Understanding Treatment Response Heterogeneity Using Crossover Randomized Controlled Trials: A primer for exercise and nutrition scientists

Crossover randomized controlled trials (RCTs) are common in exercise and nutrition sciences. Since researchers randomize participants to different sequences of the treatment and comparator/control conditions, crossover RCTs are powerful for detecting mean treatment effects under certain circumstances. We aim to review the information that can be derived from crossover RCTs about treatment response heterogeneity—a fundamental issue in precision medicine for tailoring treatments to individuals. After covering the fundamental design issues, we describe the variance components that underlie observed data. The crucial person-by-treatment variance component can be quantified from a repeated or “replicate” crossover RCT by exposing participants to multiple cycles of trial conditions. As a type of n-of-1 trial, replicate crossover RCTs have important design and statistical power considerations, which we describe. By synthesizing findings from our six published replicate crossover RCTs, we also compare various data analysis approaches. We find general agreement between these approaches, and a link between within-person consistency of response and the detection of person-by-treatment interactions. We postulate that a paired “variance comparison,” for example, the Pitman–Morgan test, provides some preliminary information regarding response heterogeneity from a typical single-cycle crossover RCT. Nevertheless, underlying assumptions are critical, rendering these comparisons as merely exploratory until an n-of-1 or replicate crossover RCT is undertaken. Multiple-cycle n-of-1 trials and replicate crossover RCTs are underused but are informative for treatment response heterogeneity. However, these trials are still only one component of the process for predicting individual magnitude of response from any personal traits, which is the “holy grail” of personalized treatment.