Development and Temporal Validation of Staged Prediction Models for Acute Respiratory Distress Syndrome After Thymic...

Background/Objectives: To develop and temporally validate staged models for predicting acute respiratory distress syndrome (ARDS) after thymic epithelial tumour resection. Methods: This single-centre retrospective study included 596 adults with pathologically confirmed thymic epithelial tumours. Patients treated in 2005-2020 formed the development cohort (n = 395), and those treated in 2021-2026 formed the temporal validation cohort (n = 201). Postoperative ARDS was adjudicated using the 2024 global definition. A preoperative model and a postoperative landmark model incorporating intraoperative, pathological, and early postoperative variables were developed using least absolute shrinkage and selection operator and multivariable logistic regression. Performance was assessed using discrimination, calibration, Brier score, decision curve analysis, bootstrap internal validation, and temporal validation. Results: ARDS occurred in 116 development patients (29.4%) and 58 validation patients (28.9%). The postoperative landmark model included the preoperative linear predictor, one-lung ventilation duration, Masaoka-Koga stage, perioperative transfusion, and early postoperative glucose, albumin, and C-reactive protein. It achieved higher AUCs than the preoperative model in the development (0.908 vs. 0.693; p < 0.001) and validation cohorts (0.928 vs. 0.655; p < 0.001), with lower Brier scores (0.107 vs. 0.187 and 0.094 vs. 0.192, respectively; both p < 0.001). The optimism-corrected development AUC was 0.898. Calibration and decision curve analyses also favoured the postoperative landmark model. Conclusions: Incorporating perioperative information improved ARDS prediction after thymic epithelial tumour surgery. The staged models may support sequential perioperative risk assessment; however, prospective evaluation and independent multicentre validation are required before implementation.