Individualized Treatment Effects: Machine Learning Can Revolutionize Observations, but Let's Understand What We Are Observing

Mosier discusses the study by Seitz and colleagues which reimagine how we can use machine learning to evaluate the potential individualized treatment effects in critically ill patients. Clinical trial results present an average treatment effect across all participants and traditionally investigate heterogeneity by individually evaluating potential effect modifiers. Machine learning approaches provide a new way of evaluating relationships between variables and the magnitude of those relationships. Seitz and colleagues used a causal forest algorithm on prespecified predictor variables using the BOUGIE trial data as a demonstration of the potential for this machine learning approach. The causal forest algorithm used baseline patient and operator variables to find relationships between them using aggregated decision trees to predict individualized outcomes.