Development of a clinical prediction model using digital device data for interstitial lung disease in patients with unresectable stage III non-small cell lung cancer treated with durvalumab.

BACKGROUND: Early detection of interstitial lung disease including radiation pneumonitis (ILD/RP) is crucial in consolidative durvalumab therapy after chemoradiotherapy (CRT) in unresectable stage III non-small cell lung cancer. We conducted a multicenter, noninterventional pilot study to develop clinical prediction models for grade ≥2 ILD/RP following durvalumab treatment by using machine learning algorithms on continuously collected automatic measurements from a wearable device and cough-counting smartphone application, as well as electronic medical record data. METHODS: The primary objective of this prospective observational study was to develop a clinical prediction model for grade ≥2 ILD/RP. Eight machine learning algorithms were applied to develop the prediction model. After performing 4-fold cross-validation, the prediction performance was evaluated using the area under the receiver operating characteristic curve (ROC-AUC). RESULTS: Of the 145 registered patients, 123 were analyzed. The median age was 68.0 years (range: 39-85), and 79.7% were male. Grade ≥2 ILD/RP occurred in 46 patients. The XGBoost model outperformed the other seven models for the primary objective, with an ROC-AUC of 0.789 (95% confidence interval [CI]: 0.724-0.854). Heart rate showed the highest SHapley Additive exPlanations value for grade ≥2 ILD/RP. The model using patient background and wearable device data (excluding cough app) showed accuracy comparable with the model using all data sources, with an ROC-AUC of 0.794 (95% CI: 0.730-0.859). CONCLUSIONS: These findings support the feasibility and acceptability of developing a clinical prediction model for grade ≥2 ILD/RP in patients undergoing durvalumab treatment after CRT.