Determining the utility of clinical and dosimetric factors for the prediction of radiation induced oesophagitis and pneumonitis

Abstract

Background and Purpose: Predicting toxicity from radiotherapy (RT) is a complex problem because there are usually multiple organs at risk irradiated and protecting all these structures requires compromise. Multiple methods can be used to predict toxicity such as Lyman Kutcher Burman (LKB) modelling, logistic regression (LR) and supervised machine learning (ML). Several trials have used isotoxic RT to treat non-small cell lung cancer (NSCLC) patients, a technique that escalates and individualises RT doses to the tumour to improve local control. Dose escalation is often constrained by the dose unavoidably delivered to oesophagus and lungs during treatment. Here we model toxicity using data from the IDEAL-CRT trial. Methods: Data from 116 IDEAL-CRT patients were analysed in this study. Clinical data including sex, age, disease stage, forced expiratory volume (FEV), force vital capacity (FVC) and diffusing capacity of lung for carbon dioxide (DLCO) were collected for the trial. Dosimetric information was generated from RT datasets, including V5Gy, V20Gy, Mean dose (MD) and Equivalent uniform dose (EUD) for lung and V35Gy, V50Gy, D1cc and MD for the oesophagus. All doses were reported as equivalent dose in 2 Gy fractions corrected for overall treatment time. Uni-variable statistical analysis was performed on all metrics using LR, with p-values used to determine which metrics would be most useful for toxicity modelling. The bootstrap method was evaluate the accuracy of LR. This information was used to inform toxicity modelling with ML using the classification learner application in MATLAB v2020a. ML Models reported overall predictive accuracy, sensitivity, specificity and Area Under the Curve (AUC) from recover operator characteristics (ROC) analysis. Resulting ML models were compared with LKB analysis and multi-variable LR. Results: Uni-variable LR found a statistically significant (p

Date of Award	31 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Adam Aitkenhead (Supervisor)