The XGBoost machine learning model effectively identified heart failure with preserved ejection fraction in patients with type 2 diabetes mellitus, achieving an AUC of 0.852.
Cross-Sectional (n=1,444)
No
Can machine learning algorithms accurately identify heart failure with preserved ejection fraction in patients with type 2 diabetes mellitus?
The XGBoost machine learning algorithm provides a highly accurate and clinically useful model for identifying underdiagnosed HFpEF in patients with type 2 diabetes mellitus.
Absolute Event Rate: 0.852% vs 0.78%
p-value: p=0.001
Aims: Type 2 diabetes mellitus (T2DM) is commonly observed in heart failure with preserved ejection fraction (HFpEF) patients. Despite its growing prevalence, HFpEF is frequently underdiagnosed. The aim of our study is to apply machine learning algorithms for identifying HFpEF in patients with T2DM. Methods: A total of 1,444 patients with T2DM who met the criteria were consecutively enrolled. Least absolute shrinkage and selection operator (LASSO) technique was applied for feature selection to identify key clinical variables. All patients were randomly divided into a training set and a test set at a ratio of 7:3. Extreme gradient boosting (XGBoost), random forest, K-nearest neighbors, support vector machine (SVM), light gradient boosting machine, decision tree and logistic regression were used to establish diagnostic models. The diagnostic performance of models was evaluated by the area under the receiver operating characteristic curve (AUC), precision, accuracy, F1 score, and Brier score. Calibration curve and decision curve analysis (DCA) were used to assess the accuracy and clinical validity of the model. Results: Patients were divided into HFpEF group and non-HFpEF group. XGBoost model (precision 0.812, accuracy 0.770, sensitivity 0.719, AUC 0.852, F1 score 0.741, Brier score 0.148) and SVM model (precision 0.784, accuracy 0.765, sensitivity 0.681, AUC 0.857, F1 score 0.745, Brier score 0.166) had the highest diagnostic performance. However, the calibration curve of the SVM model depart from the line of perfect calibration which confirmed poor calibration. Therefore, XGBoost was demonstrated to be best ML model in identifying HFpEF in patients with T2DM. Rankings of variable importance based on the Gain metric showed that B-type natriuretic peptide over 100 pg/mL had the highest importance score, followed by albumin, E/e', age and high-sensitivity cardiac troponin T. Conclusions: This study found XGBoost to be the optimal machine learning algorithm in identifying HFpEF in T2DM patients. Additionally, the model demonstrated substantial clinical utility, as assessed by DCA.
Zhou et al. (Thu,) conducted a cross-sectional in Type 2 diabetes mellitus and suspected heart failure with preserved ejection fraction (n=1,444). XGBoost machine learning model vs. H2FPEF score was evaluated on Area under the receiver operating characteristic curve (AUC) for identifying HFpEF (p=0.001). The XGBoost machine learning model effectively identified heart failure with preserved ejection fraction in patients with type 2 diabetes mellitus, achieving an AUC of 0.852.