A Deep-Learning Model Using Chest Radiographs for Prediction of 30-Day Mortality in Patients With Community-Acquired Pneumonia: Development and External Validation.

Abstract

Chest radiography is an essential tool for diagnosing community-acquired pneumonia (CAP), although has an uncertain prognostic role in patients with CAP. To develop a deep-learning (DL) model to predict 30-day mortality in patients with CAP using chest radiographs from the time of diagnosis and to validate the model's performance in patients from different time period or institutions. In this retrospective study, a DL model was developed in 7105 patients from one institution from March 2013 to December 2019 (3:1:1 allocation to training, validation, and internal test sets) to predict risk of all-cause mortality within 30 days after CAP diagnosis using patients' initial chest radiograph. The DL model was evaluated in patients diagnosed with CAP during emergency department visits at the same institution as the development cohort from January 2020 to December 2020 [temporal test cohort (n=947)], and from two additional different institutions [external test cohort A (n=467), January 2020 to December 2020; external test cohort B (n=381), March 2019 to October 2021]. AUCs were compared between DL model and CURB-65 score (an established risk prediction tool). The combination of CURB-65 score and DL model was evaluated by a logistic regression model. AUC for predicting 30-day mortality was higher (P<.001) for DL model than CURB-65 score in temporal test set (0.77 vs 0.67); higher AUC for DL model than CURB-65 score was not significant (P>.05) in external test cohort A (0.80 vs 0.73) or external test cohort B (0.80 vs 0.72). In the three cohorts, DL model, in comparison with CURB-65 score, exhibited higher (P<.001) specificity (range, 61-69% vs 44-58%) at the sensitivity achieved by CURB-65 score. Combination of DL model and CURB-65 score, in comparison with CURB-65 score, yielded increase in AUC in temporal test cohort (0.77, P<.001) and external test cohort B (0.80, P=.04), and nonsignificant increase in AUC in external test cohort A (0.80, P=.16). A DL-based model using initial chest radiographs could predict 30-day mortality in patients with CAP with improved performance compared with CURB-65 score. The DL-based model may guide clinical decision-making in management of patients with CAP.