An optimal rule for screening subjects for clinical trials in the presence of within-person variability

In screening subjects for a continuous variable, such as blood pressure (BP), for eligibility for clinical trials, multistage rules are often used to select subjects within a target range. Multiple readings may be taken over successive visits, and the subject must meet eligibility criteria at each visit. We present an optimal rule that minimizes the number of clinic visits needed per eligible subject or the overall number of screening visits needed to achieve a given sample size of eligible subjects. We illustrate these methods with the design of phase I of the Trials of Hypertension Prevention (TOHP), which attempts to identify subjects with mean diastolic BP in the range 80- < 90 mm Hg over three screening visits. The optimal rule requires only 73% of the visits needed with a no-screening rule in which all subjects are examined at each visit, and the average BP at all screening visits is used as a criterion. The optimal rule requires only 36% of visits needed in a more standard approach in which a subject must have an observed mean diastolic BP of 80- < 90 mm Hg on each of three consecutive visits. Alternatively, one may minimize the expected cost of screening per eligible subject when screening visits differ in cost. Examples are given of optimal rules with differing relative costs per visit.