Opportunities for Causal Machine Learning in Precision Oncology

AbstractPrecision oncology aims to tailor treatment strategies for patients based on individual genetic, molecular, and clinical characteristics, yet current approaches in machine learning (ML) often fail to capture the full complexity of treatment response and individual treatment effects. Unlike traditional predictive ML, causal ML focuses on counterfactual reasoning, which can help guide treatment decisions by determining what outcomes would have been observed had a patient received treatment A as opposed to treatment B. Causal ML thereby offers a unique transformative opportunity for precision oncology by leveraging multimodal data - including phenomics, genomics, transcriptomics, radiomics, and pathology - to estimate individualized treatment effects. Its applications in precision oncology include identifying predictive biomarkers, optimizing personalized treatment strategies, and accelerating drug repurposing. Furthermore, by emulating randomized controlled trials, causal ML has the potential to generate robust evidence from real-world data, addressing gaps for underrepresented patient populations and rare cancers. As clinical trials increasingly integrate multimodal data sources, causal ML provides a framework to extract actionable insights, supporting precision medicine at scale. While challenges in validity, transportability, and data quality remain, causal ML represents a critical advancement toward more effective, data-driven decision-making in precision oncology.