Accurate Prediction of Survival Based on Kaplan–Meier Analytics

This study integrates Kaplan–Meier survival analysis with the Stochastic and Augmented Interpretable Health Analytics (SAIHA) framework to model long-term survival in pancreatic cancer, a malignancy characterized by late diagnosis, rapid progression, and poor prognosis. The Kaplan–Meier estimator was first employed to nonparametrically characterize empirical survival probabilities across the observed follow-up period, capturing censoring patterns and short-term mortality dynamics without imposing distributional assumptions. This step provided a transparent baseline representation of survival up to approximately four years post-diagnosis, where empirical data density remains sufficient for reliable estimation. To address the limitations of traditional Kaplan–Meier analysis in extrapolating beyond observed follow-up, the SAIHA framework was then applied using a Weibull survival model to propagate uncertainty, incorporate population heterogeneity, and generate probabilistic survival projections into the long-term horizon. The Weibull distribution was selected for its flexibility in modeling monotonic hazard functions commonly observed in aggressive cancers and for its interpretability within clinical contexts. Parameter uncertainty was explicitly modeled to reflect variability in disease progression and treatment response across patients. The combined model predicts a pronounced decline in survival beyond year four, with the most likely five-year survival probability estimated near 3% and a median six-year survival approaching 1.5%. These projections align with known epidemiological patterns of pancreatic cancer and underscore the persistent lethality of the disease despite advances in therapy. Importantly, the SAIHA framework provides full survival distributions rather than point estimates, enabling clinicians and researchers to assess uncertainty bounds and tail risks associated with long-term outcomes. Overall, the integrated Kaplan–Meier–SAIHA approach extends classical survival analysis by combining empirical rigor with stochastic, distribution-aware forecasting. This methodology offers a robust and interpretable framework for high-risk clinical prediction, supporting more informed decision-making in oncology research, population health modeling, and precision medicine applications.