Aim: To assess the independent determinants of tumor-induced mortality in different age subgroups after considering competing risk (CR).
Methods: Data were extracted from the SEER database. The independent determinants of tumor-induced mortality were defined by CR analysis and validated by conditional inference trees. A CR nomogram was created based on the proportional subdistribution hazard model.
Results: The different age subgroups had their own independent determinants of tumor-induced mortality. Using these variables, a CR nomogram was built with good discrimination and calibration.
Conclusion: When conducting population-based cohort studies, a CR analysis is recommended for cancers with short survival and high mortality. A CR nomogram represents the first attempt at a predictive model for quantifying tumor-induced mortality.
Keywords: competing risk; machine learning; nomogram; primary glioblastoma; survival.