Figure 2.

Evaluation of XGB-RF predictive performance under increasingly stringent validation settings. We consider single clinic and all-clinics cross-validation, where models are trained and evaluated in a cross-validation framework using data from either a single clinic or all clinics combined, with results from the latter stratified across the three datasets. As multiple models are constructed, we show the distribution of these results. Finally, external validation, shows the result of training a model on two cohorts and evaluating on the remaining cohort. These results are a single score for a single model and hence are shown as a horizontal line.