Multimodal AI-based 28-day mortality prediction of pneumonia patients at ED discharge: a multicenter study

Abstract

This study develops and evaluates an artificial intelligence (AI)-driven model to predict the 28-day mortality in patients with pneumonia by integrating AI-interpreted chest radiographs (CXR) and clinical data available at the time of emergency department (ED) disposition. This multicenter retrospective study included patients who visited the ED with pneumonia at a tertiary academic hospital in South Korea, as well as recorded in the Medical Information Mart for Intensive Care (MIMIC-IV, v3.1) database during study periods. To compare AI-driven models with a traditional clinical scoring system, three survival prediction models were developed using a baseline CURB-65 score. Five variable sets were constructed by combining the CURB-65 score, AI-interpreted CXR findings, and additional clinical information. A total of 2,874 ED visits were analyzed. The random survival forest (RSF) model using the all-feature set (CURB-65, CXR interpretation, and clinical information) achieved a concordance index (C-index) of 0.872 (95% confidence interval [CI]: 0.861–0.886) in the test set, significantly outperforming the RSF model excluding the CXR interpretation information, which had a C-index of 0.865 (95% CI: 0.854–0.879). This study highlights the potential utility of a multimodal AI-driven prediction model to support prognosis estimation and clinical decision-making for patients with pneumonia in ED.

Ähnliche Arbeiten