The Relation of Serum Potassium Concentration with Cardiovascular Events and Mortality in Community-Living Individuals

Study Populations

The MESA enrolled 6814 participants between 2000 and 2002 who were free of CVD (defined as physician diagnosed myocardial infarction (MI), angina, nitroglycerine use, stroke or transient ischemic attack, heart failure, current atrial fibrillation, or having undergone cardiovascular procedures [coronary artery bypass grafting, angioplasty, valve replacement, pacemaker or defibrillator implantation, any surgery on the heart or arteries]). The population was recruited to be multiethnic (black, Hispanic, Asian, and non-Hispanic white Americans) and was drawn from six cities across the United States (7). From this group, we excluded 346 who were missing serum potassium measurements or covariate data, resulting in an analytic sample of 6468 participants within the MESA.

Participants in the CHS were all aged ≥65 years at baseline and there was no exclusion for prevalent CVD (7,8). The CHS enrolled 5201 participants between 1989 and 1990 from four regions across the United States. An additional 687 black participants were enrolled between 1992 and 1993. From these 5888 participants, 711 lacked eGFR measurements, 80 lacked serum potassium measurements, 25 lacked covariate measurements, and 1926 had prevalent CVD (defined as MI, stroke, coronary heart disease [CHD], angina, coronary bypass, transient ischemic attack, or heart failure). As these groups were not mutually exclusive, these exclusions resulted in a sample of 3206 CHS participants. Twenty-three individuals had participated in both the MESA and CHS, so after excluding them from the MESA (and retaining them in the CHS), we ultimately derived an analytic dataset of 9651 participants for pooled analysis.

The institutional review boards for both MESA and CHS sites, and the University of California, San Diego approved this study, and all participants provided written informed consent.

Serum Potassium

In the MESA, serum samples were collected at baseline, between 2000 and 2002, and frozen at −80°C. They were thawed in 2011 for measurement at the Kidney Research Institute at University of Washington utilizing indirect potentiometry with an ion selective electrode on an automated clinical chemistry analyzer (UniCel DxC 800; Beckman Coulter, Inc., Brea, CA). Coefficients of variation (CVs) for this assay were 1.3%–2.0%.

In the CHS, serum potassium was measured in 5132 samples from the 1989–1990 and 1992–1993 exams at the University of Vermont Laboratory for Clinical Biochemistry Research following standard procedures for collection and storage, and using the Kodak Ektachem 700 Analyzer (Eastman Kodak, Rochester, NY), a colorimetric method (9). CVs were approximately 2.0% (10).

Cardiovascular Events

In both the MESA and CHS, a CVD event was defined as a composite of MI, resuscitated cardiac arrest, definite angina, probable angina (if followed by revascularization), stroke, stroke death, CHD death, other atherosclerotic death, or other CVD death. Cardiovascular events were adjudicated by a centralized events committee, which entailed detailed review of medical records by a panel of physicians, compared with prespecified criteria, which came to consensus on each event (7,11).

Total and CVD Mortality

Total and CVD mortality were adjudicated by centralized events committees in both cohorts (11,12). In both the MESA and CHS, CVD death was defined as death from atherosclerotic CVD, stroke, atherosclerotic disease other than coronary disease or stroke (e.g., aortic aneurysm), or other CVD death (e.g., valvular heart disease).

In the MESA, non-CVD death was defined as a death that was not related to CVD. In the CHS, non-CVD death was first classified into 19 disease and organ system categories. We then collapsed non-CVD death into four categories representing the most common causes: pulmonary (e.g., chronic obstructive pulmonary disease, pneumonia), cancer, neurologic (e.g., Parkinson disease, dementia, amyotrophic lateral sclerosis), or other, consistent with prior CHS manuscripts (13).

Sudden cardiac death (SCD) was defined according to the National Heart, Lung, and Blood Institute working group on SCD, where SCD is a pulseless condition of cardiac origin in a previously stable person, occurring out of the hospital or in the emergency department (14). Investigation of SCD in the CHS was performed ancillary to the collection of main events, and was followed until 2006.

Covariates

Covariates were measured at baseline concurrent with serum potassium measurement. Participants reported age, sex, race/ethnicity, highest attained education, smoking status, pack-years of smoking, and ever having been diagnosed with cancer. Medication use was obtained through a validated medication inventory, as described previously (7,15). Medications of particular interest for this study included diuretics, NSAIDS (both prescribed and over-the-counter), potassium supplements, angiotensin-converting enzyme (ACE) inhibitors/angiotensin II receptor blockers (ARB), β agonists, β blockers, and any other antihypertensive medications (15). ARBs were not available at the CHS baseline visit (1989–1990), and were therefore not included in the CHS. Diabetes mellitus was defined as fasting blood glucose ≥126 mg/dl or the use of hypoglycemic medications. eGFR<60 ml/min per 1.73 m² was calculated using serum concentrations of both creatinine and cystatin C (16). Trained study personnel measured systolic and diastolic BP, height, and weight. Standard clinical analyzers were used to measure fasting blood glucose, total cholesterol level, triglycerides, and urine albumin and creatinine, the latter two of which were combined to create the urine albumin-to-creatinine ratio in analysis in the MESA.

Statistical Methods

Before analysis, we elected to model serum potassium by categories: <3.5, 3.5–3.9, 4.0–4.4, 4.5–4.9, and ≥5.0 mEq/L. We examined the distribution of covariates by categories of serum potassium using ANOVA for continuous variables and chi-squared tests for categorical variables.

In the MESA, participants provided risk time from their baseline visit (2000–2002) until April 13, 2011. In the CHS, participants were considered at risk from their baseline visit (1989–1990 or 1992–1993, depending upon date of each participant’s enrollment) until December of 2011, except for sudden death, which was available up to 2006.

We calculated the unadjusted incident rates of CVD events, and total and cause-specific mortality for each category of serum potassium concentration separately. We then used Cox proportional hazards models to examine associations of serum potassium concentrations with each outcome, setting the 4.0–4.4 mEq/L category as the referent category. With regard to cause-specific mortality, we utilized competing risks models as described by Lunn and McNeil (17). Analyses were performed in sequential models. Model 1 adjusted for age, sex, and race/ethnicity. Model 2 additionally added time-varying eGFR. Model 3 further added diabetes mellitus and systolic BP, and model 4 added current smoking, pack-years of smoking, ever having cancer, study cohort (MESA versus CHS), ACE/ARB, diuretics (potassium sparing and all other diuretics), NSAIDs, potassium supplements, β agonists, β blockers, and use of any other antihypertensive medication.

To examine the functional form of fully adjusted associations between unit change in serum potassium concentration with each outcome, we used penalized smoothing splines with evenly spaced knots among the inner 95% distribution (18).

In order to address whether associations differed by eGFR<60 ml/min per 1.73 m² versus eGFR≥60 ml/min per 1.73 m², ACE/ARB use, NSAIDs use, or diuretics use, we evaluated multiplicative interaction terms, and conducted stratified analyses when statistically significant interactions were observed. Additionally, we evaluated interactions by sex and race/ethnicity for the main outcomes only: all-cause mortality, CVD death, non-CVD death, and CVD events.

All statistical tests were two-sided, and P<0.05 was considered statistically significant for all analyses, including interaction terms. No adjustments were made for multiple comparisons, as our a priori hypothesis focused on these specific outcomes (19). Analyses were conducted using SAS, version 9.4 (SAS Institute, Inc., Cary, NC) except for spline analysis, which was conducted in R software (2013; http://www.R-project.org/).