Risk prediction model for one-month post-discharge hyperkalemia in hospitalized heart failure patients: a nationwide registry cohort study in China
H Wang, K Chai, Y Wan, M Du, Y Li, X Yu, W Li, Y Liao, Y Dong, X Li, J Zhou, Y Huo, J YangAbstract
Background
Heart failure (HF) poses a significant burden in China, where hyperkalemia (HK) represents a critical challenge in HF management, particularly among patients receiving renin-angiotensin-aldosterone system inhibitors (RAASi). While HK in HF has been widely studied globally, short-term risk prediction models for Chinese HF patients are limited.
Purpose
To develop a prediction model for the short-term (one-month) risk of post-discharge hyperkalemia in Chinese HF patients.
Methods
This prospective, nationwide multicenter cohort study utilized the China Cardiovascular Association Database-Heart Failure Center Registry. Between Jan 2017 and Dec 2020, adult hospitalized patients with HF were enrolled. HK was defined as a serum potassium (sK+) level >5.0 mmol/L. Patients with one-month post-discharge sK+ were included to construct prediction model for HK and randomly divided into training and validation sets (7:3). Random forest (RF) and least absolute shrinkage and operator (LASSO) regression were used for variable selection, followed by multivariable logistic regression model. A nomogram was developed to facilitate individualized prediction.
Results
A total of 14,727 patients with HF were included (training set: 10,309; validation set: 4,418). The mean age was 67.9±13.9 years and 39.8% were female. Most patients (56.2%) were New York Heart Association (NYHA) class III/IV. Mean baseline sK+ was 4.2 ± 0.5 mmol/L and 34.2% of patients had an estimated glomerular filtration rate (eGFR) <60 ml/min/1.73m2. Renin-angiotensin system inhibitor (RASi) and mineralocorticoid receptor antagonist (MRA) use at discharge was 76.2% and 77.7%, respectively. Common comorbidities included hypertension (58.5%), coronary heart disease (55.8%), atrial flutter/fibrillation (33.7%) and diabetes (28.2%). After integrating the overlapping variables identified by LASSO regression and RF screening in the training set, and considering clinical interpretability, five key predictors were selected. Higher baseline sK+ level, lower eGFR, presence of diabetes, and use of RASi and MRA at discharge were significantly associated with higher risk of HK, as shown in the nomogram (Figure 1A). The model demonstrated good predictive value, with an area under curve (AUC) of 0.746 (95% CI: 0.722-0.766) in the training set (Figure 1B), and 0.723 (95% CI: 0.703-0.743) in the validation set (Figure 1C).
Conclusion
This study developed a practical nomogram based on five predictors including sK+, eGFR, diabetes, RASi and MRA to predict the short-term risk of post-discharge HK in hospitalized HF patients. The model demonstrates satisfactory discriminative ability and enables early identification of high-risk individuals, facilitating personalized monitoring strategies and the safe optimization of RAASi therapy in the critical post-discharge period.