A comparison of interpretable XGBoost and artificial neural network model for the prediction of severe acute pancreatitis

Abstract

Introduction: Acute pancreatitis (AP) that progresses to persistent organ failure is defined as severe acute pancreatitis (SAP) which has a relatively high mortality. Early establishment of a prediction model is crucial for the improvement of disease prognosis.

Objectives: The aim of this study was to evaluate the accuracy of Extreme Gradient Boosting (XGBoost) and artificial neural network model (ANN) for predicting SAP.

Patients and methods: A total of 648 patients with AP were enrolled. XGBoost and ANN models were developed and valuated in the training set (519 patients) and test set (129 patients), respectively. The accuracy and results of XGBoost and ANN models were evaluated both by area under the receiver operating characteristic curves (AUC) and the area under precision recall curve.

Results: 15 variables were selected for model construction through univariable analysis. The AUCs of XGBoost model and ANN model in five-fold cross-validation of the training set were 0.92 (95%CI, 0.87–0.97) and 0.86 (95%CI, 0.78–0.92), respectively. AUCs of XGBoost model and ANN model for the test set were 0.93 (95%CI, 0.85–1.00) and 0.87 (95%CI, 0.79–0.96). XGBoost outperformed ANN in terms of both diagnostic accuracy and the area under the precision recall curve. Individualized prediction by XGBoost model was explained by local interpretable model-agnostic explanations (LIME) plot.

Conclusions: An interpretable XGBoost model showed higher discriminatory efficiency in predicting SAP compared to ANN.