Cook NR, Hebert PR, Manson JE, Buring JE, Hennekens CH. Self-Selected Posttrial Aspirin Use and Subsequent Cardiovascular Disease and Mortality in the Physicians' Health Study. Arch Intern Med. 2000;160(7):921-928. doi:10.1001/archinte.160.7.921
The randomized aspirin component of the Physicians' Health Study (PHS) was terminated early, after 5 years, primarily because of the emergence of a statistically extreme (P<.00001) 44% reduction of first myocardial infarction (MI) among those assigned to aspirin. As a result, there were insufficient numbers of strokes or cardiovascular disease (CVD)–related deaths to evaluate these end points definitively.
Data on self-selected aspirin use were collected until the beta carotene component ended as scheduled after 12 years. Posttrial use of aspirin was assessed at the 7-year follow-up among 18 496 participants with no previous reported CVD. Randomized and posttrial observational results in the PHS were compared, and differences between those self-selecting aspirin and those not were examined.
At 7 years, 59.5% of participants without CVD reported self-selected aspirin use for at least 180 d/y, and 20.8% for 0 to 13 d/y. Use was significantly associated with family history of MI, hypertension, elevated cholesterol levels, body mass index, alcohol consumption, exercise, and use of vitamin E supplements. In multivariate analyses, self-selected aspirin use for at least 180 vs 0 to 13 d/y was associated with lower risk for subsequent MI (relative risk [RR], 0.72; 95% confidence interval [CI], 0.55-0.95), no relation with stroke (RR, 1.02; 95% CI, 0.74-1.39), and significant reductions in CVD-related (RR, 0.65; CI, 0.47-0.89) and total mortality (RR, 0.64; CI, 0.54-0.77).
These associations between self-selected aspirin use and CVD risk factors increase the likelihood of residual confounding and emphasize the need for large-scale randomized trials, such as the ongoing Women's Health Study, to detect reliably the most plausible small to moderate effects of aspirin in the primary prevention of stroke and CVD-related death.
LOW-DOSE ASPIRIN has demonstrated clear net benefits in randomized trials of secondary prevention of cardiovascular disease (CVD), including a wide range of previous occlusive diseases as well as acute evolving myocardial infarction (MI).1 Evidence from primary prevention trials supports a clear reduction in first MI among men, but the balance of benefits vs risks for stroke and CVD-related mortality has not yet been evaluated definitively. Data from observational studies on these questions of small to moderate effects have the benefits of larger numbers of end points and longer duration, but also have the inherent limitation that their results may result at least in part from confounding by unmeasured, unmeasurable, or unknown risk factors. The Physicians' Health Study (PHS),2 a randomized trial of low-dose aspirin in the primary prevention of CVD, offered a unique opportunity to compare the randomized and posttrial observational results and to examine whether and the ways in which those who self-selected to use aspirin after the end of the randomized aspirin component differed from those who did not.
The early impact on first MI found in the PHS would be expected to translate into a reduced risk for CVD-related mortality,3 but this has yet to be demonstrated in primary prevention trials, including the British Doctors' Trial4 and the Thrombosis Prevention Trial,5 primarily because of inadequate numbers of deaths. The impact of aspirin on stroke also remains unclear. Overviews of secondary prevention trials indicate that aspirin produces clear and consistent reductions in stroke,6 but in primary prevention there have been insufficient numbers of strokes to evaluate this end point definitively. In the PHS, a possible but nonsignificant observed increase in strokes resulted primarily from an apparent excess of a small number of hemorrhagic strokes in the aspirin group (23 vs 12; 95% confidence interval [CI], 0.96-4.77).2
Observational studies of these questions have, not surprisingly, been inconsistent.7- 12 In all observational studies, confounding by indication, such as a possible increased (or decreased) self-selection of aspirin use among those at higher (or lesser) risk for CVD, is difficult, if not impossible, to quantify. Our report evaluates randomized and observational posttrial data from the PHS and examines the relation of self-selection of aspirin use to risk factors for CVD as well as to subsequent morbidity and mortality after the early termination of the aspirin component of the PHS.
The PHS was a randomized, double-blind, placebo-controlled 2 × 2 factorial trial of aspirin and beta carotene in 22,071 US male physicians. The methods and results have been described in detail previously.2,13 Beginning in 1982, participants, aged 40 to 84 years, with no history of MI, stroke, transient cerebral ischemia, or cancer (excluding nonmelanoma skin cancer), were randomized to one of the 4 treatment arms. Participants received monthly calendar packs containing 325 mg of aspirin (Bufferin; provided by Bristol-Myers, Princeton, NJ) with placebo on alternate days, 50 mg of beta carotene (Lurotin; provided by BASF Corporation, Ludwigshafen, Germany) with placebo on alternate days, both active drugs, or both placebos. Eligible participants had no contraindications to aspirin use and were not regularly taking aspirin, other platelet-active medications, or supplements of beta carotene (vitamin A). Written informed consent was obtained from all study participants, and the research protocol was reviewed and approved by the institutional review board at the Brigham and Women's Hospital, Boston, Mass.
On January 25, 1988, the aspirin component of the trial was terminated early, after an average of 60.2 months of follow-up.2 The reported consumption of aspirin or other platelet-active drugs was 85.7% in the aspirin group and 14.2% in the placebo group at this time. After this date, the beta carotene component of the trial continued uninterrupted, but all physicians were asked whether they preferred active aspirin or placebo to be included in their calendar packs.
The beta carotene component of the trial continued until its scheduled end on December 31, 1995.13 At this time, follow-up information had accrued for an average of 12 years. At the end of 11 years of follow-up (the last year completed by all participants), 99.7% of participants were providing morbidity information, and mortality information was complete for all but 1 of the 22 071 participants. At that time, 78.7% of participants were still taking beta carotene or its placebo.
Throughout the entire follow-up period, written informed consent was requested to review the participant's medical records when a relevant end point was reported. When necessary, details were requested from hospitals and treating physicians. Reports of CVD or cause of death were considered confirmed or refuted only after the examination of all available information by an end-points committee consisting of 2 internists, a cardiologist, and a neurologist, all of whom were unaware of treatment assignments and aspirin exposure. When written consent or relevant records could not be obtained, a reported event was not classified as confirmed but remained unrefuted.
Diagnoses of nonfatal MI were confirmed using World Health Organization criteria.14 Nonfatal stroke was defined as a focal neurologic defect, sudden or rapid in onset, that lasted more than 24 hours and was attributed to a cerebrovascular event. Death due to a cardiovascular cause was confirmed by convincing evidence from available sources, including death certificates, hospital records, and (for death outside the hospital) observers' accounts.
By the 7-year follow-up questionnaire, all participants had entered the posttrial period (past January 25, 1988), and self-selected aspirin use at this time was the primary exposure in these analyses. On this and subsequent questionnaires, participants were asked, during the past 12 months, how many days they had taken the white pills from their calendar packs, with possible response categories of 0, 1 to 13, 14 to 30, 31 to 60, 61 to 90, 91 to 120, 121 to 180, and more than 180 days. Participants were also asked on how many days they had taken additional aspirin or medication containing aspirin, using the same response categories. Total aspirin use was estimated from the white pill count and reported outside use. Responses were collapsed into the following 4 categories for analysis: 0 to 13, 14 to 120, 121 to 179, and at least 180 days of self-selected aspirin use in the past year.
Analyses of randomized aspirin use have been published previously and are based on person-years from time of randomization to the time of first CVD event or termination of the randomized aspirin component.2 Analyses of self-selected aspirin use at the 7-year follow-up considered events occurring during the observational posttrial period from 7 years through database closure on October 24, 1995, an approximate 5-year follow-up period. All analyses of the posttrial period excluded participants with any previous report of CVD, including MI or other ischemic heart disease, stroke, transient ischemic attack or other cerebrovascular disease; atrial fibrillation; coronary artery bypass graft; percutaneous transluminal coronary angioplasty; carotid artery surgery; or angina. A total of 18 496 participants remained available for analysis. Analyses of cancer or total mortality also excluded those with a diagnosis of cancer before the 7-year follow-up (n=595).
Known cardiovascular risk factors were considered as correlates of self-selected aspirin use. These included age, smoking, body mass index (BMI; calculated as weight in kilograms divided by the square of height in meters), hypertension, cholesterol level and treatment, diabetes, previous pulmonary embolism or deep vein thrombosis, intermittent claudication, migraine, family history of MI, use of warfarin sodium (Coumadin) or heparin, exercise, use of alcohol, and use of supplements of vitamin E. All of these variables were assessed at the 7-year follow-up, except for BMI and family history of MI (assessed at baseline), smoking (assessed at 5 years), and exercise (assessed at 3 years). Crude means and proportions of these variables were computed for the 4 categories of self-assessed aspirin use. Multivariate odds ratios were assessed using logistic regression with the 4-category aspirin variable using the CATMOD procedure of SAS software (SAS Institute, Cary, NC), comparing each aspirin group to the 0- to 13-d/y category. To test for trend across these aspirin categories, cumulative logistic regression with the LOGISTIC procedure of SAS software was used.
The association between self-selected aspirin use and subsequent CVD or mortality in the posttrial period was assessed using Cox regression models. Trend in risk across the 4 categories was tested with an ordinal variable. All analyses presented also control for the cardiovascular risk factors described above, as well as for randomized aspirin assignment. Interactions of self-selected use with randomized assignment and age were also examined.
During the 5-year period of the randomized aspirin component of the PHS, there were 378 MIs, 217 strokes (173 ischemic and 35 hemorrhagic), 164 CVD-related deaths, and 444 total deaths (Table 1), as reported previously.2 At this time, there was a highly significant (P<.00001) 44% reduction in first MI among those assigned to active aspirin. There was also a possible but nonsignificant 22% increase in total stroke, which was largely confined to a possible but nonsignificant increase in hemorrhagic stroke. The death rate due to CVD was much lower than originally anticipated,3 so the 95% confidence interval for this end point was wide. Using a combined end point of MI, stroke, or CVD-related death, there was a significant 18% reduction due to aspirin use (relative risk [RR], 0.82; 95% CI, 0.70-0.96).2 The total number of deaths was similar in both groups.
After the randomized trial period, participants were asked whether they wanted the white pill in their calendar packs to contain active aspirin or placebo. At the 7-year follow-up, nearly 86.6% requested active aspirin (98.6% of those in the randomized aspirin group, and 74.6% in the placebo group). Actual use as assessed by study questionnaire, however, was lower. Of the 18 496 participants with no previous reported CVD, 11 010 (59.5%) reported taking aspirin at least 180 days during the past year; 2136 (11.6%), 121 to 179 days; 1501 (8.1%), 14 to 120 days; and 3849 (20.8%), 0 to 13 days. Use was higher among those who had been randomized to active aspirin, with 65.5% reporting use of at least 180 days, compared with 53.5% in the placebo group. These proportions remained relatively stable through the remainder of follow-up. This compares with use in the randomized period of 65.3% taking aspirin at least 180 days at the 5-year follow-up in the active aspirin group (86.0% taking it ≥90 days) and 6.8% in the placebo group taking aspirin at least 180 days.
We compared characteristics according to category of self-selected posttrial aspirin use at the 7-year follow-up, excluding those in whom CVD had developed before 7 years (Table 2). All reported odds ratios and P values control for all other characteristics considered. Several statistically significant differences between groups emerged. First, posttrial aspirin use differed by previous randomized assignment. Of those who chose to take aspirin at least 180 d/y, 55.1% had been randomized to active aspirin, compared with 37.8% among those with no or little use of aspirin. There were slight differences in age and BMI at baseline, with frequent users at 7 years being slightly older and heavier. Larger differences were apparent in the proportions with family history of MI, with frequent users of aspirin more likely to have such a history. There was significantly less use of warfarin or heparin, although 8 frequent users of aspirin also reported use of these anticoagulants. There were no significant differences in previous reports of diabetes or migraine, although those with migraine reported slightly less subsequent use of aspirin. There was also no difference in previous report of headache (data not shown).
There were significant differences in treatment of hypertension and high cholesterol level, as well as differences in cholesterol levels, among the self-selected aspirin groups. Treatment for hypertension was more prevalent among those who chose to use aspirin at least 180 d/y. After controlling for this variable, blood pressure level itself did not predict posttrial use (data not shown). The relationship of aspirin use with cholesterol was more complex. Treatment for elevated cholesterol levels was more prevalent among those choosing aspirin. However, besides treatment, level of cholesterol also predicted aspirin use, and there was a significant interaction of cholesterol level and treatment on the choice of aspirin. Among those not receiving medication for lowering of cholesterol levels, there was a positive relation of cholesterol level with aspirin use. Among those using such medication, those with lower cholesterol levels were more likely to choose aspirin. Thus, those with cholesterol levels under greater control were more likely to use aspirin.
Although there were no significant differences in smoking status among the groups, there were strong differences in alcohol use and exercise, even after adjusting for the other cardiovascular risk factors. Frequent users of aspirin tended to report more frequent use of alcohol (at least up to daily use) and exercise, particularly when compared with those who did not use aspirin. Frequent aspirin users were also more likely to take supplements of vitamin E.
We assessed the relationship of self-selected posttrial aspirin use at 7 years with subsequent CVD and mortality in the period from 7 to 12 years of follow-up among those with no major CVD before this time (Table 3). All analyses were adjusted for all of the risk factors considered in Table 2 as well as randomized aspirin assignment. During this 5-year posttrial follow-up period, there were 311 unrefuted reports of MI, 266 strokes (including 185 ischemic and 34 hemorrhagic), 205 CVD-related deaths, and 782 total deaths, 652 of these among persons with no previous cancer. For MI, there was a statistically significant 28% lower rate of events in the frequent users (≥180 d/y) compared with the nonusers (0-13 d/y) (RR, 0.72; 95% CI, 0.55-0.95). The test for trend across all 4 categories was also significant (P=.02). There were no apparent effects of self-selected aspirin use on stroke, including ischemic or hemorrhagic stroke.
For CVD-related mortality, we observed a statistically significant 35% lower rate among frequent users of aspirin compared with nonusers (RR, 0.65; 95% CI, 0.47-0.89). Rates were also lower in the 2 intermediate groups, and the test for trend was significant (P=.03). Although the numbers of events were small, the difference was more apparent for deaths due to cerebrovascular disease than for those due to acute MI. There was also a marginally significant 22% lower rate of death due to total cancer in the frequent user vs nonuser group (RR, 0.78; 95% CI, 0.58-1.05), after excluding those with reported cancer before the 7-year follow-up. The test for trend across the 4 groups was statistically significant (P=.03), with a lower rate in the 121- to 179-d/y aspirin group as well. The rate of death due to non-CVD and noncancer causes was also lower in the frequent user group (RR, 0.53; 95% CI, 0.39-0.74). These differences were reflected in the rates of total mortality, which was 36% lower in the frequent user group than in the nonuser group, even after excluding those with previous cancer (RR, 0.64; 95% CI, 0.54-0.77). The trend across the 4 groups was highly significant (P<.001), with the 121- to 179-d/y group also experiencing lower mortality.
Randomized aspirin assignment was not a significant predictor of any of these outcomes in the posttrial period after controlling for observational aspirin use. No statistically significant interactions of randomized aspirin assignment and posttrial self-selected use were found for MI or stroke (Table 4), although the risk reductions for MI and CVD-related mortality were stronger among those who had been randomized to placebo. The interaction was significant and in the same direction for cerebrovascular-related deaths. For cancer-related deaths, other (non-CVD and noncancer) deaths, and total deaths among those with no previous cancer, the RR estimates were comparable among those who had been randomized to aspirin and to placebo. In separate analyses, no significant interactions of age and posttrial aspirin use were found (data not shown).
In the randomized trial, aspirin decreased the risk for a first MI by 44% (95% CI, 30%-55%). There were, however, insufficient numbers of events to reliably determine whether there were reductions for stroke or CVD-related mortality. In the posttrial period, those who self-selected for aspirin use of at least 180 d/y had rates of MI that were 28% lower than those who took aspirin 0 to 13 d/y, a difference that was statistically significant and consistent with the 44% reduction seen in the randomized trial. There was no significant decrease or increase in total or ischemic stroke in the randomized or posttrial periods. Although based on very small numbers, the randomized results suggested a possible increase in hemorrhagic stroke with aspirin use that was not apparent in the observational data. In addition, in the posttrial period, there was a significant reduction in CVD-related mortality with self-selected aspirin use, including mortality due to cerebrovascular causes, cancer, and noncancer and non-CVD causes and, as a result, in total mortality, findings that were not seen during the randomized period.
DATA FROM from randomized trials show a clear impact of aspirin on CVD, although data on mortality are limited. In primary prevention, an overview of the PHS and British Doctors' Trial found a 33% reduction in nonfatal MI, but no clear effect on CVD-related death or nonfatal stroke.15 In addition, there was no significant effect on ischemic stroke, but a possible 1.9-fold increase in risk for hemorrhagic stroke, based on small numbers. Secondary prevention trials, as well as trials of suspected evolving MI,16 have shown a clear benefit of aspirin use on stroke as well as on MI and death due to vascular causes. In an overview of 142 trials of antiplatelet therapy in high-risk patients who had survived a previous occlusive event, nonfatal MI and nonfatal stroke were reduced by one third, and death due to vascular causes was reduced by one sixth.6 Aspirin alone was associated with a highly significant 25% reduction in vascular events. A recent meta-analysis of primary and secondary prevention trials with stroke subtype information found significant reductions in MI, stroke, and CVD-related mortality, with an increase in risk for hemorrhagic stroke.17 The totality of evidence from randomized trials thus suggests a definite benefit of aspirin on MI, stroke, and CVD-related death in secondary prevention and in treatment of a suspected evolving MI. The evidence concerning primary prevention indicates a clear protective effect for MI, at least in men, but insufficient data are available for stroke or CVD-related death, and a possible small increased risk for hemorrhagic stroke remains plausible.
With respect to observational studies of aspirin, the earliest show a reduction in risk for first MI among men and women7 but no significant benefit concerning deaths due to coronary heart disease.8 A study among the elderly found a nonsignificant reduction in MI among men, but possible increases in ischemic heart disease among women.9 A more recent prospective study among middle-aged women in the Nurses' Health Study found a significant reduction in first MI among those taking aspirin 1 to 6 times a week compared with those taking none10 and nonsignificant reductions in CVD-related deaths and important vascular events. No effects on stroke were seen. In addition, women who self-selected aspirin 7 or more times a week had no decreases in MI, stroke, or death. In the Cardiovascular Health Study, examining short-term predictors of stroke among the elderly, an increase was seen among men and women self-selecting aspirin that appeared to be stronger in the subgroup with no previous CVD.11 In a more recent analysis after 4.2 years of follow-up in the same population, women, but not men, who used aspirin frequently experienced higher rates of ischemic stroke, and in both sexes combined there was an increase in hemorrhagic stroke.12 These observational findings need to be interpreted with caution, however, because questions remain concerning the reasons for aspirin use and doses used, and because the observed increased risk for ischemic stroke is not consistent with the totality of evidence from randomized trials.18
In the analysis of posttrial self-selected aspirin use in the PHS, different dose-response effects were seen for various outcomes, although the numbers in the middle 2 categories of aspirin use were small. The observed reduction in MI was restricted to those taking aspirin at least 180 d/y. These results are consistent with an earlier analysis in the PHS that showed the strongest benefit on MI among those with the highest levels of adherence.19 The confidence intervals for the 2 middle posttrial aspirin groups were wide, however, and could not exclude even a large reduction. For mortality, significant risk reductions were also observed in lower categories of use, especially for CVD mortality. These reductions in mortality, however, especially for non–CVD-related mortality, are not consistent with trial results.
The differences seen according to frequency of aspirin use as well as apparent discrepancies among the trial and observational results, particularly for mortality, may be due to residual confounding. Several risk factors were highly predictive of self-selected aspirin use in these data, including age, BMI, family history of MI, treatment of hypertension, and elevated cholesterol level. Smoking was not predictive of aspirin use, but other lifestyle factors were associated with more frequent self-selection of aspirin, including exercising at least once per week, more frequent alcohol consumption, and use of vitamin E supplements. Risk factors for CVD as well as an interest in prevention, as assessed through measures such as exercise, antioxidant use, and lipid-lowering medications, are thus associated with the frequency of aspirin use among these physicians.
Correlates of self-selected aspirin use seen in this study were similar to those seen among women in the Nurses' Health Study.10 Predictors in common included the traditional CVD risk factors of hypertension, high cholesterol levels, higher body weight, and family history of MI. In the Nurses' Health Study, however, aspirin users were more likely to smoke and exercised less than nonusers, although they also reported higher alcohol consumption. Both study populations differ in that less than 10% of the nurses taking aspirin were doing so for primary prevention of CVD. Although not undergoing assessment in the PHS, the latter proportion is likely higher in the population of physicians enrolled in a trial in which aspirin clearly reduced the risk for first MI. The prevalence of self-selected aspirin use was also much higher in the PHS, with 60% vs 22% in the Nurses' Health Study reporting use at least once every other day.
Thus, despite multivariate adjustment for a large number of risk factors, uncontrolled confounding by indication and unmeasured health behavior is likely to remain in these observational data. The unexpected reductions seen in non–CVD-related mortality, in particular, are unsupported in trial data, and may be due to such residual confounding. In primary prevention, the benefit of aspirin on MI is clear, but effects on stroke and vascular death remain unclear because of inadequate numbers of these events in randomized trials. Thus, the most reliable evidence on the balance of risks and benefits of aspirin will accrue from large-scale and long-term randomized trials, most notably the ongoing Women's Health Study, testing the effect of low-dose aspirin and vitamin E on CVD and cancer among 39,876 female health professionals.20
Accepted for publication June 29, 1999.
This work was supported by grants CA-34944, CA-40360, HL-26490, HL-34595, and HL-58476 from the National Institutes of Health, Bethesda, Md.
The authors are indebted to Martin Van Denburgh for his expert programming assistance.
Corresponding author: Nancy R. Cook, ScD, Division of Preventive Medicine, Brigham and Women's Hospital, 900 Commonwealth Ave E, Boston, MA 02215-1204 (e-mail: email@example.com).