The following is a summary of the “Machine-Learning Model for Mortality Prediction in Patients With Community-Acquired Pneumonia: Development and Validation Study,” published in the January 2023 issue of Chest by Cilloniz, et al.

Machine learning (ML) and other forms of artificial intelligence are being looked at as viable ways to improve the prediction capacity of existing clinical tools, such as prognostic scores. However, there needs to be more research assessing the performance of ML techniques in improving the predictive capacity of existing scores for community-acquired pneumonia (CAP). This study aimed to implement and verify a causal probabilistic network (CPN) model for predicting mortality in patients with CAP. This research involved a retrospective derivation-validation analysis at 2 hospitals affiliated with Spanish universities. 

In this study, they compared the predictive ability of the Sequential Organ Failure Assessment (SOFA), the Pneumonia Severity Index (PSI), the quick Sequential Organ Failure Assessment (qSOFA), and the CURB-65 criteria (confusion, urea, respiratory rate, BP, age 65 years) to a CPN adapted for CAP (SeF-ML) that was originally developed to predict mortality in sepsis. It’s important to note that the SeF models are proprietary. The DeLong technique for correlated receiver operating characteristic curves was used to evaluate the dissimilarities between the curves.

There were 4,531 patients in the derivation cohort and 1,034 in the validation cohort. AUCs for 30-day mortality prediction using SeF-ML, CURB-65, SOFA, PSI, and qSOFA were 0.801, 0.759, 0.671, 0.799, and 0.642 in the derivation cohort, respectively. SeF-AUC ML’s was 0.826 in the validation study, which matched its AUC (0.801) in the training data (P =.51). SeF-AUC ML’s was 0.764, which was significantly higher than CURB-65’s (0.764, P =.03) and qSOFA’s (0.729, P =.005). However, it was not significantly different from PSI (0.830; P =.92) or SOFA (0.771; P =.14). Using structured health data, SeF-ML shows promise in enhancing patient mortality prediction with CAP. To strengthen generalizability, more studies using external validation are needed.

Source: sciencedirect.com/science/article/abs/pii/S0012369222012430