substantial proportion of former smokers had smoked for only brief periods or had quit many years before study entry, this could also bias our estimate of the relative risk among former smokers toward unity.

The self-reported smoking status of the physicians was not confirmed by any biochemical tests. However, we believe that the self-reported smoking information supplied by the physicians is valid. A previous report from the Physicians' Health Study [48] described results of a validation substudy of self-reported angina pectoris. Among 90 physicians reporting this diagnosis, medical records confirmed the self-reported information in 88 (97.8%). In a study [49] of male health professionals (dentists, veterinarians, pharmacists, optometrists, osteopathic physicians, and podiatrists), the correlation between self-reported and technician-measured weight was 0.97.

Our data suggest that smoking is strongly related to risk for stroke. Assuming causality, approximately 6% of the incidence of stroke among physicians was attributable to current cigarette smoking. However, had the proportion of physicians currently smoking been the same as in the U.S. population (25.7%) [50], approximately 15% of stroke incidence would be attributable to smoking. In the United States, where more than 500 000 nonfatal strokes and 145 000 fatal strokes occur annually [51], smoking may account for a substantial amount of stroke-associated morbidity and mortality.