An interpretable stacked ensemble framework for evaluating respiratory rehabilitation outcomes under traditional...

Evaluating treatment effectiveness in patients undergoing respiratory rehabilitation integrated with Traditional Chinese Medicine (TCM) remains challenging because recovery trajectories are clinically heterogeneous, interventions are multimodal, and treatment effects may be nonlinear. Conventional assessment strategies often rely on score changes or threshold-based criteria and therefore provide limited support for individualized evaluation and clinically grounded interpretation.

This multicenter retrospective cohort study included 1,483 patients who received respiratory rehabilitation combined with TCM-based interventions. Predictor variables covered demographic and clinical characteristics, disease severity, rehabilitation prescription and implementation intensity, TCM intervention features, laboratory and physiological indicators, and baseline functional status. Data were randomly divided into training and test sets at a ratio of 7:3 for internal validation. Within the training set, feature selection was performed using the least absolute shrinkage and selection operator (LASSO), and the selected variables were entered into a stacked ensemble framework. Model performance was evaluated using discrimination and calibration metrics. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) to quantify both global and individual-level feature contributions.

The stacked ensemble model achieved the best predictive performance in the independent test set, with an area under the receiver operating characteristic curve of 0.903, outperforming all single learners. SHAP analysis indicated that baseline functional capacity and symptom burden were dominant predictors of rehabilitation outcomes. Measures reflecting TCM intervention intensity, particularly exercise frequency and duration of herbal exposure, also contributed substantially to prediction and showed nonlinear dose-response patterns. These findings suggest that TCM-related treatment characteristics provided additional clinically interpretable information beyond baseline patient status alone.

The proposed interpretable stacked ensemble framework provides a transparent approach for evaluating respiratory rehabilitation outcomes under TCM-integrated care. By identifying nonlinear effects and patient-specific drivers of treatment response, the model may support more individualized outcome assessment and more informed clinical decision-making. Further validation across independent external cohorts is warranted before broader clinical implementation.