Estimating main and interaction effects of a multi-component randomized controlled trial via simulation meta-heuristics

Randomized controlled trials often are conducted on multi-component interventions in all-or-none manners for practical, logistic, or statistical purposes, typically allowing researchers the ability only to estimate the overall effect of the intervention en masse. In parallel, we propose a simulation-based approach to estimate main and interaction effect sizes of an intervention's sub-components based on conducting meta-heuristic parameter search on intra-trial longitudinal input, output, and context data. This approach is illustrated with a recent application to a healthcare intervention consisting of three information technology patient safety tools tested as a single intervention in a multi-unit staggered cluster crossover RCT design, with the overall objective being to reduce falls, infections, and other adverse events. (AEs) A high fidelity simulation of individual and combined use of these tools was developed and validated with retrospective data and then applied to prospective longitudinal data as clinical units varied in occupancy, staffing, patient risk, care team composition, and tool adherence, with parameter search estimating main and interaction effect sizes to maximally reproduce observed data. Computational results and implications are discussed.