Abstract

BACKGROUND

Artificial intelligence (AI) may support multidisciplinary NSCLC decision-making

AIMS

To develop AI to predict postoperative complications (POC) after lung resection for NSCLC

METHODS

Retrospective review of clinical records of lung resection for NSCLC in 2020-2021

Multivariate analyses defined statistically significant POC predictors; logistic regression model was developed. Patient files were allocated to training set and test set. AI was taught to predict POC occurrence on training set; model was validated on test set

RESULTS

1552 consecutive patients included (M/F ratio 0.98, median age 64 [19-87]): 545 lobectomies, 18 bilobectomies, 32 pneumonectomies, 492 segmentectomies

COPD (p=0.035), FEV1% (p=0.043), smoking pack/year (p=0.047) significantly associated with POC occurrence

AI was trained and validated to predict POC with all identified variables (training set=981, test set=571), achieving 89% sensitivity/96% specificity

CONCLUSIONS

AI was able to predict, with adequate accuracy, POC in a retrospective dataset

AI may support multidisciplinary management by identifying patients requiring specific perioperative care, improving surgical planning

However, large-scale multicentre investigations are necessary to substantiate our findings