Abstract:
|
Precision medicine (PM) methods aim to discover an optimal individualized treatment rule (ITR), which recommends a treatment based on patient characteristics at the time of treatment decision. The goal of this study is to understand the ability of PM methods to correctly identify optimal ITR for count outcomes. A diverse collection of PM methods is used including generalized linear models, tree-based models, and doubly robust estimators. Methods are evaluated with cross validation using the estimated value function, which is the average treatment effect in the population had the treatment choice followed the estimated ITR. In simulations, we study the impact of sample size and different levels of treatment effect heterogeneity. In the case study of a randomized clinical trial, we compared optimal ITRs between two multiple sclerosis treatments, dimethyl fumarate and glatiramer acetate. Results show that model performance depends on the proportions of neutral, moderate, and high responders to treatment, and a subset of methods outperforms the “one-size-fits-all" rule, where all patients receive the same treatment, especially with larger sample sizes.
|