Making predictions under intervention to avoid causal blind spots in treatment decisions - Niels Peek, THIS Institute, University of Cambridge
Prediction models are increasingly used to support medical decisions, but fail to address the special role of treatments, leading to inappropriate use. In this talk I will discuss the limitations of using standard prediction models for treatment decision support, and describe "causal blind spots" in three common approaches to handling treatments in prediction modelling, with illustrations how this can occur in examples of implemented prediction models. I will argue that prediction models should specify upfront the treatment decisions they aim to support and target a prediction estimand in line with that goal. This will improve the efficacy of prediction models in guiding treatment decisions and prevent potential negative effects on patient outcomes. Finally I will review causal reasoning and inference techniques that can be used to develop and validate models for making predictions under intervention.
You can also Register to attend virtually.
You can also watch recordings of previous BSU Seminars
If you have a question about this seminar, please email: alison.quenault@mrc-bsu.cam.ac.uk