Causal machine learning reveals age-dependent radiation dose effects on mandibular osteoradionecrosis

TL;DR

This study uses causal machine learning to identify how radiation dose effects on mandibular osteoradionecrosis vary with age, enabling personalized treatment planning to reduce risk.

Contribution

It introduces a causal inference framework applying generalized random forests to clinical radiotherapy data, revealing age-dependent dose-response effects on ORN risk.

Findings

All dosimetric factors have significant positive causal effects on ORN.

Patients aged 50-60 show the strongest dose-response relationships.

Patients over 70 have minimal causal effects.

Abstract

Distinguishing causal relationships from statistical correlations remains a fundamental challenge in clinical research, limiting the translation of observational findings into interventional treatment guidelines. Here we apply causal machine learning to establish causal effects of radiation dose parameters on mandibular osteoradionecrosis (ORN) in 931 head and neck cancer patients treated with volumetric-modulated arc therapy. Using generalized random forests, we demonstrate that all examined dosimetric factors exhibit significant positive causal effects on ORN development (average treatment effects: 0.092-0.141). Integration with explainable machine learning reveals substantial treatment effect heterogeneity, with patients aged 50-60 years showing the strongest causal dose-response relationships (conditional average treatment effects up to 0.229), while patients over 70 years demonstrate minimal effects. These results suggest that age-stratified treatment optimization and personalized treatment planning for the dosimetric factors could reduce ORN risk. Our findings demonstrate that causal inference methods can transform clinical retrospective radiotherapy data into personalized treatment recommendations, providing a methodological framework applicable to toxicity prediction across oncology and other clinical domains where treatment decisions depend on complex dose-response relationships.