Predicting severe community-acquired pneumonia in adults: a machine learning approach using C-reactive protein to albumin ratio.

Background: Community-acquired pneumonia (CAP) can rapidly progress to severe disease, markedly increasing mortality. This study aimed to investigate the value of C-reactive protein to albumin ratio (CAR) and its incorporation into machine learning models for predicting the progression of CAP to severe pneumonia in adults. These findings may provide new insights and a new tool for early risk assessment and clinical decision-making for patients with CAP. Methods: Clinical data of 117 patients diagnosed with CAP and hospitalized in our institution between June 2023 and April 2024 were used. The patients were divided into severe and non-severe CAP groups based on discharge diagnoses. Univariate analysis and random forest (RF) modeling were employed to select feature. For inclusion in five machine learning models logistic regression (LR), multilayer perceptron (MLP), support vector machine (SVM), extreme gradient boosting (XGBoost), and RF. The predictive performance of these models was evaluated using the accuracy, F1-score, and area under the receiver operating characteristic curve (AUC). Results: Based on univariate analysis and RF CAR, age, and neutrophil count were selected for model inclusion. All models demonstrated good predictive performance with accuracy exceeding 78%. The MLP model performed best across all three evaluation metrics (accuracy =85.11%, F1-score =0.70, AUC =0.89). The CAR contributed most to the predictive model. Conclusions: The CAR, along with the age and neutrophil count, effectively predicts the severe pneumonia risk in patients with CAP, and the MLP model enhances prediction accuracy for early risk assessment and clinical decision-making.