Predicthor: AI-Powered Predictive Risk Model for 30-Day Mortality and 30-Day Complications in Patients Undergoing Thoracic Surgery for Lung Cancer

Abstract

Objective: To assess the predictive performance of Predicthor, an artificial intelligence model, for 30-day mortality and complications following major pulmonary resections. Background: The significance of predicting postoperative complications in thoracic surgery lies in the impact on patient outcomes and the efficient allocation of healthcare resources. The longstanding use of the Thoracoscore for over 15 years in hospital settings emphasizes the opportune moment for an update, leveraging new artificial intelligence methodologies to enhance predictive precision and relevance. Methods: The EPITHOR French population-based database linked to the National Institute of Statistics and Economic Studies database has been queried from January 1, 2016, through December 31, 2022, on 6 selected hospital centers (Rouen, Dijon and Toulouse CHUs, Strasbourg CHRU, Centre Hospitalier Général de Bayonne, and Assitance Publique des Hopitaux de Marseille) with curated data collection. A total of 6508 patients who have undergone primary lung cancer surgery via lobectomy or bilobectomy, aged over 18 years, and with anAmerican Society of Anesthesiologists (ASA) physical status classification system score under 4, were selected. In a retrospective analysis using a 3-dataset scheme (training cohort, internal and external validation on 118 other centers), we assessed the predictive performance of Predicthor for 30-day complications and mortality following major pulmonary resections. Results: < 1E-16). Conclusions: Predicthor's predictive performance for 30-day mortality and complications highlighted its potential as a valuable tool in clinical decision-making. The study's methodology and comprehensive dataset contribute to its relevance in using machine learning on large available databases for shaping thoracic surgery practices and patient management.

Ähnliche Arbeiten