A novel artificial intelligence-based methodology to predict non-specific response to treatment

Abstract

Non-specific response to treatment (NSRT) is the primary contributor to the failure of randomized clinical trials in major depressive disorder (MDD). The objective of this study is to develop artificial neural network (ANN) models to predict the individual probability for NSRT. Pre-randomization data from a failed antidepressant trial were considered as potential predictors of the NSRT probability (prob-NSRT) using the response endpoint in subjects randomized to placebo. The inverse of the individual prob-NSRT (NSRT propensity score) was used as a weight in the mixed-effects model applied to assess treatment effect (TE). The comparison of the results obtained with and without the NSRT propensity score indicated that the weighted analyses provided an estimate of TE significantly larger than the conventional analyses. The propensity score weighted (PSW) analysis, adjusting for inter-individual variability in prob-NSRT, enhanced signal detection of TE. These findings support the potential role of PSW methodology for analyzing RCTs and determining TE. However, external validation of these ANN models in at least one independent trial is needed before advocating regulatory or broader clinical use.