Fig. 3
From: Dynamic updating of clinical survival prediction models in a changing environment
Cohort with replacement simulation results for a scenario with calibration drift. The left graphic shows the average C-index for each updating method across the 600 simulated datasets at each of the 5 prediction times for a scenario where the event rate decreased over time from 5% per year in Q1 to 2% per year in Q5. On the right, the average calibration intercept is shown for the same scenario. Results for ‘Recal once’ and ‘Refit once’ strategies are ordered by update time with the earliest time on the left