Multiple-stage screening and mortality in the Multiple Risk Factor Intervention Trial

Abstract

During the design phase of a clinical trial, sample size estimates should take into account medical screening criteria, the 'healthy volunteer' effect, consequences of run-in phases, and secular trends in the event rate of interest. All of these have been shown to relate to subsequent event rates, and hence trial power to detect intervention effects. The Multiple Risk Factor Intervention Trial (MRFIT) used three successive screenings of 361 662 men to enroll 12 866; observed coronary heart disease (CHD) mortality after a mean of 6.9 years was substantially lower than projected during design. We explore factors which may have contributed to these mortality differences and whether they persisted throughout follow-up. Proportional hazards models were used to compare 25-year mortality according to trial eligibility, self-exclusions, medical exclusions, and participation. After adjustment for baseline risk factors and age, there was higher mortality among men excluded for presence of disease [hazard ratio (HR) 1.52, 95% confidence interval (CI) 1.43-1.61, for total; HR 1.92, CI 1.75-2.11, for CHD] compared to those not excluded which persisted throughout follow-up. Volunteers had lower total (HR 0.82, CI 0.76-0.87) and CHD (HR 0.79, CI 0.70-0.88) mortality than those discontinuing participation. Men excluded with characteristics deemed likely to interfere with adherence had higher total (HR 1.19, CI 1.07-1.33) and non-cardiovascular disease (CVD) (HR 1.32, CI 1.14-1.53) mortality but no higher CVD (HR 1.04, CI 0.88-1.23) or CHD (HR 0.98, CI 0.80-1.20) mortality compared to those not excluded. Differences in mortality were stronger during the first five years, but declined only slightly over 25 years. 25-year mortality was significantly higher for non-volunteers and exclusions. Differences between observed and predicted six-year total mortality for trial participants were largely attributable to volunteers and exclusions, but there were additional differences for CHD mortality, which were likely due to downward secular trends. These results emphasize the importance of anticipating these factors during clinical trial design, even for trials of short duration.