Abstract: PO2371
Predict Hyperkalemia in Advanced CKD Patients Using Machine Learning Algorithms
Session Information
- CKD: Insights from Recent Clinical Trials and Large Real-World Effectiveness Studies
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: CKD (Non-Dialysis)
- 2102 CKD (Non-Dialysis): Clinical, Outcomes, and Trials
Authors
- Chang, Hsin-Hsiung, Paochien Hospital, Pingtung, Taiwan
- Wu, Chia-Lin, Changhua Christian Hospital, Changhua, Changhua, Taiwan
Background
Hyperkalemia is a common and fatal problem in advanced chronic kidney disease patients. The incidence rate was about 40-50%. It may cause muscle weakness, paralysis, and even cardiac arrhythmia. Our goal is to develop machine learning models to predict hyperkalemia in advanced chronic kidney disease patients, which could help physicians make clinical decisions.
Methods
We collected clinical data for advanced CKD (CKD stage 4 and 5, eGFR < 30 ml/min/1.73m2) patients receiving Output Patient Care in one medical center in Taiwan from January 2010 to December 2019. 1,965 patients were included. Four machine learning models (multilayer perceptron [MLP], logistic regression with regularization, XGBoost, and random forest [RF]) were used to estimate serum potassium concentration 3 months later. 2 Nephrologists participated in human-machine competition.Area under the receiver operating characteristic curves (AUCs), sensitivity, specificity, positive (PPV) and negative (NPV) predicted values, and accuracy were used to evaluate
the performance of machine learning models with that of these physicians.
Results
In a test set including 2,074 records, the AUC of machine learning models was highest for XGBoost (0.843; 95% confidence interval [CI], 0.822–0.864). Moreover, the NPV and specificity were 0.875 and 0.943, respectively. The AUC for detecting hyperkalemia by humans was 0.602, 95% CI, 0.580–0.623. XGBoost model performed significantly better than humans (P < 0.001, using the DeLong test).
Conclusion
Machine learning models may help physicians make clinical decisions in advanced CKD patients who suffer from hyperkalemia in outpatient department care and possibly reduce cardiac arrhythmia.