Fig. 2
From: Accounting for treatment use when validating a prognostic model: a simulation study
a-d: Risk distributions in two simulated validation sets, before and after applying different approaches to correct for treatment use. 50% of individuals received an effective treatment (relative odds reduction on treatment: 0.5) (see Table 2 scenarios 2 and 1, respectively, for details). a the model was validated on the combined treatment and control group of a randomised trial. b-d the model was validated using data from a non-randomised setting where the probability of receiving treatment depended on an individual’s (untreated) outcome risk. Solid black lines represent the observed risks in the validation set after treatment. Dashed black lines represent the risks observed after applying correction methods to the data: a-b the exclusion of treated individuals, c IPW, d IPW followed by the exclusion of treated individuals. Grey lines represent the risks of the same individuals had they remained untreated