Machine learning-based prognostic prediction model of pneumonia-associated acute respiratory distress syndrome

Jing LvJuan ChenMeijun LiuXue DaiWang Deng^*

• Second Affiliated Hospital, Chongqing Medical University, Chongqing, China

Objective This study aimed to construct a machine learning predictive model for prognostic analysis of patients with p-ARDS. Methods In this single-center retrospective study, 230 patients with p-ARDS admitted to the RICU of the second affiliated hospital of Chongqing Medical University from January 2020 to November 2024 were included. Patients were divided into survival group and death group according to the 28-day prognosis results. 20% of the total samples were randomly selected as the test set, and the remaining samples were used as the training set for crossvalidation, and six different models were constructed, including Logistic Regression, Random Forest, NaiveBayes, SVM, XGBoost and Adaboost. The AUC value, AP value, accuracy, sensitivity, specificity, Brier score, and F 1 score were used to evaluate the performance of the models and pick the optimal model. Finally, the SHAP feature importance map was drawn to explain the optimal model. Results 10 key variables, namely LAR, Lac, pH, age, PO2/FiO2, ALB, BMI, TP, PT, DBIL were screened using the filtration method. Among the six algorithms, the performance of the SVM algorithm is significantly better than that of other algorithms. The AUC, AP, Brier Score, and F1 Scores in the test set were 0.77, 0.67, 0.19, and 0.60, respectively. Conclusion This study developed and visualized a machine learning model constructed based on 10 common clinical features for predicting 28-day mortality in patients with p-ARDS. The model shows good predictive performance and achieves explanatory analysis in combination with SHAP and LIME methods, providing a reliable mortality risk assessment tool for p-ARDS.