Which Risk Score Best Predicts Prolonged Air Leak After Lobectomy? A Comparative Evaluation Study

Area of Science:

• Thoracic Surgery
• Pulmonary Medicine
• Surgical Outcomes Research

Background:

• Prolonged air leak (PAL) is a frequent complication after pulmonary resection, leading to increased patient morbidity, longer hospital stays, and higher healthcare costs.
• Existing scoring systems aim to predict PAL, but their clinical validation and comparative performance require further investigation.

Purpose of the Study:

• To compare the predictive performance of three established scoring systems—the Epithor score, Gilbert score, and prolonged air leak score (PALS)—for PAL in patients undergoing video-assisted thoracoscopic surgery (VATS) lobectomy.
• To identify independent predictors of PAL within this specific surgical cohort.

Main Methods:

• A retrospective analysis of 534 patients who underwent VATS single-lobe lobectomy for primary lung cancer between 2012 and 2021.
• PAL was defined as an air leak persisting for more than 5 days postoperatively.
• Model performance was evaluated using discrimination metrics (AUC, calibration plots, Hosmer-Lemeshow test, Brier score) and clinical utility assessed via decision curve analysis (DCA).

Main Results:

• The incidence of PAL was 9.9% (53 patients). Independent risk factors identified included male sex, body mass index <25.5 kg/m², and the presence of pleural adhesions.
• The Epithor score demonstrated the highest discrimination (AUC = 0.735), while PALS showed the most accurate calibration.
• Decision curve analysis indicated that Epithor offered the greatest net benefit at low probability thresholds, Gilbert at mid-range thresholds, and PALS across typical clinical ranges.

Conclusions:

• In this single-center cohort, the Epithor score exhibited the best overall predictive performance for PAL following VATS lobectomy.
• PALS provided superior threshold-specific calibration and clinical utility.
• Combining the Epithor and PALS may enhance individualized risk stratification for PAL, though definitive superiority of any single model was not established.