This retrospective analysis of the prospective cohort study was performed at the HF Center of Fuwai Hospital between December 2006 and December 2017. The patients, including chronic decompensated HF and new-onset HF patients, were continuously enrolled. The diagnosis and assessment of hospitalized HF patients were based on symptoms/signs of fluid overload or hypoperfusion and relevant laboratory, functional, and imaging tests (such as measurements of N-terminal pro-B-type natriuretic peptide (NT-proBNP), echocardiography, electrocardiogram, and chest X-ray). The inclusion criteria were as follows: at least one of the signs and symptoms of HF; New York Heart Association (NYHA) class Ⅱ–Ⅳ; and NT-proBNP levels $>$300 pg/mL.

We did not include patients with missing serum K${}^{+}$ level information or missing follow-up data. Patients without outcome event times were excluded. In addition, those who died while hospitalized were eliminated. This study included 3114 hospitalized patients with HF in the analysis (Supplementary Fig. 1). Serum potassium values were measured in all patients from 2 days before discharge to the day of discharge. The institutional ethics committee of Fuwai Hospital authorized the study protocol after verifying that it adhered to the principles of the Helsinki Declaration. The patients signed individual informed consent forms.

The patients in the study were categorized based on quartiles of serum K${}^{+}$ levels before discharge: K${}^{+}$ $\le$3.96 mmol/L (Q1), 3.96 K${}^{+}$ $\le$ 4.22 mmol/L (Q2), 4.22 K${}^{+}$ $\le$ 4.52 mmol/L (Q3), and K${}^{+}$ $>$4.52 mmol/L (Q4). The serum K${}^{+}$ range of the Q2 group was a reference for statistical analysis. The normal K${}^{+}$ range was 3.5–5.0 mmol/L. Hypokalemia and hyperkalemia were considered K${}^{+}$ levels of 3.5 mmol/L and $>$5.0 mmol/L, respectively.

Clinical details regarding demographics, comorbidities, blood biochemistry results, echocardiograms, and medication data were recorded before discharge. The use of ACEIs or ARBs, $\beta$-blockers, MRAs, digoxin, thiazides, loop diuretics, and other diuretic drugs was assessed. To assess renal function in HF patients, the estimated glomerular filtration rate (eGFR) was employed [13, 14]. Age, sex, and serum creatinine level at baseline were utilized to assess renal function status.

Patients regularly attended followed-up appointments at the outpatient clinic and by telephone after discharge until January 2020, at least once every 3 months within the first year, and every 6 months thereafter. Patients were followed up until cardiovascular or all-cause death occurred. The medical records of patients who were followed up in the Fuwai hospital system provided information on occurrences of adverse events. For patients who were not followed up in our hospital, if necessary, the patient’s relatives and local medical personnel were contacted by telephone to obtain detailed information. Two blinded cardiologists examined and analyzed adverse event data. At 90-days, 2-years, and the maximal follow-up, the primary outcome was all-cause death. The survival duration was computed from the discharge date to the death or final follow-up date.

This study included four K${}^{+}$ intervals, with the reference interval selected as 3.96 K${}^{+}$ $\le$ 4.22 mmol/L (Q2). Continuous (mean $\pm$ SD or median) and categorical (counts and percentages) variables are displayed in the baseline table. Pearson chi-square (proportions) and ANOVA (continuous variables) were used to compare baseline variables across patients with various K${}^{+}$ levels. In addition, the Kruskal‒Wallis rank test was used to assess nonnormally distributed continuous variables. Kaplan-Meier cumulative mortality curves are shown for the quartiles of the serum K${}^{+}$ intervals and illustrate the trend of mortality over time. Clinical comorbidities, laboratory parameters, and important cardiovascular medications were considered covariates in the analysis. A Cox regression model was applied to evaluate the relationship between serum K${}^{+}$ levels and mortality within 90 days, 2 years, and maximal follow-up after adjusting for the defined covariates. The proportional hazard assumption was fulfilled by the Cox regression model. The adjusted variables were chosen based on clinical knowledge, the findings of univariate analyses, and their potential relevance to hypokalemia or hyperkalemia and/or outcomes. Moreover, restricted cubic splines were employed to examine the link between serum K${}^{+}$ levels and all-cause mortality. In a two-sided test, p 0.05 was considered to indicate statistical significance. Multiple imputation statistical methods were used to address missing data. R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) was used for the analysis.

This analysis enrolled 3114 hospitalized HF patients in total, with a mean follow-up of 4.14 years. The majority of HF patients (93.3%) had normal K${}^{+}$ values (3.5–5.0 mmol/L), whereas 52 (1.7%) and 158 (5.1%) patients had hypokalemia and hyperkalemia, respectively. On admission, the average K${}^{+}$ value was 4.02 $\pm$ 0.51 mmol/L, and it was 4.27 $\pm$ 0.45 mmol/L before discharge. Supplementary Fig. 2 shows the distribution of K${}^{+}$ levels before discharge. The patient characteristics based on the quartiles of serum K${}^{+}$ levels before discharge are summarized in Table 1. The mean age was 56.93 $\pm$ 16.04 years, and 2208 patients (70.9%) were male. In addition, 1496 (48.0%) and 1038 (33.3%) had a history of hypertension and diabetes, respectively. Among patients with various serum K${}^{+}$ levels, there was no statistically significant distinction in the usage of drugs (digoxin, ACEIs or ARBs, beta-blockers, MRAs, and diuretics) compared to the Q2 group.

Patients in the Q4 group had a higher proportion of coronary heart disease, hypertension, diabetes, and NYHA class IV than patients in the Q2 group. Patients in the Q4 group had lower levels of hemoglobin, albumin, and eGFR but higher NT-proBNP and high-sensitivity C-reactive protein and a higher age than patients in all other groups.

A total of 1300 deaths (41.7%) occurred during follow-up. The 2-year mortality rates in the quartiles of the serum K${}^{+}$ levels from the lowest (K${}^{+}$ $\le$3.96 mmol/L, Q1) to the highest (K${}^{+}$ $>$4.52 mmol/L, Q4) were 23.3%, 21.2%, 26.1%, and 32.0%, respectively. At the 90-day, 2-year, and maximal follow-up, the crude survival rate of patients in the Q4 group was the worst (Fig. 1). At 2 years, patients in the Q4 group had worse survival than those for all other groups, whereas those for the Q1 and Q3 groups were comparable. The Q2 group patients had a greater 2-year survival rate than the other groups.

Fig. 1.

Kaplan-Meier curve of survival probability of patients based on serum potassium quartiles. Ninety-day (A), 2-year (B), and maximal (C) follow-up survival by four groups: red, K${}^{+}$ $\le$3.96 mmol/L (Q1); yellow, 3.96 K${}^{+}$ $\le$ 4.22 mmol/L (Q2); blue, 4.22 K${}^{+}$ $\le$ 4.52 mmol/L (Q3), and green, K${}^{+}$ $>$4.52 mmol/L (Q4).

Those with hypokalemia (K${}^{+}$ 3.5 mmol/L) and hyperkalemia (K${}^{+}$ $>$5.0 mmol/L) had a substantially greater risk of all-cause mortality than those in the Q2 group. Patients with K${}^{+}$ levels $>$5.0 mmol/L had the lowest survival rate among all the groups (Supplementary Fig. 3). In addition, with cardiovascular mortality as the endpoint, the results were similar to those obtained for all-cause mortality.

A univariate Cox regression model was utilized to identify significant variables that influence all-cause mortality. To evaluate the utility of serum K${}^{+}$ levels in predicting all-cause mortality, we established three multivariate models. After multivariate adjustment, compared with the Q2 group, patients in the Q1 group did not demonstrate an increase in mortality at 90-days, 2-years, or maximal follow-up; however, patients in the Q4 group had an increased mortality rate. Table 2 shows the results of maximal follow-up assessments obtained using the multivariable-adjusted Cox model analysis with the Q2 group as a reference. In the adjusted analysis of Model 3, mortality did not significantly increase in patients in the Q1 group (hazard ratio [HR] 1.12, 95% confidence interval [CI]: 0.95–1.32, p = 0.180) and Q3 group (HR 1.12, 95% CI: 0.95–1.31, p = 0.178) but significantly increased in patients in the Q4 group (HR 1.28, 95% CI: 1.09–1.49, p = 0.002). During the 90-day, 2-year, and maximal follow-up, individuals with hypokalemia and hyperkalemia showed a substantial increase in mortality. The normal range of 4.52 K${}^{+}$ $\le$ 5.0 mmol/L (HR 1.20, 95% CI: 1.02–1.42, p = 0.033) indicated an elevated risk of all-cause mortality, as shown in Fig. 2.

Fig. 2.

Hazard ratios for maximal survival associated with serum potassium levels in heart failure patients. The reference interval is the K${}^{+}$ interval of 3.96–4.22 mmol/L. The adjusted variables were the same as those used for Model 3 in Table 2. HR, hazard ratio.

The model was adjusted for demographic and clinical comorbidities and the use of relevant medications. The spline curve indicates that individuals with hypokalemia and those with hyperkalemia have an elevated risk of all-cause mortality. The spline curve also revealed that the lowest mortality risk was related to a serum K${}^{+}$ level of 4.25 mmol/L. Fig. 3 depicts a J-shaped restricted cubic spline curve.

Fig. 3.

Restricted cubic splines of the hazard ratios for all-cause mortality. Unadjusted risk of mortality at 90-day (A), 2-year (B), and maximal (C) follow-up. The adjusted risk of mortality at 90-day (D), 2-year (E), and maximal (F) follow-up. Adjusted variables are the same as in Fig. 2. HR, hazard ratio.

Using normal K${}^{+}$ values of 3.5–5.0 mmol/L as a reference, with an adjusted HR of 1.53 (95% CI: 1.27–1.85, p 0.001), an abnormal K${}^{+}$ level (hypokalemia and hyperkalemia) was independently linked to an elevated risk of all-cause mortality. We found no significant interaction between abnormal K${}^{+}$ levels and the relevant clinical subgroups or the use of baseline therapies (use of MRAs and ACEIs/ARBs). Hospitalized HF patients with abnormal K${}^{+}$ levels are at an increased risk of death, regardless of whether they have diabetes or renal insufficiency (Supplementary Fig. 4).