BACKGROUND
Pneumonia, a common reason for patients to be admitted to the Intensive Care Unit, was one of the leading causes of morbidity and mortality worldwide. Prediction of pneumonia mortality is essential for individualized prevention and treatment programs. However, due to their lack of interpretability, most mortality prediction models have not yet reached clinical practice.
OBJECTIVE
The aim of this study was to develop an interpretable model to predict the mortality risk for patients with pneumonia in intensive care units (ICUs) and used the Shapley Additive Explanation (SHAP) method to explain the extreme gradient boosting (XGBoost) model and explore prognostic factors for pneumonia.
METHODS
In this retrospective cohort study, we used the electronic health records of all adult patients with pneumonia with the free and open Collaborative Research Database between from 2014-2015 (eICU-CRD). The data during the first 24 hours of each ICU admission was extracted. The data set was randomly divided, with 70% used for model training and 30% used for model validation. The prediction model was developed based on XGBoost and natural language processing. F1 score, area under the receiver operating characteristicWe included 10,962 patients with pneumonia, and the in-hospital mortality was 16.33% In this study, the XGBoost model showed a better performance to predict the mortality of pneumonia with an area under the curve (AUC) of 0.734. Compared with the traditional Apache scoring system (0.677), this model has made significant progress in its results. The SHAP m0ethod reveals the top 20 predictors of pneumonia according to the importance ranking, and the average of the blood urea nitrogen was recognized as the most important predictor variable. curve (AUC) were used as the main criteria to evaluate model performance. We used the SHAP method to explain the XGBoost model.
RESULTS
We included 10,962 patients with pneumonia, and the in-hospital mortality was 16.33% In this study, the XGBoost model showed a better performance to predict the mortality of pneumonia with an area under the curve (AUC) of 0.734. Compared with the traditional Apache scoring system (0.677), this model has made significant progress in its results. The SHAP m0ethod reveals the top 20 predictors of pneumonia according to the importance ranking, and the average of the blood urea nitrogen was recognized as the most important predictor variable.
CONCLUSIONS
Interpretable predictive models help physicians more accurately predict the risk of death in patients with pneumonia in the ICU, and interpretable frameworks can increase the transparency of models. We found that age is the most important factor affecting pneumonia mortality, followed by serum AST concentration. Through which these indicators are strongly correlated with mortality, can provide clinical better treatment plans and optimal resource allocation for their patients