Predicting the risk factors of diabetic ketoacidosis-associated acute kidney injury: A machine learning approach using XGBoost

Abstract

Objective: The purpose of this study was to develop and validate a predictive model based on a machine learning (ML) approach to identify patients with DKA at increased risk of AKI within 1 week of hospitalization in the intensive care unit (ICU).

Methods: Patients diagnosed with DKA from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database according to the International Classification of Diseases (ICD)-9/10 code were included. The patient's medical history is extracted, along with data on their demographics, vital signs, clinical characteristics, laboratory results, and therapeutic measures. The best-performing model is chosen by contrasting the 8 Ml models. The area under the receiver operating characteristic curve (AUC), sensitivity, accuracy, and specificity were calculated to select the best-performing ML model.

Results: The final study enrolled 1,322 patients with DKA in total, randomly split into training (1,124, 85%) and validation sets (198, 15%). 497 (37.5%) of them experienced AKI within a week of being admitted to the ICU. The eXtreme Gradient Boosting (XGBoost) model performed best of the 8 Ml models, and the AUC of the training and validation sets were 0.835 and 0.800, respectively. According to the result of feature importance, the top 5 main features contributing to the XGBoost model were blood urea nitrogen (BUN), urine output, weight, age, and platelet count (PLT).

Conclusion: An ML-based individual prediction model for DKA-associated AKI (DKA-AKI) was developed and validated. The model performs robustly, identifies high-risk patients early, can assist in clinical decision-making, and can improve the prognosis of DKA patients to some extent.

Keywords: XGBoost; acute kidney injury; diabetic ketosis; machine learning; outcome.

Figure 1

Overall flowchart of this study. MIMIC-IV, Medical Information Mart for Intensive Care IV; CKD, chronic kidney disease; ICU, intensive care unit; DKA, v; AKI, acute kidney injury; HR, heart rate; RR, respiratory rate; WBC, white blood cell count; PLT, Platelet count; Hb, hemoglobin; CHF, congestive heart failure; XGBoost, eXtreme Gradient Boosting; GNB, Gaussian Naïve Bayes; CNB, Complement Naive Bayes; MLP, multi-layer perceptron neural network; SVM, support vector machine; LASSO CV, least absolute shrinkage and selection operator cross-validation.

Figure 2

Lasso CV method was used to conduct feature selection. LASSO CV, least absolute shrinkage and selection operator cross-validation, BUN, blood urea nitrogen; BG, blood glucose; PLT, platelet count.

Figure 3

Comparing the different ML models’ AUC in the training (A) and validation (B) sets. ML, machine learning; AUC, area under the receiver operating characteristic curve; XGBoost, eXtreme Gradient Boosting; GNB, Gaussian Naïve Bayes; CNB, Complement Naive Bayes; MLP, multi-layer perceptron neural network; SVM, support vector machine.

Figure 4

DCA (A) and calibration curve (B) of the XGBoost and simplified model. DCA, decision curve analysis; XGBoost, eXtreme Gradient Boosting.

Figure 5

BUN, blood urea nitrogen; PLT, platelet count, BG, blood glucose.