A Deep Learning-Based Prediction Model for Mortality in Severe COVID-19 Patients: A Multi-Center Registry Analysis in Korea

Résumé fourni par la source

Introduction Coronavirus disease 2019 (COVID-19) has caused numerous deaths globally. While artificial intelligence (AI) models improve mortality prediction, research on critically ill patients is limited. For practical use, model requires minimal, accessible data for rapid assessment. Objective This study aimed to develop a deep learning model to predict mortality among severe COVID-19 patients requiring intensive care unit (ICU) for rapid risk assessment. Methods This retrospective study analyzed 1,080 critically ill COVID-19 patients from 22 centers in Korea who were admitted to the ICU between January 1, 2020, and August 31, 2021. We use deep learning-based tabular model and feature reduction strategies for developing model. Results Among 1,080 patients, 418 (38.7%) survived and 662 (61.3%) died. Non-survivors were older, had worse clinical and chest X-ray (CXR), lower respiratory rate-oxygenation (ROX) index, and required more vasopressors and continuous renal replacement therapy (all p < 0.01). An initial 67-variable model achieved an area under the curve (AUC) of 87.7, which was reduced to 7 variables (Age, C-reactive protein, lactate dehydrogenase, pro-B-type natriuretic peptide, CXR, and ROX index) using stepwise method, resulting in a minimized model with an AUC of 82.1%. Conclusion We developed a practical AI model for clinical decision support in critically ill COVID-19 patients. erj;66/suppl_69/PA1000/F1 F1 F1

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.