Abstract

Stratified or personalised medicine is an attempt to move beyond a `one size fits all’ approach based on comparing group-level average outcomes to improve patient-level outcomes by identifying personalised treatment recommendations (PTR). A PTR maps a set of predictive markers to a decision of whether or not to treat an individual patient. A PTR can be estimated from a weighted sum of predictive markers and the treatment effect using either regression models, inverse probability weighting (IPW), augmented IPW , or classification methods.

Once estimated, PTRs can be evaluated by testing if the expected outcome under the PTR improves on the expected outcome under an alternative policy – such as one where either every patient receives the treatment or every patient receives the control condition. Evaluating a PTR differs from the evaluation of prognostic or diagnostic models because the object of inference (whether a subject benefited from treatment) remains unobserved.

In this talk, we will describe the statistical methods for estimating a PTR . Monte-Carlo simulations are used to compare the statistical properties of the estimation methods under a range of data generating scenarios. These methods will be demonstrated with application to data from a randomised controlled trial in Chronic Fatigue Syndrome, using our new user-written Stata command ptr.