Issues in Data Analysis

The Physicians' Health Study is a randomized, double-blind, placebo-controlled trial designed to determine whether low-dose aspirin (325 mg every other day) decreases cardiovascular mortality and whether beta carotene reduces the incidence of cancer. The aspirin component was terminated earlier than scheduled, and the preliminary findings were published. We now present detailed analyses of the cardiovascular component for 22,071 participants, at an average follow-up time of 60.2 months. There was a 44 percent reduction in the risk of myocardial infarction (relative risk, 0.56; 95 percent confidence interval, 0.45 to 0.70; P less than 0.00001) in the aspirin group (254.8 per 100,000 per year as compared with 439.7 in the placebo group). A slightly increased risk of stroke among those taking aspirin was not statistically significant; this trend was observed primarily in the subgroup with hemorrhagic stroke (relative risk, 2.14; 95 percent confidence interval, 0.96 to 4.77; P = 0.06). No reduction in mortality from all cardiovascular causes was associated with aspirin (relative risk, 0.96; 95 percent confidence interval, 0.60 to 1.54). Further analyses showed that the reduction in the risk of myocardial infarction was apparent only among those who were 50 years of age and older. The benefit was present at all levels of cholesterol, but appeared greatest at low levels. The relative risk of ulcer in the aspirin group was 1.22 (169 in the aspirin group as compared with 138 in the placebo group; 95 percent confidence interval, 0.98 to 1.53; P = 0.08), and the relative risk of requiring a blood transfusion was 1.71. This trial of aspirin for the primary prevention of cardiovascular disease demonstrates a conclusive reduction in the risk of myocardial infarction, but the evidence concerning stroke and total cardiovascular deaths remains inconclusive because of the inadequate numbers of physicians with these end points.

A new type of research, termed meta-analysis, attempts to analyze and combine the results of previous reports. We found 86 meta-analyses of reports of randomized controlled trials in the English-language literature. We evaluated the quality of these meta-analyses, using a scoring method that considered 23 items in six major areas--study design, combinability, control of bias, statistical analysis, sensitivity analysis, and application of results. Only 24 meta-analyses (28 percent) addressed all six areas, 31 (36 percent) addressed five, 25 (29 percent) addressed four, 5 (6 percent) addressed three, and 1 (1 percent) addressed two. Of the 23 individual items, between 1 and 14 were addressed satisfactorily (mean +/- SD, 7.7 +/- 2.7). We conclude that an urgent need exists for improved methods in literature searching, quality evaluation of trials, and synthesizing of the results.

To display a number of estimates of a parameter obtained from different studies it is common practice to plot a sequence of confidence intervals. This can be useful but is often unsatisfactory. An alternative display is suggested which represents intervals as points on a bivariate graph, and which has advantages. When the data are estimates of odds ratios from studies with a binary response, it is argued that for either type of plot, a log scale should be used rather than a linear scale.