Development and validation of an interpretable model for pre-chemotherapy triage of fatigue-pain-sleep disturbance burden in lung cancer.

PURPOSE: The aim of the study is to develop an interpretable machine learning model that can effectively identify patients at risk of high fatigue-pain-sleep disturbance (FPS) burden before chemotherapy. METHODS: A two-centre observational study was conducted in two tertiary hospitals. Candidate predictors were assessed approximately two weeks before chemotherapy initiation, whereas fatigue, pain, and sleep disturbance outcomes were assessed after chemotherapy completion. Pre-chemotherapy demographic, clinical, laboratory, and patient-reported variables were collected, and latent profile analysis (LPA) of post-chemotherapy FPS identified symptom-burden subgroups. Machine-learning models were trained in one hospital-based development cohort and validated in an independent institutional cohort from a second hospital; performance was assessed with discrimination, calibration, and clinical-utility metrics, and the best model was interpreted using Shapley Additive Explanations (SHAP). RESULTS: A total of 1444 patients were included for model development and independent institutional validation. The macro area under the receiver operating characteristic curve (ROC-AUC) of four candidate models in institutional validation ranged from 0.761 to 0.821. The Multi-XGBoost model showed the best discrimination in the independent institutional validation cohort (micro AUC = 0.853; macro AUC = 0.821), with classification results summarised by the confusion matrix. CONCLUSION: The interpretable machine-learning model developed and evaluated through independent institutional validation demonstrated good predictive performance. This model may help oncology nurses and clinicians identify high-risk patients before chemotherapy and support earlier, risk-stratified symptom-management planning.