SCSRF: A Stable Clinical Survival Risk Framework for PBC Survival-Risk Prediction

Predicting likelihood of death in primary biliary cholangitis (PBC) is unreliable must be handled with great care for small samples, censoring, model selection, etc. Predictor stability this study propose the following Stable Clinical Survival Risk Framework. The model was built using (SCSRF), which combines the local feature engineering technique of folding with the cross-validation technique. Evaluation that takes censorship into account, and stability analysis. Among three learners, al
though the random survival forest’s mean C-index value (0.855) was the highest, the differences between the pairs were not significant post Holm correction. Its pooled out-of-fold C-index was 0.832 (95% CI: 0.800–0.862), with an IPCW C-index AUC
of 0.876, time-dependent AUC of 0.788 and integrated Brier score of 0.117. Even with just 15 features, the model’s ability to predict survival risk was similar to using all the features (C index = 0.830). The curves of the out-of-fold risk groups were
similar when plotted with the Kaplan–Meier method. There was consistent separation of risk groups across three repeats (median log-rank p = 2.42 × 10−28). Internal transportability analysis yielded a C-index of 0.786. These findings indicate that SCSRF is a a framework for survival-risk stratification based on reliability; But it must be independently externally validated to make it generalisability and clinical usefulness.