A Randomized Trial of Low-Dose Aspirin in the Prevention of Clinical Type 2 Diabetes in Women

RESEARCH DESIGN AND METHODS—

The WHS was a 2 × 2 factorial trial evaluating the balance of risks and benefits of low-dose aspirin (100 mg every other day; Bayer Healthcare) and vitamin E (600 IU α-tocopherol every other day; Natural Source Vitamin E Association) in the primary prevention of cardiovascular disease and cancer (13–15). The dose of 100 mg every other day was chosen to be the lowest dose that would have a cardioprotective effect while minimizing gastrointestinal side effects. Although not a prespecified primary end point of the WHS, clinical diabetes was a key outcome of interest and prospectively ascertained throughout the duration of the trial.

Written informed consent was obtained from all women. The study was approved by the institutional review board of the Brigham and Women's Hospital and monitored by an external data and safety monitoring board.

Setting and participants

Details of the study design have previously been described (13,16). Women were eligible if they were at least 45 years of age without a previous history of cancer (except nonmelanoma skin cancer), cardiovascular disease, or other major chronic illness; had no history of adverse effects from aspirin; were not taking aspirin or nonsteroidal anti-inflammatory drugs (or were willing to forgo their use during the trial); and were not taking anticoagulants or individual supplements of vitamin A, E, or β-carotene more than once a week. A total of 39,876 women were willing, eligible, and compliant during a 3-month placebo run-in period and underwent random assignment: 19,934 were assigned to receive aspirin and 19,942 to receive placebo. In the present analyses, we excluded women with reported physician-diagnosed diabetes at baseline (n = 1,160), leaving a total of 38,716 women free of clinical diabetes at entry into the trial; 19,326 were assigned to receive aspirin and 19,390 to receive placebo. Additional details including a Consolidated Standards of Reporting Trials (CONSORT) flow diagram are provided in supplemental data and supplemental Fig. 1 (available in an online appendix at http://dx.doi.org/10.2337/dc08-1206).

Randomization and interventions

With use of a computer-generated table of random numbers, treatment assignments were made within seven age-groups (45–49, 50–54, 55–59, 60–64, 65–69, 70–74, and ≥75 years) using block sizes of 16. Random assignment took place from 30 April 1993 through 24 January 1996. Each year, women received calendar packs that contained amber capsules (vitamin E or placebo) and white pills (aspirin or placebo) on alternate days. Every 6 months for the first year and annually thereafter, they also received follow-up questionnaires inquiring about compliance with pill-taking, potential adverse effects, occurrence of end points, and risk factors. Study medications were continued in blinded fashion through the scheduled end of the trial (31 March 2004).

Compliance

On the basis of self-reported adherence, compliance, defined as taking at least two-thirds of their aspirin or matching placebo tablets, was 76.1% at 5 years and 67.0% at 10 years. Averaged throughout the trial, compliance was slightly higher in the placebo (73.7%) group than in the active (72.5%) group (P = 0.004). Nontrial use of aspirin or aspirin-containing products on ≥4 days/month (“drop-ins”) was 11.6% at 5 years and 19.2% at 10 years. Averaged throughout the trial, it was somewhat higher in the placebo (13.0%) group than in the active (12.7%) group (P = 0.10).

Outcomes and follow-up

Details regarding the ascertainment of incident diabetes in the WHS have been reported previously (17). Information on newly diagnosed diabetes was collected on every follow-up questionnaire from baseline through the end of the trial. All participants were asked annually “In the past year, were you newly diagnosed with diabetes mellitus?” Subjects also provided the month and year of diagnosis. Confirmation of diabetes was conducted in a blinded fashion using American Diabetes Association diagnostic criteria (18). Self-reported cases were investigated by either telephone interview conducted by a physician or a previously validated self-administered supplemental questionnaire that inquired about symptoms, diagnostic testing, and use of diabetes medications (19). The response rate was high with >90% of women who reported incident diabetes responding to either the telephone interview or supplemental questionnaire. On the basis of responses to the supplemental questionnaire, 77.2% of those with confirmed cases reported use of antidiabetic agents. Only women with confirmed cases were analyzed in this report. Because the vast majority of diabetes diagnosed at age ≤45 years is of the type 2 variant, incident diabetes in the WHS is considered to be type 2 diabetes.

Glucose screening rates were assessed during follow-up. When asked about screening for diabetes on the 9-year questionnaire, 71.8 and 68.2% of nondiabetic women reported having a fasting glucose test performed within the preceding 5 and 3 years, respectively. Screening rates were equivalent between the two treatment arms: 68.3 versus 68.1% in the preceding 3 years in the aspirin and placebo groups, respectively (P = 0.78). These values are similar to contemporaneous diabetes screening rates; among patients in a U.S. managed care population, the occurrence of any glucose testing (random or fasting) over a 3-year period was 71.5% for women ≥45 years (20).

Statistical analysis

All primary analyses were performed on an intention-to-treat basis. We used Cox proportional hazards models to estimate the rate ratio (RR) and 95% CIs, comparing event rates in the aspirin and placebo groups after adjustment for age and other randomized treatment assignments (vitamin E and β-carotene, which was a component of the trial for a median of 2.1 years) (21). The proportionality assumption of constant hazards over time was tested using an interaction term of aspirin with the logarithm of time. The divergence of diabetes incidence over time between groups was estimated using Kaplan-Meier survival curves and the log-rank test was computed to compare curves.

Exploratory subgroup analyses were conducted to examine the effect of aspirin according to the baseline presence of major risk factors for type 2 diabetes including age-group, BMI group, family history of diabetes in a first-degree relative, physical activity, menopausal status, and hormone therapy. Categories are specified in Table 1. Among women providing baseline blood specimens (n = 27,167), subgroup analyses were performed after stratification by levels of total cholesterol, LDL cholesterol, HDL cholesterol, the total cholesterol–to–HDL cholesterol ratio, non-HDL cholesterol, A1C, and high-sensitivity C-reactive protein (hsCRP).

To estimate the effect of treatment duration, we fit two separate proportional hazards models to the experience of the first 5 years and after 5 years of follow-up. To assess the impact of potentially undiagnosed diabetes at baseline, we conducted sensitivity analyses in which women with diabetes diagnosed during the first 2 and 5 years of follow-up were excluded. To examine the effect of actual as opposed to assigned aspirin use, we performed additional analyses in which subjects were censored if and when they stopped taking at least two-thirds of their study pills. To assess the effect of nontrial aspirin use, we also performed analyses in which censoring occurred when women either stopped taking at least two-thirds of their study pills or reported outside use of aspirin or aspirin-containing products for ≥4 days/month.

All analyses were performed with SAS (version 9.1; SAS Institute, Cary, NC). A two-sided significance level of 0.05 was used.

RESULTS—

The aspirin and placebo groups were similar with respect to baseline characteristics (Table 1). Among women providing blood specimens, there were also no significant differences in lipid levels, A1C, or hsCRP. The median duration of follow-up was 10.2 years (mean 9.8 years). At completion of the trial, 1,696 cases of confirmed incident clinical type 2 diabetes had occurred. There were 849 cases in the aspirin group and 847 cases in the placebo group, with no significant risk reduction (RR for aspirin versus placebo 1.01 [95% CI 0.91–1.11]).

The cumulative incidence curves according to treatment assignment were similar throughout follow-up (Fig. 1) (log-rank P = 0.92). A test of the proportionality hazards assumption showed no deviation from proportionality (P = 0.27). In separate analyses that considered newly diagnosed diabetes in the first 5 years of follow-up versus thereafter, aspirin therapy was associated with an RR of 0.99 (95% CI 0.86–1.15) during the first 5 years and 1.01 (0.90–1.15) after 5 years. We found no difference when patients with potentially undiagnosed diabetes documented during the first 2 or 5 years of follow-up were excluded. In these analyses the RRs were 0.99 (0.89–1.09) and 1.01 (0.90–1.15), respectively.

There was no evidence that any diabetes risk factors considered modified the effect of aspirin on diabetes incidence (Table 2) (P_interaction ≥0.2 for clinical risk factors). Although we did find statistically significant risk reductions among women who had total cholesterol levels >6.21 mmol/l (RR 0.77 [95% CI 0.61–0.97], P = 0.028) and LDL cholesterol >4.13 mmol/l (0.76 [0.58–1.00], P = 0.049), with borderline nonsignificant findings for women with total cholesterol–to–HDL cholesterol ratio ≥6.0 (0.81 [0.66–1.00], P = 0.053), these findings must be interpreted with caution given both the lack of a consistent risk increase across biomarker categories and the large number of subgroups examined. There was no statistically significant interaction for any of the biomarkers assessed (P_interaction = 0.39 for total cholesterol, 0.26 for LDL cholesterol, 0.13 for the total cholesterol–to–HDL cholesterol ratio, and ≥0.2 for all others).

Because compliance diminished over time, sensitivity analyses were performed that censored noncompliant women at the time they stopped taking at least two-thirds of their study pills during the preceding year. In this analysis, there was also no significant benefit of aspirin (RR 1.02 [95% CI 0.91–1.15]). When women were censored at the time they either stopped taking at least two-thirds of their study pills or started outside aspirin or aspirin-containing medications on ≥4 days/month, findings were similarly nonsignificant. In addition, there was no evidence of effect modification by other randomized treatments. With regard to safety, as expected, there were slightly higher rates of clinically significant bleeding episodes and other side effects in the aspirin arm compared with placebo: 4.5 vs. 3.7% (RR 1.22, P < 0.001) for any bleeding event, 0.6 vs. 0.4% (RR 1.37, P = 0.03) for transfusion-requiring bleeding, 2.7 vs. 2.1% (RR 1.30, P < 0.001) for peptic ulcer, 15.2 vs. 14.4% (RR 1.06, P = 0.02) for hematuria, and 19.0 vs. 16.5% (RR 1.17, P < 0.001) for epistaxis.

CONCLUSIONS—

In this large-scale, long-term trial of initially healthy women, there was no association of 100 mg aspirin on alternate days with the overall incidence of clinical type 2 diabetes. Treatment duration, ≤5 years or >5 years, did not have an impact on our results, and the treatment effect did not vary significantly across subgroups of women at high risk for diabetes due to the presence of clinical risk factors, dyslipidemia, elevated A1C, or hsCRP. There was no difference when women with early cases, presumably undiagnosed at baseline, were excluded or in analyses accounting for adherence to randomized assignment. Aspirin therapy was associated with a higher incidence of clinically important bleeding events.

The increasing incidence of diabetes, high treatment costs, and disproportionate impact on cardiovascular disease in women highlight the need to identify prevention strategies applicable on a broad population basis. Preventive measures with potential dual effects on cardiovascular and diabetes risk reduction are particularly appealing because “cardiometabolic” risk factors often coincide in the same individual. The main findings of the Women's Health Study were published previously and demonstrated that long-term treatment with low-dose aspirin, although resulting in no significant benefit or harm on the end point of any first major cardiovascular event, did significantly reduce the risk of stroke overall (RR 0.83 [95% CI 0.69–0.99]) and myocardial infarction among women aged >65 years (0.66 [0.44–0.97]) (13). The current analysis addresses the important question of whether low-dose aspirin has added benefits for diabetes prevention in this large population of otherwise healthy women. This report provides the only randomized data available in this regard.

One of the least appreciated pharmacological actions of aspirin is its ability to lower glucose levels. Attempts to delineate this effect have been made in several small clinical studies. Among both nondiabetic and diabetic patients, short-term high-dose aspirin (3–10 g/day for up to 3 weeks) is consistently associated with higher basal and stimulated insulin concentrations and reduced glucose excursion during glucose tolerance testing (8–10,22). However, whether aspirin has a net beneficial contribution to glucose homeostasis has been controversial, with several studies suggesting that any favorable hypoglycemic action is offset by deterioration in insulin sensitivity (23,24). Higher doses of aspirin given over a longer period were more recently found to have broad therapeutic benefits in a detailed study of nine patients with overt type 2 diabetes (10). At the end of 2 weeks, aspirin at a dose of 7 g/day lowered fasting glucose (∼25%) and C-reactive protein (∼15%) while improving glucose tolerance (∼20%), reducing basal hepatic glucose production (20%), and increasing insulin-stimulated peripheral glucose uptake (20%). Importantly, however, the high dose of aspirin used in all of these studies is known to cause serious side effects.

The current analysis directly evaluated whether a low dose of aspirin acceptable for long-term use in routine clinical practice is effective in reducing clinical type 2 diabetes. Although we found no evidence to support this approach, there are three main possibilities that could account for our null results. First, the dose of aspirin used in this study may be insufficient to impart a clinical benefit. As noted, prior studies evaluating the glucose metabolic effects of aspirin have used far higher doses. Data pertaining to potential hypoglycemic actions of low-dose aspirin have not been available. However, short-term treatment with doses as low as 30–81 mg/day have been shown to improve several systemic inflammatory parameters, including lowering soluble CD40 ligand and promoting counterregulatory anti-inflammatory mediators such as 15-epi-lipoxin A₄ (11,12). Despite these latter findings, our data demonstrate that chronic treatment with 100 mg aspirin on alternate days does not prevent clinical diabetes. However, intermediate doses as high as 1.3 g/day have been used in long-term clinical trials of cardiovascular disease prevention, and these settings may offer additional opportunities to evaluate this issue if diabetes incidence was also ascertained.

Second, women enrolled in the WHS were generally at low risk for diabetes as evidenced by the low prevalence of obesity, predominantly non-Hispanic white ethnicity, and lower rates of clinical diabetes compared with similarly aged women in the overall U.S. population (25). It is possible that the use of aspirin in a higher risk group may have led to detection of a beneficial effect. However, in subgroup analyses of women with high-risk features we did not find strong support for differential effects. Furthermore, there was no benefit among women having evidence of inflammation as reflected by elevated hsCRP.

Third, underdiagnosis of diabetes may have influenced our results. Women did not undergo systematic screening for diabetes or glucose intolerance as a part of the study; thus, we detected clinical diabetes as ascertained during routine clinical practice rather than all women with biochemical evidence of disease. However, any misclassification due to unrecognized diabetes would be nondifferential between treatment arms and unlikely to lead to important alterations in the estimation of relative effects. In addition, we are reassured that reported diabetes screening rates among participants in our study were similar to contemporaneous screening rates in the general population and equivalent in the two treatment groups.

Major strengths of this analysis include the randomized setting, large study population, long treatment duration, large number of events, and ability to examine several large high-risk subgroups of women. In addition, baseline blood specimens were provided by roughly 70% of women in whom hsCRP and A1C levels were available. Importantly, systematic data were also collected on the occurrence of significant bleeding events. Limitations of this study have been alluded to previously. An important limitation is the lack of systematic screening for more sensitive measures of glucose intolerance and insulin resistance during follow-up. While the costs associated with such diagnostic testing are prohibitive in this large-scale setting, we cannot with the current data determine the potential impact of low-dose aspirin on these subclinical markers of incipient disease.

In summary, aspirin at a dose of 100 mg on alternate days is not effective for the prevention of clinical type 2 diabetes among otherwise healthy women at generally low risk for this disease. Our data do not pertain to other salicylate agents currently being evaluated for diabetes treatment or to intermediate or high doses of long-term aspirin in primary prevention. However, even at the low dose evaluated in this trial, the use of aspirin was associated with a significant increase in clinically important bleeding events and any potential benefit at higher doses, if found, must take into account this potential for excess risk.