Predicting response to neuromodulators in refractory chronic cough: An interpretable machine learning approach

BackgroundNeuromodulators (e.g., Gabapentin, Baclofen, and Flupentixol-melitracen) are widely prescribed for refractory chronic cough (RCC); however, heterogeneous treatment responses remain a significant challenge for optimal agent selection.ObjectivesThis study aimed to develop an interpretable machine learning (ML) model to predict individual patient responses to these therapies.DesignSingle-center retrospective observational study.MethodsThis retrospective cohort study included 730 patients with RCC treated with neuromodulators from 2016 to 2023. We extracted comprehensive clinical data, including demographics, patient-reported outcomes, medication history, and laboratory findings. Six ML algorithms underwent rigorous hyperparameter tuning and 10-fold cross-validation. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration and decision curve analysis (DCA). The optimal model was interpreted using SHapley Additive exPlanations (SHAP).ResultsThe overall neuromodulator response rate was 56.8%. A 14-variable Random Forest (RF) model demonstrated the best performance, yielding a training AUC of 0.825 (95% CI: 0.785-0.865; calibration slope: 0.984) and a robust cross-validated AUC of 0.718 (95% CI: 0.655-0.765; calibration slope: 0.956). SHAP analysis demonstrated that pre-treatment cough severity, history of esophagitis, and specific gastroesophageal reflux-related symptoms scores were identified as the top predictive features associated with treatment outcomes. The free online clinical decision-support application was deployed.ConclusionWe developed and internally validated an interpretable ML model capable of predicting neuromodulator efficacy in RCC. The integration of SHAP clarifies feature attributions to support individualized, data-driven treatment selection.