Patient-Centric Approach for Utilising Machine Learning to Predict Health-Related Quality of Life Changes During Chemotherapy

Patients undergoing chemotherapy often experience adverse effects, which can lead to changes in health-related quality of life (HRQOL) and have detrimental effects on patients’ physical and psychological wellbeing. This study aims to apply machine learning (ML) models to patient-reported, clinical, and demographic data to predict changes in physical well-being, social functioning, role functioning, usual activities, and mobility at 6, 12 and 18 weeks from starting chemotherapy. A patient-centric approach is followed as outcome variables were selected after consultation with patients and a clinician, who also was involved in the study design. Logistic regression, random forest, extreme gradient boosting, and multilayer perceptron were developed and their performance of predicting improvement and deterioration in HRQOL was evaluated with accuracy, recall, specificity, and area under the ROC curve (AUC). Model performance was generally better when predicting improvement, with best models giving AUC of 0.904 for predicting mobility improvement at 12 weeks and AUC of 0.898 for predicting usual activities improvement at 18 weeks. The results encourage involving stakeholders in research and support the view that ML can be used to predict outcomes meaningful to patients. They also highlight that although some outcome variables can be valuable for patients, they may not be predicted well by ML models. This study can inform future work on patient-centric ML methods contributing to treatment decisions in oncology.