Predictive capacity of competing risks models

L. Garmendia Bergés, I. Barrio, G. Gómez Melis

Competing risks situations appear often in survival analysis when the endpoint of interest (i.e., recovery) is precluded by another event (i.e, death). The probability of failure in the presence of competing risks can be modeled by means of cause-specific hazards or relying on the incidence function. These models can be as well used to predict the course of future individuals and in this case their predictive capacity has to be analyzed. The area under the time-dependent ROC curve (AUC(t)) is commonly used to quantify the ability of a survival model to correctly predict future events (non-events) at a time t. The objective of this work is to propose a global concordance measure to quantify the predictive capacity of a competing risks model by means of the partial AUC(t)s obtained in each transition. With that aim we have studied different estimators of the AUC(t).

Keywords: Competing risks, AUC, concordance measure

Statistical Models
November 7, 2023  11:40 AM
CC3: Room 1

D. Nieto Barba, I. Montes Gutiérrez, E. Miranda Menéndez

V. Cueva López, M. J. Olmo Jiménez, J. Rodríguez Avi

C. Sebastián Martínez-Cava, C. González Guillén

B. Pérez-Sánchez, C. Perea, G. Durán, J. J. López-Espín