Serum Potassium, End-Stage Renal Disease and Mortality in Chronic Kidney Disease

The kidneys play a crucial role in potassium homeostasis, and the importance of their contribution in the maintenance of potassium balance is reflected by the high rates of potassium disorders in patients with decreased kidney function [1,2]. Indeed, individuals with chronic kidney disease (CKD) and those with end-stage renal disease (ESRD) can experience both hyperkalemia and hypokalemia; the former typically occurs by virtue of low kidney function or as a consequence of drugs such as Angiotensin Converting Enzyme inhibitors (ACEIs) or Angiotensin-Receptor Blockers (ARBs), whereas the latter is typically a consequence of diuretic administration [3].

Among dialysis patients, hyperkalemia has been associated with higher mortality due to its arrhythmogenic effects [4,5]. Hypokalemia has also been linked to higher mortality in dialysis patients, not only because of the electrophysiological effects of potassium, but also because hypokalemia often reflects a poor underlying nutritional status [6]. In CKD patients not on dialysis, however, the impact of potassium disorders on outcomes is unclear. Recent evidence suggests that the relationship between serum potassium and mortality follows a U-shaped curve in which both low and high serum potassium are associated with an increase in mortality [7,8]. However, the data is heterogeneous with a variable follow-up ranging from 1 day to 4 years and variable number of potassium values per patient [9]. In addition, low serum potassium has been postulated to be a risk factor for the progression of CKD, as it had been shown to induce kidney damage via modulation of renal inflammation and impaired angiogenesis [10,11,12]. To our knowledge, the association between serum potassium and ESRD has not been studied in detail. Therefore, we examined the associations between serum potassium and all-cause mortality and ESRD in a cohort of stage 3 and stage 4 CKD patients in our health care system.

We conducted an analysis using our preexisting electronic health record (EHR)-based CKD registry. The development and validation of our EHR-based CKD registry at Cleveland Clinic are described in detail elsewhere [13].

Patients who met the following criteria from January 1, 2005 to September 15, 2009, were included in the study population: (1) had at least one face-to-face outpatient encounter with a Cleveland Clinic health care provider, (2) had 2 or more estimated GFR (eGFR) values <60 ml/min/1.73 m² more than 90 days apart, and (3) had a potassium measured on the date of the second eGFR <60 ml/min/1.73 m². Patients aged <18 years old, who entered the registry after September 15, 2009 and those who were diagnosed with ESRD needing dialysis or renal transplant before the date of second eGFR <60 ml/min/1.73 m² were excluded.

Renal Function

We applied the CKD-EPI equation to patients who had 2 outpatient serum creatinine levels between January 1, 2005 and September 15, 2009, to calculate eGFR [14]. All creatinine measurements were performed by the modified kinetic Jaffe reaction, using a Hitachi D 2400 Modular Chemistry Analyzer (Roche Diagnostics, Indianapolis, IN) in our laboratory. CKD was defined according to current guidelines as follows: stage 3 CKD (eGFR 30-59 ml/min/1.73 m²) and stage 4 CKD (eGFR 15-29 ml/min/1.73 m²). We further categorized stage 3 into CKD stage 3a (eGFR 45-59 ml/min/1.73 m²) and stage 3b (eGFR 30-44 ml/min/1.73 m²).

Serum Potassium

Only outpatient serum potassium laboratory measures obtained with a same reference range were included in this analysis. Serum potassium levels measured on the day of CKD confirmation (second eGFR <60 ml/min/1.73 m² at least 90 days after the first eGFR), as described, were used for the analysis in which single potassium measured was considered. For our time-dependent repeated measures analysis, we included the baseline potassium as well as the first potassium with an appropriate reference range measured each month during the study follow-up. We used carry-forward values to fill in data for months with no potassium measurement.

Comorbid Conditions and Laboratory Parameters

Demographic details were extracted from the EHR. Diabetes mellitus, hypertension, coronary artery disease, and other comorbid conditions were defined using pre-specified criteria and validated [13]. Serum potassium levels and other relevant outpatient laboratory details were obtained from our electronic laboratory records.

Outcomes

The primary outcomes of interest, all-cause mortality and ESRD were ascertained from our EHR and linkage of our CKD registry with the Social Security Death Index and United States Renal Data Services (USRDS). Patients were followed from their date of inclusion in the registry (date of second qualifying eGFR) until September 15, 2009.

CKD patients with measured outpatient serum potassium values at the time of second eGFR <60 ml/min/1.73 m² were further classified into three groups: <3.5, 3.5-3.9, 4.0-4.9, 5.0-5.5 and >5.5 mmol/l based on clinical relevance and previous reports. Associations of the baseline characteristics for these groups were assessed using Chi-square and ANOVA tests for categorical and continuous variables respectively.

Two separate logistic regression analyses were conducted to examine the association of low (<3.5 mmol/l) or high (>5.0 mmol/l) serum potassium levels with potential covariates. For each of these models, the reference group was all patients with normal serum potassium levels based on our laboratory reference values (3.5-5 mmol/l). Covariates were based on information known prior to second eGFR <60 and chosen a priori based on factors previously shown or thought to be related to serum potassium, ESRD and mortality. These include age, gender, race, body mass index (BMI), eGFR, diabetes, hypertension, malignancy, coronary artery disease, heart failure, chronic obstructive pulmonary disease and/or asthma, use of ACE/ARB, and use of beta blockers.

To evaluate whether survival and ESRD among persons with CKD was associated with serum potassium levels, we used Kaplan-Meier plots and log-rank tests with the date of second eGFR <60 ml/min/1.73 m² as the time of origin. Progression to ESRD and pre-ESRD death are competing events; therefore, we fitted cumulative incidence functions that adjust for competing risks and compared these results to the traditional cause specific analysis. In addition, a separate analysis that included all deaths (both before and after ESRD) was also conducted. We used Cox proportional hazards models to assess the association between serum potassium and mortality and ESRD while adjusting for other covariates mentioned earlier, as well as cerebrovascular disease, peripheral vascular disease, potassium sparing diuretics, non-potassium sparing diuretics, potassium supplementation, serum bicarbonate, log glucose and albumin. To incorporate serum potassium results obtained after inception, we fitted a Cox proportional hazards model with time-dependent repeated measures of potassium while adjusting for all variables mentioned earlier. We used splines to relax linearity assumptions for continuous variables included in the models. We also examined the relationship between continuous baseline serum potassium and each outcome using restricted cubic splines.

We tested two-way interactions between baseline serum potassium and the following covariates: age, gender, race, diabetes and eGFR in the adjusted Cox proportional hazards model. The patients who had missing serum glucose and/or bicarbonate data were 0.3% and 13% of them had missing albumin. We used mean value imputation to include all patients in the Cox model. To evaluate the effect of the imputation, we also fit a model with time-dependent serum potassium with complete case data. We did not adjust for proteinuria in our main models because we had data only in 49% of the sample. However, we performed a sensitivity analysis fitting Cox models with time-dependent serum potassium as described earlier and adjusting for proteinuria in a sensitivity analysis.

All data analyses were conducted using Unix SAS version 9.4 (SAS Institute, Cary, N.C., USA) and R 3.0.1 (The R Foundation for Statistical Computing, Vienna, Austria). The cmprsk package was used for competing risk analysis. The CKD registry and this study were approved by the Cleveland Clinic Institutional Review Board.

There were 42,912 patients with stages 3 and 4 CKD from our CKD registry, and of those, 36,359 (85%) had serum potassium levels measured on the date of second eGFR <60 ml/min/1.73 m² with reference ranges 3.5-5 mmol/l. This group constituted the study population (online suppl. fig. 1; for all online suppl. material, see www.karger.com/doi/10.1159/000437151). Patients who had had measurements with different assays (n = 4,725, 11%) were excluded. The mean age of the study population was 72 ± 11.9 years with 55% being females and 13% African Americans. Demographic variables and comorbid conditions were significantly different among patients with various ranges of serum potassium levels (table 1). Serum potassium <3.5 mmol/l was noted among 3% and >5.0 mmol/l was observed among 11% of the study population.

Low Serum Potassium Levels (<3.5 mmol/l)

In the multivariable analysis, the following variables were associated with higher odds of having low potassium levels compared to normal: African American race, lower eGFR, malignancy, and heart failure (table 2). Conversely, the following variables were associated with having lower odds of having low serum potassium levels: older age, male gender, being overweight, diabetes, hypertension and use of ACE/ARB (table 2).

High Serum Potassium Levels (>5.0 mmol/l)

In the multivariable analysis, the following variables were associated with higher odds of having high potassium levels compared to normal: male gender, lower eGFR, BMI <18.5 kg/m², diabetes, malignancy, and use of ACE/ARB (table 2). The following variables were associated with having lower odds of having high serum potassium levels: higher age and higher BMI (table 2).

Among our study population, 6,114 of them died before reaching ESRD and 772 reached ESRD during an average follow-up of 2.3 years. A total of 6,379 patients died overall. Both the Kaplan-Meier and competing risk analyses showed a significant difference in overall mortality and ESRD among those with different potassium levels (p < 0.001; fig. 1) and the results were similar in both analytical approaches. In the multivariate model (time-dependent repeated measures analysis) adjusting for demographics, comorbid conditions, medications use, lower levels of serum potassium (<3.5 and 3.5-3.9 mmol/l) were associated with higher risk for mortality. Similar higher risk for mortality was noted for those with higher levels of serum potassium (5.0-5.4 and >5.5 mmol/l; table 3). No such associations between potassium levels and ESRD were noted (table 3). In a model in which only baseline serum potassium was considered the explanatory variable, serum potassium levels >5.5 mmol/l were associated with higher risk for death (online suppl. table 1). There was a U-shaped association between serum potassium levels and mortality when serum potassium was examined as a continuous variable (fig. 2).

We found significant two-way interactions between serum potassium and age and eGFR on mortality, which indicated that the increased mortality hazard associated with abnormal levels of serum potassium was different at different ages (p = 0.04) and eGFR stages (p < 0.001). Age appears to modify the association with younger patients showing stronger association and seems to either attenuate or reduce the statistical significance in elderly population (online suppl. table 2). The interaction with CKD stage suggested that serum potassium is associated with increased mortality among patients with stage 3a but not in other stages of CKD (data not shown).

In a sensitivity analysis in which we included patients who had proteinuria data (n = 17,790), similar results were noted (data not shown). Results from the model including only patients with complete data (complete case analyses) were qualitatively similar to those in the model using mean value imputation and a class level for missing BMI groups.

In this large cohort of patients with stages 3 and 4 CKD, the prevalence of hypokalemia (<3.5 mmol/l) and hyperkalemia (>5.0 mmol/l) was 3 and 11%, respectively. Hyperkalemia was more frequently observed in patients with lower eGFR, treated with ACEI/ARB and those with conditions known to cause cellular shifts such as diabetes. On the other hand, hypokalemia was more often seen in African Americans and in patients taking diuretics. We observed a U-shaped association between serum potassium and death with both lower and higher levels of potassium being associated with increased risk of death. However, neither low nor high serum potassium levels were associated with progression to ESRD.

Hypokalemia and hyperkalemia are frequently observed in patients with CKD. As the eGFR falls, the renal excretion of potassium is reduced, and the prevalence of hyperkalemia increases from 2% in patients with eGFR >60 ml/min/1.73 m² to 42% in patients with eGFR <20 ml/min/1.73 m² [15,16]. Other factors that have been linked to hyperkalemia include cellular shifts from underlying acidosis or diabetes and treatment with RAAS blockers such as ACEI or ARB [17,18,19,20]. Hypokalemia, on the other hand, is often seen in the setting of diuretic use and has been proposed as a marker of poor nutritional status [21]. Our findings were concordant with the previous studies: hyperkalemia was more frequently observed in patients with lower GFR, treated with ACEI/ARB and those with diabetes. These same factors were associated with lesser risk for hypokalemia. On the other hand, diuretics were associated with higher odds of hypokalemia and lower odds of hyperkalemia. Black race was also associated with higher odds of hypokalemia. Previous reports have shown that African Americans have lower mean serum potassium concentration than whites in non-CKD population and the risk for unprovoked hypokalemia was 5 times that of Whites [22,23,24,25]. Our findings extend this result to African Americans with CKD and suggest the need to monitor them closely for development of hypokalemia.

Hypokalemia has been previously linked to increased mortality in patients with heart failure and acute myocardial infarction [26,27]. In a propensity-matched analysis, Bowling et al. [28] reported an association between low serum potassium (defined as <4 mmol/l) and mortality in those with CHF and CKD. In a CKD population with a mean eGFR of 25.4 ml/min/1.73 m², an association between low serum potassium and mortality was reported. Our findings extend these observations to those without heart failure and to those with stage 3 CKD. Low serum potassium levels affect the myocardial resting membrane potential, which can lead to ventricular arrhythmias and sudden cardiac death [29]. Low potassium levels can also be a reflection of a poor nutritional status, which in turn portends poor prognosis [30] but in the repeated measures analysis, there exists an association between hypokalemia (both in the 3.5-3.9 and <3.5 mmol/l) and mortality despite adjusting for serum albumin.

In our study population, serum potassium >5.0 mmol/l is also associated with increased mortality in non-dialysis-dependent CKD patients, a finding that is typically attributed to the arrhythmogenic effects of potassium. In a secondary analysis of the Renal Research Institute (RRI) study CKD cohort, an increased risk of the composite of death or cardiovascular events requiring hospitalization was observed but did not show an increase in overall mortality [8]. The differences between our study and the RRI study could be attributed to the reference ranges used (4-5.5 mmol/l in RRI vs. 4.0-4.9 mmol/l in our study). However, another study included hospitalized CKD patients and demonstrated higher risk for death within 1 day among those with hyperkalemia [31]. Our study shows a clear association between serum potassium and mortality with hypokalemia <4.0 mmol/l and hyperkalemia ≥5.0 mmol/l (table 3).

However, when looking at the interaction between age groups and potassium, we found a statistically significant association between serum potassium of 3.5-3.9 mmol/l and mortality for patients aged <80 years old but not in those aged >80 (online suppl. table 2). A potential explanation could relate to the high morbidity load of patients aged above 80 that would put them at risk of death regardless of their serum potassium. Baseline kidney function seem to modify the associations between serum potassium and outcomes as patients with hypo/hyperkalemia and stage 3a CKD (eGFR 45-59 ml/min/1.73 m²) were more likely to die than those with stages 3b and 4 CKD (eGFR <45 ml/min/1.73 m²). Such findings could be attributed to the fact that higher occurrence of chronic hyperkalemia in CKD patients might be better tolerated than acute hyperkalemia. This in turn might lead to a reduced sensitivity to cardiac complications [32]. Further studies are warranted to ascertain whether such phenomenon exists for hypokalemia.

It has been suggested that potassium disorders could lead to progression of renal disease. Hypokalemia in particular could contribute to interstitial renal scarring, and experimental studies suggest that chronic hypokalemia can lead to renal fibrosis via modulation of renal inflammation and local activation of the RAAS [11,33,34]. Current clinical evidence provide conflicting findings, as some studies suggest that hypokalemia was associated with progression to ESRD, while others did not observe such association after adjustment for nutritional indices (serum albumin) [8,35]. Hayes et al. [7] noted that hypokalemia was associated with renal progression but the association with hard renal outcome, such as reaching ESRD was unclear. Table 3 shows that there could be a higher risk for ESRD in those with serum potassium 3.5-3.9 mmol/l, but the overall p value for the model wasn't significant, suggesting that no associations exist between hypokalemia and ESRD. Despite the large sample size and inclusion of relevant confounders, we didn't observe an association between hypokalemia and kidney disease progression.

Strengths of this analysis include a large diverse patient population with stages 3 and 4 CKD, availability of several confounding variables and the use of time-dependent repeated measures analyses. However, retrospective analyses are prone to residual confounding. While we included several variables that could affect serum potassium levels and are related to mortality (diabetes, serum bicarbonate, medications etc.), we lacked details about nutritional data. Further, our patients have been followed in a health care system and hence these data might not be applicable to the community-dwelling adults with CKD.

In summary, hypokalemia and hyperkalemia are often noted in those with CKD. Several clinical and demographic factors are associated with potassium disorders. Both serum potassium levels <4.0 and >5.0 mmol/l are associated with all-cause mortality but not with ESRD in those with stages 3 and 4 CKD. These associations are modified by the stage of kidney disease and age and further studies are needed to understand the potential mechanisms that explain these findings.

The authors wish to thank Welf Saupe, Vicky Konig and John Sharp of Cleveland Clinic who helped in data extraction during the development of the registry.

SDN: R01DK101500; JVN: DK094112 and SEJ: 1K23DK091363. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The creation of the CCF CKD registry was funded by an unrestricted grant from Amgen, Inc., to the Department of Nephrology and Hypertension Research and Education Fund. Parts of this manuscript were presented as a poster at the annual National Kidney Foundation Meeting held in Dallas, TX in March 2015.

The authors have no relevant financial interest in the study.