Uncovering treatment effect heterogeneity in pragmatic gerontology trials

March 1, 2026

A recent publication by Heather Allore, PhD, and Fan Li, PhD, both members of the IMPACT Design and Statistics Core, and Guangyu Tong, PhD, a faculty scholar, demonstrates how embedding machine learning methods such as Bayesian Additive Regression Trees (BART), can help find treatment effect heterogeneity in gerontologic trials.

The authors of this paper address the challenges of detecting and measuring treatment effects in pragmatic trials involving older populations when traditional subgroup analyses often fail to detect complex patterns of variation in treatment response.

The researchers applied an integrated machine learning approach for pragmatic cluster randomized trials that incorporated mixed-effects BART into both participant subgroup and outcome models and reanalyzed the Whole Systems Demonstrator telecare trial to show how treatment effects among survivors can differ from overall intent-to-treat results.

This approach showed that some subgroups did see benefit from the intervention, which provides an opportunity for more personalized intervention strategies. The findings show that embedding machine learning methods within a principled causal inference framework can offer deeper insights into trial data with complex features.

Access the full journal article, titled “Uncovering treatment effect heterogeneity in pragmatic gerontology trials,” in Experimental Gerontology.