Modeling personalized treatment effects using multiple data sources

In this chapter, we will discuss statistical methods for developing models to predict conditional average treatment effects using data from multiple randomized and non-randomized studies. We begin with prediction models that assume a constant treatment effect and explore how to account for heterogeneity in baseline risk. We then introduce two main strategies for modeling treatment effect heterogeneity in the context of individual participant data meta-analysis (IPD-MA): the risk modeling and effect modeling approaches. For each, we first describe how to combine IPD from randomized controlled trials comparing the same two treatments, before extending these methods to incorporate multiple treatments, real-world data, and aggregate published results. We also discuss available statistical software and provide example code. Real-world case studies are used throughout to illustrate key concepts and methods.