Innovations in biomarker stratification for precision oncology

Biomarker stratification underpins precision oncology, yet survival analysis often relies on arbitrary thresholds that undermine reproducibility and clinical relevance, particularly for continuous biomarkers. This review focuses on methodological approaches for stratifying continuous biomarkers within survival analysis frameworks, examining conventional strategies alongside data-driven and machine learning methods in the context of threshold selection and clinical interpretability. We evaluate the extent to which these approaches address key challenges including heterogeneity, confounding, and overfitting, and critically appraise their strengths and limitations for clinically actionable risk stratification. By synthesising current evidence, we highlight opportunities for more robust and reproducible prognostic modelling and outline future directions to improve the reliability of biomarker-driven decision-making in oncology.