An interpretable prognostic model for early unfavorable outcomes after bronchiectasis surgery

BACKGROUND: Bronchiectasis is a heterogeneous airway disorder characterized by irreversible bronchial dilatation, chronic sputum production, recurrent infections, and acute exacerbations. Although lung resection remains a valuable option for patients with localized, refractory disease despite optimized medical therapy, early postoperative recovery varies substantially among individuals. An interpretable tool for individualized risk stratification remains lacking in clinical practice. METHODS: This retrospective study enrolled patients who underwent bronchiectasis-related lung resection at our center between January 2020 and August 2025 (n = 142). The primary endpoint was a composite outcome assessed at approximately 1 month postoperatively. A favorable outcome was defined as the absence of Clavien-Dindo grade II or higher complications in conjunction with a Quality of Life Questionnaire-Bronchiectasis Respiratory Symptoms Scale (QoL-B-RSS) score > 50; all other cases were classified as unfavorable outcomes. Candidate predictors (n = 40) were prespecified before model development. Elastic Net-penalized logistic regression was employed as the primary modeling approach, based on which a nomogram was constructed. Nested cross-validation was used for hyperparameter tuning and internally unbiased performance evaluation. Out-of-fold (OOF) predicted probabilities were used to assess discrimination (area under the curve [AUC]), overall accuracy (Brier score), calibration, and clinical utility via decision curve analysis (DCA). A random forest (RF) model, using identical outer-fold partitions, was developed as a direct comparator, and a conventional multivariable logistic regression model was additionally fitted using the predictors retained in the primary model. RESULTS: The incidence of early unfavorable outcomes was 57.7% (82/142). The final Elastic Net model retained five predictors: pCO₂, left ventricular ejection fraction (LVEF), forced expiratory volume in 1 s as a percentage of predicted (FEV1%pred), maximal voluntary ventilation as a percentage of predicted (MVV%pred), and total Bronchiectasis Severity Index (BSI) score. Internal OOF validation yielded an AUC of 0.820 and a Brier score of 0.173 for the primary model, compared with an AUC of 0.831 and a Brier score of 0.179 for the RF model. Overall, the RF model did not demonstrate a meaningful performance advantage over the primary model. Calibration analysis indicated good agreement between predicted and observed probabilities, and DCA suggested potential net clinical benefit across a range of threshold probabilities. CONCLUSIONS: This internally validated Elastic Net-based nomogram showed good discrimination, acceptable calibration, and potential clinical utility for predicting early unfavorable outcomes after bronchiectasis surgery. Based on routinely available preoperative variables, the model may provide quantitative support for individualized perioperative risk stratification and early follow-up planning. Prospective external validation and clinical impact assessment are warranted to confirm its generalizability and inform future clinical implementation.