Machine learning-based prediction model for bronchoalveolar lavage efficacy in severe pneumonia

Back to news list

Source: Frontiers Medicine

Original: https://www.frontiersin.org/articles/10.3389/fmed.2025.1710355...

Published: 2026-01-07T00:00:00Z

The study constructed and validated a machine learning-based predictive model for the clinical efficacy of bronchoalveolar lavage (BAL) in 206 patients with severe community-acquired pneumonia. The patients were divided into a training set (n=144) and a validation set (n=62) in a ratio of 7:3. Independent risk factors for low efficacy were age ≥60 years, chronic obstructive pulmonary disease, procalcitonin ≥2 ng/mL, C-reactive protein ≥100 mg/L, PaO₂ <60 mmHg, and PaCO₂ ≥50 mmHg (all p<0.05). The Random Forest (RF) model achieved the best AUC values ​​of 0.799 in the training set and 0.778 in the validation set, outperforming the KNN and GB models. In the training set, 32 patients (22.22%) and in the validation set 13 patients (20.97%) had poor efficacy. The RF model effectively identifies key factors influencing BAL outcomes. The study is single-center and requires multicenter validation.