Cutoff assignment strategies for enhancing randomized clinical trials

Abstract

The randomized clinical trial (RCT) is the preferred method for assessing the efficacy of treatments. Recent ethical and logistical criticisms suggest that new variations of the traditional RCT are needed. Some of these criticisms may be addressed with new hybrid designs that combine random assignment with assignment by one or more cutoff values on a baseline variable (e.g., severity of illness). In a simple version of such a "cutoff-based" RTC, persons scoring below a cutoff score on a baseline measure (i.e., the least severely ill) are automatically assigned to the control-treated group, those scoring above a second, higher cutoff (i.e., the most ill) are automatically assigned to the test-treated group, and those scoring in the interval between the cutoff scores (i.e., the moderately ill) are randomly assigned to either group. Depending on the baseline score, the patient is assigned to treatment either randomly or by the need-based, clinically related baseline score. Six cutoff-based design variations are studied via simulations and compared with the traditional RCT and the single-cutoff (i.e., regression-discontinuity) design. All variations yield unbiased estimates of the treatment effect but estimates differ in efficiency, with the RCT being most efficient and the single-cutoff design being least efficient. Secondary analyses of data from the Cross-National Collaborative Study of the Effects of Alprazolam (Xanax) on panic are conducted for each variation by selectivity discarding cases from the original dataset to stimulate cutoff-based assignment. The results confirm the simulations and illustrate how cutoff-based designs might look with real data.