Abstract: PO1128
Machine Learning Models to Predict Cardiovascular and Renal Outcomes and Mortality in Hyperkalemic Patients
Session Information
- Salt, Potassium, and Water Balance: Clinical
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: Fluid, Electrolyte, and Acid-Base Disorders
- 902 Fluid, Electrolyte, and Acid-Base Disorders: Clinical
Authors
- Kanda, Eiichiro, Kawasaki Medical School, Kurashiki, Japan
- Okami, Suguru, AstraZeneca K.K., Osaka, Japan
- Kohsaka, Shun, Keio University, Tokyo, Japan
- Ma, Xiaojun, IQVIA Solutions K.K., Tokyo, Japan
- Okada, Masafumi, IQVIA Solutions K.K., Tokyo, Japan
- Kimura, Takeshi, Real World Data, Co.Ltd., Kyoto, Japan
- Yajima, Toshitaka, AstraZeneca K.K., Osaka, Japan
Background
Hyperkalemia is associated with increased risks of mortality and adverse clinical outcomes. To date, limited evidence is available for personalized risk evaluation in this heterogeneous and multifactorial pathophysiological conditions.
Methods
We developed prediction models using extreme gradient boosting (XGB), logistic regression (LR), and neural network. Models were derived and cross-validated in a retrospective cohort of hyperkalemic patients with either heart failure or stage ≥3a chronic kidney disease and aged ≥18 years from a Japanese administrative hospital database (April 1, 2008–September 30, 2018). The outcomes of interest included all-cause death, introduction of renal replacement therapy (RRT), hospitalization for heart failure (HHF), and cardiovascular events within 3 years after first hyperkalemic episode. The best performing model was further validated using a separate hospital-based database.
Results
24,949 adult patients with hyperkalemia were selected for the model derivation and internal validation. The mean age was 75 years and 54% were male. Among machine learning algorithms tested, XGB outperformed other models, showing AUROC of XGB vs. LR for all-cause death, RRT, HHF, and cardiovascular events as 0.823 vs. 0.809, 0.957 vs. 0.947, 0.863 vs. 0.838, and 0.809 vs. 0.798, respectively. In the external validation set including 86,279 patients, AUROC of XGB for all-cause death, RRT, HHF, and cardiovascular events were 0.747, 0.888, 0.673, and 0.585, respectively. The Kaplan-Meier curves of high-risk predicted group showed a significant differentiation from that of low-risk predicted group for all outcomes (Figure).
Conclusion
These findings suggest the possible use of machine learning models for real-world risk assessment as a guide for treatment decision making that may lead to the improvement of cardiovascular and renal outcomes, and mortality in hyperkalemic patients.
Funding
- Commercial Support –