42 D. Canter Methods Patient population Three-hundred-and-forty-four patients were entered into the studies, 303 men and 41 women. Three patients were withdrawn from the trials having been screened, randomized and allocated a study number but prior to receiving drug or placebo, two at the patient's own request and one because of a myocardial infarction. All 344 patients were analysed for endpoints on an intention to treat principle, but only 341 patients were analysed for evidence of drug toxicity. For each patient the trial records were reviewed, and periods of placebo, washout, known non-compliance and the taking of an alternative lipid-regulating drug were calculated to the nearest month and deducted from the total time of the trial. All references to time therefore represent the total time of exposure to gemfibrozil. For six studies, the entry and exclusion criteria were the same and no patient had previously had a myocardial infarction. The plasma lipid acceptance criteria for all six trials were a total serum triglyceride level >1·7 mmol/l and/or a total serum cholesterol > 7·7 mmol/l. In the seventh study (Kaukola et al. 1981) all patients had had a myocardial infarction at least 3 months prior to entry. The daily dose of gemfibrozil was 1200 mg/day for all of the long-term, open-label phases in the trials. Statistical methods Laboratory parameters These variables were summarized as means and standard deviations at each time point (yearly intervals). Means and standard deviations of serial changes and paired t-tests were done on the original data for total WBC and on log-transformed data for ESR and ALT, which are not normally distributed. An analysis was also done for haemoglobin on values above and below the laboratory's normal range using the exact binomial version of McNemar's test. Both increases and decreases in haemoglobin were judged to be detrimental. Analysis of erythrocyte sedimentation rate (ESR) examined the proportions above the upper normal limit and above three times the upper normal limit. The upper limit of normal for the ESR in two studies (Centres P66 and P91) was <10 mm/h and in the remaining studies was <7mm/h. In order that the results of all trials could be compared by parametric analysis, the ESR values from Centres P66 and P91 were first multiplied by 0.7. This would affect the means and standard deviations but not the serial changes. The view was taken that the actual recorded values were less meaningful than the relationship to the laboratory's reference range. A change in the laboratory method for alanine aminotransferase (ALT) analysis during the course of the studies produced a change in the upper limit of normal for ALT from <17 IU/l to <40 IU/1 for all centres. All values following the change were identified, and were multiplied by 0-425 so as to express all results relative to a reference range < 17 IU/1, and thus allow parametric analysis. This would not affect the evaluation of serial changes. Studies with gemfibrozil in Finland Analysis of endpoints 43 The incidence rates per 1000 patient-years were calculated together with 95 per cent confidence limits based on the Poisson distribution. Analysis was based on the "first" endpoint only, and repeated "events" for patients who remained in the trial did not count twice. For example, a patient who had a myocardial infarction, but continued in the trial and subsequently had a second myocardial infarction, would appear in the analysis of endpoints as a single myocardial infarction. The six trials in which a previous myocardial infarction was not an inclusion criterion were analysed separately from the seventh trial (Kaukola), because this trial represents a group of men with a much higher risk of a second myocardial infarction and a higher risk of premature death. The standardized mortality ratio (SMR) is the ratio of observed deaths in the population under study to expected deaths in the standard population, specific for coronary heart disease. This ratio is multiplied by 100 and expressed as a percentage (Armitage 1971). If the observed deaths are equal to the expected deaths then the SMR will equal 100 per cent. The SMR for mortality attributable to coronary heart disease was calculated using rates for the whole of Finland, by sex and 10-year age groups. These mortality rates were not specific for men and women with dyslipidaemias. All statistical tests were two-tailed and the level of significance was p<0.05. Results This analysis represents 1203 patient-years of medication with gemfibrozil. A total of 344 patients were entered into the trials, 303 men (mean age at entry ± SD: 43.8±6.9 years; age range 18-70 years) and 41 women (52.0±9.5 years; age range 19-68 years). The maximum period of exposure to gemfibrozil was 8 years; the mean duration was 3.6 years. Figure 1 shows the length of time in years that each patient was exposed to gemfibrozil. Figure 1. Duration of patient exposure to gemfibrozil (n = 344) One-hundred-and-sixty-four patients were still taking gemfibrozil at the cut-off point for this analysis in November 1983. Of those entering, 180 patients (47.7 per cent) had dropped out of the trial for reasons listed in Table 1. All adverse events other than gastro-intestinal symptoms were grouped together and no attempt is made to differentiate between those events which are attributable to the drug and those which are probably not. Twenty-six patients (7.6 per cent of the total) had gastro-intestinal upset which was the commonest drug-attributable side-effect requiring withdrawal from a trial. Laboratory results Haemoglobin (or haematocrit) The largest differences from baseline values occurred at 1 year and using the exact binomial version of McNemar's test there were no statistically significant changes. The changes from baseline which occurred with increasing exposure to gemfibrozil were less than those at 1 year and were therefore not worth testing for statistical significance (Table 2). If a data value was not available within ±1 month of the assessment point it was regarded as 'missing'. No patient was withdrawn from any of the studies due to an abnormal haemoglobin or haematocrit value. White blood cell count (WBC) There was a consistent decrease comparing baseline with each yearly interval for year 2 to year 8 and this decrease was statistically significant using a t-test in years 4, 5 Studies with gemfibrozil in Finland Table 2 45 Changes in haemoglobin (or haematocrit) values from final pre-trial assessment to 1 year Serial changes in WBC from last baseline to 1 to 8 years. Summary information based on all seven trials and 6. The decreases were small and no patient either had any clinical evidence of leucopenia or was withdrawn from a study because of a decrease in the white cell count. The minimum value for the total WBC count for any patient was 2·2× 109/1. Table 3 summarizes the data at yearly intervals. Erythrocyte sedimentation rate (ESR) A 'normal' value for ESR is difficult to define. When the reference ranges used by the laboratories (either < 10 or <7 mm in the first hour) and an arbitrary value three times this level (either <30 or <21 mm) were compared to our population group at baseline, they corresponded to the 37th and 87th percentiles respectively. The pattern that emerged was a general decrease in ESR when the mean value at each yearly interval was compared to the mean baseline value. This decrease was apparent at year 1 and was subsequently maintained from year 2 to year 7 (Table 4). Aspartate aminotransferase (ALT) The changes in ALT were highly conflicting among the seven trials. Two trials showed a consistent increase in ALT compared to baseline, three showed a consistent decrease 46 Table 4 Serial changes in ESR from last baseline to 1 to 8 years D. Canter Serial changes in ALT from last baseline to 1 to 8 years by different trial centres Geometric mean change as percentage from last baseline in trial for each centre and the remaining two were equivocal (Table 5). From baseline to year 7, between 7 per cent and 12 per cent of patients had ALT values above the upper normal limit (either < 17 IU/l or <40 IU/l) at each assessment point. Four patients were removed from the trials specifically for elevated liver enzymes and one patient was removed with alcoholic cirrhosis. These five patients had been in the trials for 1, 4, 5, 6 and 2 years respectively. Analysis of endpoints The cause of all deaths was known. There were 10 deaths from myocardial infarction (MI), two from cerebrovascular accidents (CVA) and two from trauma (one drowning and one fall whilst intoxicated). There was a total of 10 fatal and four non-fatal myocardial infarctions. Five of these 14 patients came from the trial of 59 men who had all previously had a myocardial infarction (Kaukola). All five of these men died following their second myocardial infarction. The mean age at death of these five men suffering their second MI was 46-4 years (SEM±1·3) whilst the mean age at |