Predictive Modeling in Precision Oncology

Abstract

Precision oncology aims to match each patient to the therapy most likely to benefit them on the basis of tumour molecular profile, clinical features, and treatment history, yet the predictive models that underpin this matching remain inconsistent in their clinical deployment. We evaluated 232 predictive modelling programmes applied to precision oncology across cancer centres in Spain, Switzerland, and Germany between 2015 and 2024, spanning five modelling domains: treatment response prediction, survival outcome modelling, biomarker discovery pipelines, tumour subtype classification, and drug combination optimisation. A Precision Oncology Modelling Index (POMI) was constructed from five sub-scores-- predictive discrimination, biological plausibility, external validation breadth, clinical workflow integration, and prospective trial evidence -- with weights from regression against sustained adoption into molecular tumour board decision-making. POMI correlated with adoption at r = +0.84 and discriminated adopted from non-adopted models with an AUC of 0.886. Treatment response prediction models scored highest (mean POMI 0.824), while drug combination optimisation trailed at 0.598. Only 35.8 percent of programmes exceeded the 0.75 threshold. Predictive discrimination carried the largest regression weight (beta = +0.278), followed by external validation breadth (beta = +0.230). These findings indicate that precision oncology models should prioritise multi-centre validation and clinician-interpretable outputs to accelerate adoption into routine treatment planning.