Fasting serum potassium and long-term mortality in healthy men

The Oslo Ischemia Study is a comprehensive health survey established in 1972, aimed to examine risk factors for coronary heart disease and other cardiovascular diseases in a healthy middle-aged male population. A total of 2341 male employees in five companies with occupational health service in Oslo, Norway, aged 40–59 years, were invited to participate. Inclusion in the study required the absence of any medical diagnosis and no medication use for the past 14 days. Among those invited, 2014 (86%) men agreed to participate and completed the study protocol. At inclusion, all participants underwent a comprehensive review of the medical history, clinical and physical examinations (including measurements of body height and weight), a panel of blood tests, and a maximal exercise tolerance bicycle test. Information on lifestyle variables (e.g. smoking habits) was collected with a questionnaire. Detailed selection and implementation procedures have been reported previously [3, 12‐14].

Information about vital status and cause of death were obtained by linkage to the Cause of Death Registry, using the 11-digit personal identification number assigned all Norwegian citizens. The Cause of Death Registry covers all deaths in Norway (i.e reporting is mandatory by law) including deaths of Norwegians who died abroad. A physician determined the cause of death according to the current revision (7th, 8th, 9th or 10th) of the World Health Organization’s International Classification of Diseases (ICD). The underlying cause of death was used to identify deaths related to cancer (Neoplasms; ICD-10 C00-D48), cardiovascular disease (Diseases of the circulatory system; ICD-10 I00-I99) and other causes (details in supplementary Table 1).

At inclusion in the 1970s, all participants provided informed consent according to the Declaration of Helsinki. After increased awareness of privacy during the 1990s, the Data Protection Authority retrospectively reviewed the inclusion process for the Oslo Ischemia Study and judged participation to be in accordance with provisions on written informed consent. This study was approved by the Regional Committees for Medical and Health Research Ethics, Norway.

Blood samples for measurement of electrolytes were collected after at least 12 h fasting and 8 h abstaining from smoking. Repeated analysis of the blood samples has shown high accuracy [12]. In the present study, the variable of interest was serum potassium concentration. The continuous level of potassium was given per 1 unit (mmol/L) and per 1 standard deviation (SD) increase. Potassium levels below or above the normal range were categorized as hypokalemia (≤3.5 mmol/L) or hyperkalemia (> 5.0 mmol/L), respectively. The normal range of potassium was divided into three equally spaced categories (normo-low [3.6–4.0], normo-intermediate [4.1–4.5], and normo-high [4.6–5.0 mmol/L]). Due to the small number of observations in the hypo- and hyperkalemia range, potassium levels were also categorized as (low [≤4.0], middle [4.1–4.5], and high [≥4.6 mmol/L]).

The erythrocyte sedimentation rate (mm/hour) was determined by the Westergren method. In addition, fasting blood glucose (mmol/L), serum creatinine (μmol/L), triglyceride (mmol/L), total cholesterol (mmol/L) and sodium (mmol/L) were measured. Body mass index (BMI) was calculated based on the individual’s measured body height and weight at baseline (kg/m²). Blood pressure (mm Hg) was measured in the supine position, with a manual mercury sphygmomanometer. Physical fitness was measured as the total work (i.e. sum of the work performed in the bicycle test) divided by body weight (kJ/kg). Information on smoking habits was self-reported from a questionnaire asking about present and past smoking habits, categorized as “never”, “former”, or “present” smokers. We had complete variable information in all subjects except for fasting blood glucose and sodium, which was missing for 5 and 88 men, respectively. Of the 2014 participants, we excluded 15 men due to cancer diagnosis prior to baseline and 10 men for whom values for potassium were not obtained. Thus, 1989 men remained in the study sample for statistical analyses.

The primary outcome was cancer mortality. Further, cardiovascular death, death related to other causes and all-cause mortality were studied as secondary outcomes. The cohort of men was followed for death throughout 2017.

The absolute risk of death, cause-specific and all-causes, was presented as age-adjusted mortality rates per 1000 person-years. Graphically, the cumulative mortality by cause of death was presented for potassium in three categories, during 45 years of follow-up.

The relative risk of death was explored by using Cox proportional hazard models with age as the time scale. Results were presented as hazard ratios (HR) with 95% confidence intervals (CI). Adjustment was performed for BMI, systolic blood pressure, physical fitness, fasting blood glucose, sodium, triglycerides, cholesterol, creatinine, and sedimentation rate. Variables included in the multivariable model were chosen a priori based on potential confounding effects on the association between serum potassium and mortality, and what have been adjusted for in previous studies [5‐11, 15, 16]. The proportionality assumption was checked using Schoenfeld residuals. The models were stratified by smoking to allow for the difference in baseline hazards between never, former and present smokers. Likelihood ratio tests for interactions between potassium and the covariates yielded no significant findings. We tested for trend by modeling the variables as continuous. To illustrate the trends in HR over levels of potassium, restricted cubic spline function was performed with four knots chosen according to Harrell’s recommended percentiles [17].

To exclude the possibility that undetected morbidity had caused increased potassium (i.e. reverse causality) we performed restricted analysis, excluding the first 10 years of observation (including 122 deaths), leaving 1867 men in the analysis. Further, to explore the importance of smoking as effect modifier, we stratified the analyses of trends in HR by smoking status.

To overcome the difficulties in interpretation of HRs [18], we also calculated restricted mean survival times (RMST) as a supplement. The RMST at time t can be interpreted as the t-year life time expectancy [19] and were thus calculated for all-cause deaths. The RMST model was adjusted for the same variables as in the Cox model. The difference in RMST for middle and high levels was compared to low potassium levels for every 10 year (t = 10, 20, 30, 40 years) of follow-up.

All statistical tests were 2-sided, with a significance level of .05. All statistical analyses were performed using Stata, version 15 (StataCorp LLC).

Our results are consistent with the main findings of three previous studies based on data from the general population [5‐7]. Hughes-Austin et al. reported higher cancer mortality (HR 1.2, 95%CI, 1.0–1.5) in individuals with 4.5–4.9 mmol/L compared to those with 4–4.4 mmol/L, and a similar, but non-significant, increase was also observed for levels ≥5.0 [5]. Fang et al. reported a non-significant increasing trend in cancer mortality by serum potassium level [6]. The study by Wannamethee et al. found an increased cancer mortality (HR 1.8, 95% CI, 1.0–3.2) with higher levels of potassium (≥5.2 mmol/L) compared to lower levels (< 5.2 mmol/L), restricted to current smokers [7]. Our results add to previous studies by examining the continuous association in individuals free of any medication, and by extending follow-up to a median of 30 years.

Based on this cohort, we have previously shown a positive and linear relationship between serum potassium and cancer incidence through 30 years of follow-up [3]. The fact that we in the current study observe similar findings for cancer mortality, strengthens the conclusions of both studies. Most likely, the differences in cancer mortality are the result of different risk of being diagnosed with cancer.

We previously proposed three hypotheses that could explain the relationship between potassium and cancer [3]. Firstly, higher level of serum potassium may promote cancer development and growth through, for example immune mechanisms such as suppressing the T-cell activity [20]. Secondly, there may be some inherent individual factors that contribute to the regulation of serum potassium that also facilitate cancer processes. For example, some genetic variations in ion channels and pumps that are involved in potassium homeostasis appear to be associated with genes that have a role in cell proliferations and differentiation. Thirdly, a reverse causation may be observed if necrotic cancer cells - in an undiscovered cancer at baseline - released potassium extracellularly. However, a persistent increase in cancer mortality after excluding the first 10-years of observation reduces the likelihood of the last hypothesis.

Hoppe et al. have performed a systematic review and meta-analysis of observational studies assessing the association of abnormal serum potassium levels and cardiovascular mortality in the older general population. They reported, combined results of six studies, that high potassium levels were associated with a 1.4-fold increased cardiovascular mortality, while no association was detected for low potassium levels [8]. Our results, as shown in Fig. 2b, were in line with their findings, although the results seem to be less clear. Relevant to this comparison maybe that our men were completely medication free at baseline, and that they found the strongest association with higher potassium among users of diuretics [5, 6]. Possible reasons for the increased cardiovascular mortality could result at least partly from fatal ventricular arrhythmias in the course of hyperkalemia [8].

The positive linear association between potassium and death due to other causes is in accordance with the two previous publications, that observed a linear association [7], and an increased risk within the normal range (HR 1.5, 95%CI, 1.1–1.9 for 4.5–4.9 vs. 4.0–4.4 mmol/L) [5].

Regarding all-cause mortality, conflicting results have been reported, with a linear association [7, 9, 15], a U-shaped association [10, 16], association only for high potassium levels [5, 6], and no association [11] has been observed. The study most similar to ours is a large population-based Swedish cohort study, including healthy individuals in early mid-life, with a median follow-up of 29 years and with fasting measures of serum potassium [9]. They reported a linear association between potassium, per mmol/L increase, and all-cause mortality, HR 1.2, 95%CI, 1.1–1.3. Further, an interaction with serum sodium was found. We observed a slightly stronger linear association, HR 1.6, 95%CI, 1.4–1.8, when potassium was considered continuous, which was most pronounced for high potassium levels. We did not reveal any interaction with sodium. In addition to what was found in the Swedish study, we present RMST for all-cause mortality. The RMST quantified that the average life-years lost was 1.3 and 2.5 years in men with high potassium levels compared with low levels during 30 and 40 years, respectively.

We demonstrate that abnormal values of potassium are not common in a healthy population. Although our study suggests that normo-high levels and hyperkalemia may lead to increased mortality. Our data are observational and cannot demonstrate causality, but the results highlights need for further studies on the subject of mechanisms linking serum potassium to mortality.

There are certain strengths and limitations to our study. The Oslo Ischemia cohort is a comprehensively measured and apparently healthy population-based cohort without use of any medication during the last 14 days that has been followed for death through complete national registries for over 40 years, with no loss to follow-up. The cohort has shown similar cancer incidence to men of the same age, time-period and counties of residence [13]. Measurements of potassium were performed fasting, thus minor bias due to potassium intake was expected. However, as a limitation, the potassium level was measured at baseline only, which may both involve measurement error and imprecise estimates of long-term levels, thus leading to diluted association estimates in our study [21]. Outcomes could occur up to 45 years later. Thus, we report not on the immediate impact of potassium levels but the long-term impact by some inherent individual factors, which could be mediated by other events, such as changes in lifestyle variables. However, any bias by this would have been toward null. Residual confounding is a concern in all observational studies, and cannot be entirely ruled out. E.g. information about alcohol consumption would have been preferable. Smoking was self-reported and may suffer from errors in accuracy. We have adjusted for many potential confounders, which did not alter the associations. The strength of the association (risk ratio) of an unmeasured confounder with both the exposure and the outcome, as measured by the E-value, would have to be 1.6 to explain away the observed association between potassium (per SD) and cancer mortality [22]. The corresponding E-value would have to be 1.3 for cardiovascular mortality and 1.5 for other causes and total mortality as the outcome. The cohort was restricted to white Caucasian men, and the results from the present study may not be generalizable to female and other ethnicities. Previous interaction analysis by sex, however, has not revealed any gender differences [9].