Background
Potassium homeostasis is regulated within extracellular concentrations between 3.5 and 5.0 mmol/l [
1]. This is of importance for physiological processes, such as the negative resting membrane potential and consequently neuromuscular and cardiac excitability [
1]. Severe hyperkalaemia is associated with ventricular arrhythmia, bradycardia and cardiac arrest [
2]. Symptoms of severe hypokalaemia include muscle weakness up to paralysis, respiratory insufficiency, constipation and cardiac arrhythmias [
3]. Hypo- and hyperkalaemia are associated with an increased complication rate and mortality risk [
1‐
6].
Contributing mechanisms for potassium electrolyte disorders include disturbances in potassium ingestion, excretion and potassium shifting between intra- and extracellular compartment. Various medications interfere with potassium homeostasis [
3]. The drug with the highest risk for inducing hyperkalaemia is the intravenous potassium supplementation [
7].
The European Resuscitation Council Guidelines for Resuscitation 2015 state that there is no universal definition of hyperkalaemia [
2]. They define hypokalaemia as serum potassium concentration < 3.5 mmol/l and hyperkalaemia as > 5.5 mmol/l. According to the guideline, therapy to lower potassium concentration should be initiated > 5.5 mmol/l. Severe hypokalaemia should be treated with intravenous potassium application in emergency situations [
2]. In patients with myocardial infarction, a previous American College of Cardiology/American Heart Association (ACC/AHA) Guideline suggested to elevate potassium concentrations > 4.0 mmol/l to reduce the risk of ventricular fibrillation [
8,
9]. Evidence for potassium targets in the upper reference level originates from studies implemented before the routine use of β-blockers and reperfusion therapy in patients with myocardial infarction [
8]. Potassium repletion to 4.0 mmol/l or higher is recommended in patients with torsade de pointes episodes associated with medication-induced QT prolongation [
10]. After cardiac and thoracic surgery, guidelines recommend to correct hypokalaemia to 4.5–5.5 mmol/l in order to prevent atrial fibrillation, although they emphasise that this suggestion has never been scientifically confirmed [
11]. In summary, previous guidelines recommend potassium concentrations rather in the upper normokalaemic range. However, there is evidence that potassium concentrations above 4.5 mmol/l may be adversely associated with survival in patients with myocardial infarction [
8,
12].
Increased mortality has been shown for hyperkalaemia in ICU patients [
6]. To our knowledge, there is no valid guideline available regarding potassium target ranges and supplementation in ICU patients. The purpose of this retrospective study was to analyse mean potassium concentrations as well as potassium variability and their association with mortality in a general ICU population. We aimed to identify possible mortality differences within the normal range. Until now, this is the largest analysis considering the association between mortality and potassium concentrations in the range from 3.0 to > 5.5 mmol/l. In addition, we investigated the association of potassium supplementation with mortality in groups of potassium mean and variability.
Materials and methods
We performed a retrospective analysis of ICU patients treated in the Department of Anaesthesiology and Operative Intensive Care Medicine, Charité – Universitätsmedizin Berlin. Trial registration: German Clinical Trials Register, DRKS00016411. Retrospectively registered 11 January 2019,
http://www.drks.de/DRKS00016411.
Inclusion criteria and patient cohort
ICU patients \(\ge\) 18 years with a minimum of two serum potassium values during ICU stay who were treated in one of the ICUs of the Department of Anaesthesiology and Operative Intensive Care Medicine, Charité – Universitätsmedizin Berlin between January 2006 and May 2018 were eligible for inclusion. Potassium values > 10 mmol/l and < 2 mmol/l were excluded.
Data sources and ethics vote
We extracted data from the local patient documentation system. All patient data including blood gas analysis, laboratory measurements, medications, patient baseline characteristics and vital signs are documented in this patient data management system. The retrospective analysis of these data was approved by the local ethics committee, Ethikkommission Charité - Universitätsmedizin Berlin, Germany (EA2/187/18).
Exposure of interest and covariates
The exposure of interest was the patient’s mean potassium value during the ICU stay. Patients were categorised into groups of mean potassium: < 3.0, 3.0–3.5, > 3.5–4.0, > 4.0–4.5, > 4.5–5.0, > 5.0–5.5 and > 5.5 mmol/l. In addition, similar analyses were performed in categories of first potassium values at ICU admission. Potassium variability was measured in (a) standard deviation (SD) of serum potassium measurements per patient, classified into 1st, 2nd and ≥ 3rd SD groups and (b) coefficient of variation (CV) (SD/mean potassium), classified into 4 groups: CV1 0–10%, CV2 10–20%, CV3 20–30%, CV4 > 30%. Finally, we combined groups of mean potassium and variability and analysed in-hospital mortality risk. Potassium values were determined by blood gas analyser (Radiometer ABL 600 or 800 FLEX, Copenhagen, Denmark) or laboratory standard. Investigated covariates were age, gender, mean blood glucose concentration, blood glucose variability measured as SD, minimum and maximum blood glucose concentration, mean sodium concentration, sodium SD, acute physiology and chronic health evaluation (APACHE) II score, mean maximum sepsis-related organ failure assessment (SOFA) score, diabetes, acute kidney injury, chronic kidney disease, atrial fibrillation, length of stay in the ICU, number of potassium measurements and pH value. APACHE II score was evaluated at ICU admission whilst SOFA score was determined on a daily basis by physicians. Covariates identified by ICD-codes like hypokalaemia, hyperkalaemia, chronic kidney disease, acute kidney injury, diabetes, arterial hypertension and atrial fibrillation were taken from the patients’ hospital records. ICD-10-GM derived covariates are listed in Additional file
1.
Endpoint
The primary endpoint was all-cause in-hospital mortality. Information on vital status for the study cohort was obtained from hospital records. The censoring date was May 31, 2018.
Statistical analysis
We used IBM© SPSS© Statistics version 24, Microsoft® Excel and SigmaPlot version 12.5 and GraphPad Prism 7 for data analysis and visualisation. Results are expressed as median with interquartile range or as absolute numbers with percentages. For the variables potassium, glucose, sodium and pH values, we calculated mean values for each patient and reported median and 25th/75th percentile of the cohort. A p value < 0.05 was considered statistically significant. For the statistical analysis of group differences, we performed Mann–Whitney U test, Kruskal–Wallis test and χ2 test. To verify confounding factors, we performed a binary logistic regression analyses stepwise unadjusted, adjusted for sex, age and mean pH value, and adjusted for clinically important possible confounders as listed above. 20-fold imputations and rerun of the binary logistic regression analyses was performed to confirm the results in spite of missing values. Survival curves for mean potassium range and potassium SD groups were performed according to Kaplan–Meier and tested by Log-Rank-test.
Potassium supplementation and in-hospital mortality
In this subanalysis, we included only patients from our cohort treated between 2013 and 2018 with medication records available to compare in-hospital mortality risk of patients with and without potassium supplementation. Potassium supplementation included oral and/or intravenous substitution for at least one time. In-hospital mortality analyses were performed in each group of mean potassium concentration and potassium variability.
Discussion
There is a lack of evidence for clear recommendations on potassium management in critically ill patients. We observed lowest mortality (3.7%) in ICU patients with a mean potassium range between > 3.5 and 4.0 mmol/l and low potassium variability. In-hospital mortality increased significantly for lower and higher potassium ranges including the normokalaemic range (3.5–5.0 mmol/l). More obvious, an increased potassium variability indicated by increased potassium concentration standard deviation or coefficient of variation during hospital stay is associated with an increased in-hospital mortality. In addition, we found an increased in-hospital mortality risk in ICU patients receiving potassium supplementation.
Our findings are in line with results of a recent systematic review, investigating the association of potassium concentration with mortality and occurrence of ventricular arrhythmias in patients after myocardial infarction in 12 studies. In those cardiac risk patients, they found an increased mortality risk when potassium concentrations were above 4.5 mmol/l [
12]. In agreement, in patients with atrial fibrillation, we observed the lowest mortality when mean potassium was > 3.5–4.0 mmol/l. The few previous studies investigated the association between potassium ranges, variability and mortality in ICU patients reported a J- and U-shaped association between potassium and mortality [
5,
13]. However, their classification lacked to define a tight potassium range with the lowest mortality. Our findings are in line with results reported by Hessels et al. and Uijtendaal et al. who showed that potassium variability was independently associated with outcome [
5,
13]. McMahon et al. [
6] investigated potassium concentrations at initiation of critical care in a comparable large ICU patient cohort. They stressed the importance of tight potassium regulation, as potassium concentrations between 4.5 and 5.5 mmol/l were associated with an increase in mortality risk. Unfortunately, they excluded patients with serum potassium concentrations < 4.0 mmol/l, which was in our analysis identified as the group with the lowest mortality. Unexpectedly, in our patient cohort, in-hospital mortality was significantly lower even in patients with mild hypokalaemic concentrations (3.0–3.5 mmol/l) as compared to 4.0–4.5 mmol/l. Contrary, in patients after myocardial infarction, potassium concentrations < 3.5 mmol/l were associated with an increased risk of ventricular arrhythmias [
12].
Potassium supplementation in critically ill patients is usually performed intravenously with a high risk for causing severe hyperkalaemia [
3,
14]. It may be an issue that potassium substitution is guided according to extracellular potassium measurements, which may not necessarily correlate with the intracellular potassium pool. In our cohort, mean potassium concentrations were slightly higher in patients receiving potassium supplementation. In addition, potassium substitution may increase potassium variability. We showed a significantly higher potassium variability in patients receiving potassium supplementation. Due to the lack of standardised protocols, decisions on when to start potassium supplementation are variable [
15]. In our cohort, it was performed frequently, but was not associated with mortality reduction. Contrary, there was a tendency for an increased in-hospital mortality risk in patients who received potassium supplementation in each potassium group.
pH value is known as one strong confounding factor of potassium as metabolic acidosis causes a potassium shift from the intracellular to the extracellular space [
16]. In our data, pH has a major impact on in-hospital mortality. After adjusting for age, sex and pH value in model 1 mean potassium levels and variability were independently associated with mortality. When evaluating multiple confounding factors in model 2, the results were not that clearly after adding pH value. Complex interactions in between the confounding factors may be responsible for a bias. Interestingly, when calculating the regression model divided by pH groups < 7.36, 7.36–7.44, > 7.44, low mean potassium levels (3.5–4.0 mmol/l) were confirmed as beneficial for the acidosis group (< 7.36). This indicates that especially in those patients, a mean potassium range > 3.5–4.0 mmol/l seems to be beneficial. This may be due to prevention of hyperkalaemia. A low potassium variability is associated with improved outcome after adjusting for multiple confounders including pH.
Interestingly, similar to our results, an association between mortality and mean glucose concentrations as well as glycaemic variability has been described by Krinsley [
17‐
19]. In addition, hypoglycaemia is a risk factor for death in ICU patients [
20]. However, the regression model 2 showed that our results are independent of mean, minimum or maximum glucose concentrations, as well as glycaemic variability, as risk factors for death [
17‐
20]. Thus, the association between potassium categories and variability can be discussed in context of glucose control by insulin therapy, which is beneficial for critically ill patients with stress-induced hyperglycaemia [
21‐
23]. In human physiology, the activation of Na
+K
+-ATPase by insulin is essential for the immediate potassium uptake to avoid hyperkalaemia for example after ingestion [
24]. Insulin leads as a GLUT 4-independent effect to the activation of the Na
+/K
+-ATPase and reduction of potassium concentrations [
25,
26]. Therefore, it may have underestimated positive effects on potassium concentrations as it may favour lower normokalaemic mean potassium concentrations, which were in our study associated with lower mortality. However, it remains unclear if glucose-independent insulin effects contribute to benefits of insulin therapy in ICU patients [
27]. In critically ill rabbits, normoglycaemia in combination with elevated insulin concentrations improved myocardial contractility [
27]. In the Leuven insulin trial, there were 6% more potassium measurements below 4.0 mmol/l, whilst hypokalaemia < 3.5 mmol/l was successfully avoided in the group targeting tight glucose concentrations 80–110 mg/dl [
28]. Accordingly, the number of potassium measurements in the range of 3.5–4.0 mmol/l was slightly higher in the group with better survival.
Furthermore, insulin therapy may influence potassium variability. A recent trial, investigating potassium concentrations before and after implementing tight glycaemic control in the ICU, found a reduced potassium variability after implementation of tight glycaemic control [
13]. Since glucose control by insulin goes along with changes in potassium concentrations, close monitoring of both, potassium and glucose, is of importance. A previous prospective study showed a reduction of hyper- and hypokalaemia in ICU patients using a computer-guided potassium regulation program [
29]. For the future, a continuous monitoring system for both glucose and potassium in combination with an adequate protocol could be an interesting approach to target optimal glucose and potassium concentrations with a minimum variability.
Limitations
This retrospective study shows associations but was not designed to prove causality. ICU patients in general were pooled together. However, there may be different potassium target ranges depending on the medical condition. Potassium concentrations have numerous interactions with medications, clinical and biological conditions, which were only partly considered in the multivariable regression model. The analysis of potassium supplementation has several limitations and potential biases. A major limitation is the use of mean potassium concentrations for group categorisation, whilst short-term hypokalaemic episodes are not reported. Thus, hypokalaemic episodes and not the potential subsequent potassium supplementation may have resulted in a mortality increase. Interestingly, initial potassium values are associated with in-hospital mortality in the same manner as mean potassium levels. in addition, detailed information on potassium doses as well as time point of potassium measurement and application are lacking. We performed an analysis on potassium supplementation but did not consider nutrition or other medications such as insulin, ACE-blockers, diuretics, β-antagonists or β-sympathomimetics.
Acknowledgements
TW is participant in the Clinician Scientist Program granted by the Berlin Institute of Health (BIH). We thank Robert Röhle M.Sc. (Institute of Biometry and Clinical Epidemiology, Charité—Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Charitéplatz 1, 10,117 Berlin, Germany) for statistical support.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.