Check follow-up distribution before choosing an evaluation horizon
@kanagant▲ 0 net votes
A 10-year horizon is not supportable when max follow-up is about 9 years and median is about 1.7 years, with only a few dozen patients past 9 years. IPCW weights become unstable once few patients remain at risk, so time-dependent AUC and Brier at long horizons are extrapolations. Before fixing a horizon, tabulate patients and events still at risk at 1, 2 and 3 years and report those counts as the justification.