Decoding Patient Heterogeneity Influencing Radiation-Induced Brain Necrosis

Area of Science:

• Radiation Oncology
• Neuro-oncology
• Medical Physics

Background:

• Patient heterogeneity in radiotherapy (RT) for brain tumors obscures treatment effects and complicates the prediction and mitigation of radiation-induced brain necrosis.
• Understanding patient-specific factors is crucial for improving outcome assessments and reducing treatment toxicity in brain tumor patients undergoing RT.

Purpose of the Study:

• To develop and validate a clinically practical pipeline for identifying key variables that clarify the relationship between dosimetric features and outcomes in patients treated with proton therapy.
• To assess the impact of non-dosimetric variables on radiation-induced brain necrosis risk and improve patient stratification for personalized RT.

Main Methods:

• A cohort of 130 patients treated with proton therapy for brain and head and neck tumors was analyzed.
• An expert-augmented Bayesian network was utilized to understand variable interdependencies and assess structural dependencies, with critical evaluation using a three-level grading system.
• Markov blanket analysis, log-likelihood ratio, integrated discrimination index, net reclassification index, and receiver operating characteristic (ROC) curves were employed for statistical assessment.

Main Results:

• Tumor location and proximity to critical structures (white matter, ventricles) were identified as major determinants of necrosis risk.
• Quantitative measures confirmed the clinical significance of these non-dosimetric variables in patient stratification (log-likelihood ratio = 12.17; P = 0.016; integrated discrimination index = 0.15; net reclassification index = 0.74).
• The ROC curve area of 0.66 highlighted the discriminative value of non-dosimetric variables in predicting brain necrosis.

Conclusions:

• Key patient variables, particularly non-dosimetric factors, critical to understanding brain necrosis post-RT were identified, serving as confounders and moderators of dosimetric impacts.
• The developed pipeline enhances outcome assessments by identifying at-risk patients, offering a versatile tool for broader applications in RT.
• This approach aims to improve treatment personalization for various disease sites by revealing critical patient variables influencing RT outcomes.