Time-to-Next-Treatment (TTNT) is a critical real-world oncology endpoint that reflects treatment durability, disease progression, and clinical decision-making beyond controlled clinical trial settings. Predicting TTNT accurately enables pharmaceutical and biotech organizations to better understand real-world treatment performance, improve patient stratification, and support data-driven medical and commercial decision-making. This presentation introduces a survival-informed machine learning framework for predicting TTNT using real-world–inspired oncology data.

A patient-level dataset was developed to emulate longitudinal oncology treatment journeys, incorporating demographic attributes, clinical indicators, treatment history, response signals, and healthcare utilization patterns commonly observed in real-world evidence (RWE) sources. TTNT was operationalized as a predictive outcome to support scalable analytics deployment while preserving alignment with survival-based clinical reasoning. Multiple supervised machine learning models were evaluated, including Logistic Regression, Random Forest, Support Vector Machine, Gradient Boosting, and K-Nearest Neighbors.

Model performance was assessed using classification accuracy to ensure consistency and comparability across approaches. Logistic Regression, Random Forest, and Support Vector Machine each achieved an accuracy of 0.916, with Gradient Boosting and K-Nearest Neighbors achieving accuracies of 0.911 and 0.909, respectively. These results demonstrate that both linear and non-linear models can effectively capture patterns associated with treatment transitions when supported by well-engineered clinical and utilization features. Ensemble and kernel-based methods showed particular robustness in modeling complex interactions inherent in oncology care pathways.

Beyond predictive performance, the framework emphasizes interpretability, scalability, and operational feasibility—key considerations for enterprise adoption. Feature contribution analysis highlights clinically intuitive drivers of treatment change, supporting transparency and cross-functional trust among analytics, medical, and commercial stakeholders. The proposed approach enables proactive identification of patients at higher risk of early treatment change and supports longitudinal monitoring of treatment effectiveness.