Assessment of the 90-day mortality risk score after video-assisted thoracoscopic lobectomy in the Italian VATS Group cohort

A five-class (A-E) aggregate risk score predicting 90-day mortality after video-assisted thoracoscopic lobectomy for lung cancer, including as independent factors male sex (3 points), carbon monoxide lung diffusion capacity ≤60% (1 point), and operative time ≥150 minutes (1 point), has been recently published. This study aims to assess the effectiveness and reliability of this risk model in a large, independent cohort of patients to confirm its generalizability. From the Italian VATS Group database, we selected 2209 patients [60% males; median age 69 years (interquartile range: 63-74)] who underwent video-assisted thoracoscopic lobectomy for non-small cell lung cancer. We calculated the aggregate risk score and the corresponding class of 90-day mortality risk for each patient. The correlation between risk classes and mortality rates was tested by Spearman’s r-test. Model calibration was evaluated by the Hosmer-Lemeshow goodness-of-fit test. Class A-E 90-day mortality rates were 0.33%, 0.51%, 1.39%, 1.31%, and 2.56%, respectively. A strong uphill correlation was identified between risk classes and 90-day mortality (r=0.90; p=0.037), showing a positive correlation between increased mortality rate and classes A to E. Hosmer-Lemeshow Chi-squared value was 67.47 (p≤0.001), with overall, class D and E significantly lower 90-day mortality in our cohort than in the original one [1.04% versus 2.5% (p=0.018), 1.31% versus 5.65% (p=0.005) and 2.56% versus 18.75% (p=0.007), respectively]. Despite our data showing a positive correlation between 90-day mortality and risk classes from A to E with modest discriminatory performance, the poor calibration suggests the need for model recalibration using local data to better manage and counsel lung cancer patients eligible for video-assisted thoracoscopic lobectomy. analysis was performed using SPSS 24.0 software (IBM, Armonk, NY, USA) and R software (version 3.6.1, Action of the Toes) with standard, rcmdr, and irr packages (R Core Team, 2019. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/).