Abstract
Chronic kidney disease-mineral bone disorder is a common complication in patients with chronic kidney disease (CKD) and end-stage kidney disease (ESKD), and it increases the risk of osteoporosis and fractures. This study aimed to develop predictive machine-learning (ML) models to identify osteoporosis risk in patients with CKD stages 3-5 and ESKD. We retrospectively analyzed a de-identified osteoporosis database from a Taiwanese hospital, including 6614 patients with CKD stages 3-5 and ESKD who underwent bone mineral density (BMD) scans between January 2011 and June 2022. Nine ML algorithms were applied to predict osteoporosis: logistic regression, XGBoost, LightGBM, CatBoost, SVM, decision tree, random forest, k-nearest neighbors, and an artificial neural network (ANN). The ANN model achieved the highest predictive performance, with an area under the curve (AUC) of 0.940 on the validation and 0.930 on the test datasets. The receiver operating characteristic curve, confusion matrix, and predictive probability histogram revealed that the ANN model performed well in terms of discrimination. Calibration and decision curve analyses further demonstrated the reliability and applicability of the ANN model. The ANN model demonstrated the potential for clinical implementation in screening high-risk patients for osteoporosis.