Background: Pneumonia is a common infectious disease within communities and is a leading cause of mortality worldwide. This study aimed to develop a new simple and effective prognostic score using artificial intelligence (AI)-based chest radiograph (CXR) results to predict the outcomes of community-acquired pneumonia (CAP) and healthcare-associated pneumonia (HCAP).

Methods: Patients aged >18 years, admitted to hospital for the treatment of pneumonia between March 2020 and August 2021 were included. Results from a commercially available AI-based lesion detection software were used to develop a new prognostic score containing AI-based CXR results

Results: A total of 498 patients (median 79 years; 63.6% male), including 283 and 215 patients in training and test sets, respectively, were included. The median initial CURB-65 score and PSI were 2 and 120, respectively. The median hospital stay was 12 days and overall 28-day mortality was 19.7%. In the training and test sets, models that combined AI consolidation scores with the CURB-65 scores or PSI showed significantly higher C-indices compared with those using CURB-65 scores or PSI alone (p?0.001) for both CAP and HCAP. The C-index of the new prognostic score composed of the CURB-65 score or PSI, initial O2 requirement, AI consolidation score, and nursing home resident was significantly higher than the CURB-65 score for both types of pneumonia (0.7 vs 0.6, p?0.031).

Conclusions: A new simple prognostic score, including AI-based CXR results, in addition to conventional clinical scores, could provide an effective model for predicting pneumonia outcomes in patients with CAP and HCAP.