An introduction to counterfactual prediction and conformal inference
There is an increased interest in prediction under hypothetical future exposures, often with a view of informing medical decision making. For example, we may want to build a clinical prediction model for the risk of developing a health outcome, such as heart failure, if a patient were to start pharmacological treatment. Building such a prediction model using observational data is complicated, due to “confounding by indication” and “treatment drop-ins”.
In the first talk, we will motivate the need for adopting a causal framework when constructing such counterfactual (“what-if”) prediction models, while in the second half we will introduce conformal inference, as a way to construct prediction intervals around our counterfactual predictions.
The talk on the 17th of March can be thought of as a primer (or gentle introduction) to the topic, before the research-focus talk on the 24th of March on using conformal prediction for counterfactuals and individual treatment effects.
- Dr Karla Diaz Ordaz (LSHTM)
- Dr James Burns (LSHTM)
Please note that the time listed is Greenwich Mean Time (GMT)