Middle age serum sodium levels in the upper part of normal range and risk of heart failure 

Abstract

Aims

With increasing prevalence of heart failure (HF) owing to the ageing population, identification of modifiable risk factors is important. In a mouse model, chronic hypohydration induced by lifelong water restriction promotes cardiac fibrosis. Hypohydration elevates serum sodium. Here, we evaluate the association of serum sodium at middle age as a measure of hydration habits with risk to develop HF.

Methods and results

We analysed data from Atherosclerosis Risk in Communities study with middle age enrolment (45–66 years) and 25 years of follow-up. Participants without water balance dysregulation were selected: serum sodium within normal range (135–146 mmol/L), not diabetic, not obese and free of HF at baseline (N = 11 814). In time-to-event analysis, HF risk was increased by 39% if middle age serum sodium exceeded 143 mmol/L corresponding to 1% body weight water deficit [hazard ratio 1.39, 95% confidence interval (CI) 1.14–1.70]. In a retrospective case-control analysis performed on 70- to 90-year-old attendees of Visit 5 (N = 4961), serum sodium of 142.5–143 mmol/L was associated with 62% increase in odds of left ventricular hypertrophy (LVH) diagnosis [odds ratio (OR) 1.62, 95% CI 1.03–2.55]. Serum sodium above 143 mmol/L was associated with 107% increase in odds of LVH (OR 2.07, 95% CI 1.30–3.28) and 54% increase in odds of HF (OR 1.54, 95% CI 1.06–2.23). As a result, prevalence of HF and LVH was increased among 70- to 90-year-old participants with higher middle age serum sodium.

Conclusion

Middle age serum sodium above 142 mmol is a risk factor for LVH and HF. Maintaining good hydration throughout life may slow down decline in cardiac function and decrease prevalence of HF.

Structured Graphical Abstract

Analysis of 25 years follow-up data from Atherosclerosis Risk in Communities studies reveals increased risk to develop heart failure (HF) and left ventricular hypertrophy (LVH) among study participants with middle age serum sodium above 142 mmol/l. Left panels (beginning of the study): participants are divided into 4 groups based on serum sodium levels. Middle panels (follow-up: time-to-event analysis): increased incidence of HF in higher sodium groups with adjusted hazard ratio reaching 1.39 when serum sodium exceeds 143 mmol/l. Right panels (end of the 25 years follow-up: 70–90 year old participants): Increased prevalence of HF and LVH and accelerated hypertrophic left ventricular remodelling in higher sodium groups.

Audio Abstract

Heart failure, Left ventricular hypertrophy, Risk factors, Prevention, Hydration, Sodium

See the viewpoint for this article ‘Settling the controversy of salt substitutes and stroke: sodium reduction or potassium increase?’, by Franz H. Messerli et al., https://doi.org/10.1093/eurheartj/ehac160.

See the editorial comment for this article ‘Translating plasma sodium, stores, and hydration state from mouse to man’, by Friedrich C. Luft, https://doi.org/10.1093/eurheartj/ehac175.

Introduction

Heart failure (HF) is a major public health problem, especially in industrialized countries with ageing populations. About 6 million adults in the USA alone are living with HF as estimated in the 2021 American Heart Association statistical update¹ and an estimated 64.3 million people are living with HF worldwide.² In the USA, it is a leading cause of hospitalizations, readmissions, and outpatient visits at a cost of over $39 billion annually.³

Prevalence of the HF increases sharply with age with most deaths and prevalent cases in the USA and Europe occurring in an individuals older than 65 years.^2,4 Therefore, identification and implementation of preventive measures that are able to slow down degenerative changes within the heart and delay onset of HF become increasingly important. The analysis we present in the current study provides evidence for a hypothesis that chronic lifelong subclinical hypohydration, reflected in increased serum sodium concentration within normal range, promotes development of HF. These factors were not considered before in the studies addressing aetiology and pathophysiology of HF.² The hypothesis originated from studies utilizing a mouse model of chronic mild water restriction that elevated serum sodium concentration by 5 mmol/L.^5–7 In those studies, water restriction lasting from several weeks to lifetime promoted changes associated with cardiovascular risks: increased markers of coagulation⁷ and inflammation,⁶ activation of endothelial cells both in the microcirculation and coronary arteries and accelerated atherosclerosis.⁶ Water restriction yielded stable metabolism remodelling towards metabolic water production accompanied by increased food intake and energy expenditure and ultimately shortened mouse lifespan by 6 months (20%).⁵ Pathological examination of the mouse hearts revealed increased cardiac fibrosis.⁵

In healthy people, hypohydration is reflected in increased serum sodium concentration and tonicity.⁸ Normal serum sodium range, defined as the interval that 95% of a reference healthy population falls into, lies between 135 and 146 mmol/L. There are two mechanisms that maintain serum sodium in a narrow range in the healthy subject: thirst and antidiuretic hormone (ADH) release. Both these mechanisms are activated by hypohydration and are controlled by plasma tonicity that depends on the concentration of osmotically active plasma solutes with sodium and glucose being the main contributors. When plasma tonicity increases due to decreased water intake, central osmoreceptor cells are activated and ADH is released from the posterior pituitary gland.⁹ Antidiuretic hormone then acts on the kidney resulting in the excretion of a lower volume of more concentrated urine. In the absence of hyperglycaemia or renal failure, sodium concentration is the chief determinant of plasma tonicity representing 96–98% of its value of 275–295 mosmol/kg.⁸ Tonicity threshold that stimulates ADH secretion varies in a narrow range around 285 mosmol/kg in different studies,^8,10,11 which would correspond to ∼140–142 mmol/L of serum sodium. These processes are accompanied by decreased plasma volume and activation of renin–angiotensin system—a known contributor to pathogenesis of hypertension and pathological vascular and cardiac remodelling leading to HF.^12,13

In order to test the hypothesis about association between chronic hypohydration and HF, we analysed data from Atherosclerosis Risk in Communities (ARIC) study. ARIC study is an ongoing population-based prospective cohort study in which 15 792 45- to 66-year-old black (African American) and white men and women were enrolled from four US communities in 1987–89 and followed up for more than 25 years.¹⁴ We used serum sodium measured at Visits 1 and 2 that took place 3 years apart as a measure of study participants hydration habits. In our previous analysis, consistent with mouse studies, serum sodium was cross-sectionally associated with markers of coagulation and inflammation.⁵ In the current study, we analysed association between serum sodium and also other hydration measures such as water deficit, tonicity, and haematocrit at the beginning of the study with future development of HF and left ventricular hypertrophy (LVH). Our analysis revealed significant associations of HF and LVH risk with the hydration-related measures and identified serum sodium threshold of 142–143 mmol/L corresponding to water deficit of 1% body weight and tonicity of 290 mosmol/kg as a warning levels for increased risk of HF and LVH.

Methods

Dataset

We used data from the ARIC study. The data were obtained from the NHLBI Biologic Specimen and Data Repository Information Coordinating Center (BioLINCC). The datasets were redacted to remove personal identifiers to conform to the individual informed consent restrictions. Transfer of datasets was approved by the NIH Office of Human Subjects Research (OHSR) and was excluded from Institutional Review Board review as Not Human Subjects Research, based on the interpretation of 45 CRF 46 under ‘Research Involving Coded Private Information or Biological Specimens’ and Guidance on Engagement of Institutions in Human Subjects Research (16 October 2008).

Study population

ARIC is an ongoing population-based prospective cohort study in which 15 792 45- to 66-year-old black (African American) and white men and women were enrolled from four US communities in 1987–89. A detailed study design description has been published.¹⁴ Each community cohort was selected in the ARIC study by probability sampling to ensure that all individuals in the eligible age group had equal chances of being selected. There were three subsequent visits at approximately 3-year intervals (Visit 2 in 1990–92; Visit 3 in 1993–95; Visit 4 in 1996–98) followed by Visit 5 in 2011–13 (Figure 1). Participants have been contacted annually since baseline, to obtain information about hospitalizations and for additional data collection. The ARIC study protocol was approved by the institutional review board of each participating centre and informed consent was obtained from participants at each study visit.

Figure 1

Study analyses overview. After selecting 11 814 ARIC study participants based on exclusion criteria, time-to-event analysis for heart failure was performed with exposure variables related to hydration status measured at the beginning of the study. The associations were first assessed with continuous exposure variables followed by splitting the continuous ranges into groups and identifying thresholds for significant increase of heart failure risk. Applicability of the findings for older population was tested by repeating time-to-event analysis in 5185 participants who lived until 70–90 years old and attended Visit 5 evaluation. Association of middle age serum sodium with left ventricular hypertrophy and heart failure was assessed in the Visit 5 attendees by logistic regression analysis.

Outcome variables, exposure variables, and covariates

For current analyses, we used outcome variables, exposure variables, and covariates from the data collected and curated by the ARIC study investigators and obtained by us from BioLINCC data repository. The main outcome of interest was incident HF during a 25-year follow-up period. It is defined as self-reported incident HF that was not contradicted by a later physician HF survey. The methods used in the ARIC study for ascertainment of the self-reported HF events included annual phone interviews about interim hospitalizations, evaluations of hospital records for HF discharge codes, evaluation of death certificates for HF codes as the underlying cause of death. Follow-up time for those with incident HF event was the time from Visit 1 baseline exam until the incident event. Second outcome of interest was LVH diagnosis. The LVH diagnosis was made based on Cornell Definition (S V3 + R aVL > 28 mm for men and >20 mm for women). Cornell Voltage in mm was calculated as Cornell Voltage in UV (S V3 + R aVL) divided by 100.

The primary exposure variable was average serum sodium from Visits 1 and 2 that occurred 3 years apart. Serum sodium was used as a measure of hydration habits of the study participants. Several other hydration-related variables were tested in the current study: water deficit, tonicity, haematocrit, haematocrit/haemoglobin ratio, and potassium. See results section for information about derivation of these variables and their role in hydration assessment. We also created additional exposure variable related to salt intake by the study participants and called ‘Adding salt at the table’. It separates people who, according to dietary survey, had a habit to add salt to dishes at the table and those who did not use additional salt.

Several covariates were used for adjustment of statistical models and for exclusions. Sex and race were self-identified. Smoking status was self-reported. Blood pressure (BP) was measured 3 times after 5 min of rest and recorded as average of the two last measurements. High BP was defined as systolic BP > 140 mmHg or diastolic BP > 90 mmHg. Taking BP medications were ascertained based on current medications that participants were asked to bring to the visit or based on self-report. We created categorical variable for BP status containing four categories: (i) normal BP; (ii) high BP; (iii) normal BP/BP medications and (iv) high BP/BP medications. Prevalent HF at Visit 1 was defined as taking any medication for HF or qualification for Gothenburg criteria.¹⁵ Other variables that were used in this study analyses were plasma glucose (mg/dL, Visits 1 and 2), estimated glomerular filtration rate (eGFR) (mL/min/1.73 m², Visit 2), total cholesterol level (mg/dL, Visit 2), body mass index (BMI) (kg/m², Visit 1 and 2), haematocrit (%, Visits 1 and 2), haemoglobin (g/dL, Visits 1 and 2), and potassium (mmol/L, Visits 1 and 2). Detailed information about analytic measurements in blood samples, and about collection of information for numerical variables and their derivations for final databases is available at the ARIC study website (https://www2.cscc.unc.edu/aric/cohort-manuals).

Exclusions

Since the purpose of this analysis was to examine effects of hydration, we aimed to exclude people whose serum sodium could be affected by other factors in addition to amount of liquids they consume. Therefore, to avoid including people with possible abnormalities of water/salt balance regulation, we excluded people who had average sodium concentration from Visits 1 and 2 outside normal reference range of 135–146 mmol/L. We also excluded participants with plasma glucose level higher than 140 mg/dL at Visits 1 and 2, since hyperglycaemia, in spite of causing dehydration, results in decreased serum sodium concentration.¹⁶ Since obesity is known to alter distribution of body fluids and elevates serum sodium, we excluded participants with averaged BMI > 35 kg/m² at Visits 1 and 2.¹⁷ We also excluded participants who already had prevalent HF at Visit 1 and who developed HF by Visit 2. After these exclusions, 11 814 participants remained in the dataset (Figure 1). For each analysis, only participants who had all needed variables available were included leading to slight differences in number of people between individual analyses datasets.

Statistical analysis

Overview of the analyses performed in current study is given in Figure 1. We performed time-to-event analysis to assess association of middle age serum sodium and of other measures of hydration with risk of future HF. Due to long 25-year follow-up extending into older age (70–90 years), there was a high rate of mortality among the study participants (Figure 2A). Therefore, for all time-to-event analyses performed in this study, we used Fine–Gray subdistribution proportional hazard model accounting for competing mortality not related to HF.¹⁸ The models were adjusted for major HF risk factors: age, sex, race, hypertension status, smoking, kidney function (eGFR), total cholesterol and BMI.⁴ Covariates from Visit 2 were used for the adjustments to prevent possibility of reverse causation due to using averaged values from Visit 1 and Visit 2 for serum sodium and other hydration-related exposure variables. The results of the time-to-event analyses are presented as hazard ratios (HRs) with 95% confidence intervals (CIs).

Figure 2

Metrics of reduced hydration assessed at middle age are associated with increased risk to develop heart failure. (A) The cumulative incidence functions for heart failure and for mortality not related to heart failure over 25 years of follow-up in ARIC study participants. (B) Higher serum sodium within normal range of 135–146 mmol/L is associated with increased risk to develop heart failure. The model is adjusted for age as continuous variable and for heart failure risk factors: sex—male vs. female as reference; race—black vs. white as reference; current smoking vs. no smoking as reference used as categorical covariates, as well as body mass index, total cholesterol and estimated glomerular filtration rate used as continuous covariates. Categorical variable adjusting for blood pressure status contains four categories taking into account blood pressure measured values and use of blood pressure medications. High blood pressure is defined as systolic blood pressure >140 mm Hg or diastolic blood pressure >90 mm Hg. Lower panels show distribution histograms for continuous variables used in the model. (C) Hazard ratios for exposure variables related to hydration and salt intake resulting from Fine–Gray subdistribution PH models for risk of heart failure. The models were run with the same covariates as used in serum sodium model shown in (B). See Supplementary material online, Tables S2–S5 for full models results. Water deficit was calculated as (60 × (1 − (140 ÷ [Na⁺])); Tonicity 1: (2[Na⁺] + [Glucose]/18); Tonicity 2: (2[Na⁺] + [Glucose]/18 + 2[K⁺]). Lower panels show distribution histograms for the exposure variables used in the models. See Supplementary material online, Figure S1 for distribution histograms related to haematocrit and Supplementary material online, Tables S2–S5 for full models results.

We first evaluated continuous association of incident HF with hydration-related exposure variables (Figure 1). To identify clinically usable thresholds for increased HF risk, we subsequently applied the classification and regression trees (CART) algorithm in order to create categorical exposure variables by splitting continuous ranges of sodium and other hydration-related variables into maximally distinct groups based on cumulative HF rates.¹⁹ We then applied the same Fine–Gray subdistribution hazard models with the newly created categorical variables to identify exposure groups with significant associations with HF.

We then assessed association of middle age serum sodium with development of LVH and HF in a subset of the main analysis cohort: 5162 participants who lived into an older age and attended Visit 5 evaluation. For this analysis, we performed retrospective case-control analysis by running multiple logistic regression with LVH and HF diagnosis at Visit 5 as outcome variable, categorical serum sodium as exposure variable and along with same major HF risk factors from Visit 2 listed above as covariates.²⁰

Prevalence of HF and LVH at Visit 5 in relation to midlife serum sodium was calculated for all participants who attended Visit 5 (n = 5162) as a whole cohort and separately for four race/sex groups: white females (n = 2294), white males (n = 1918), black (African American) females (n = 579), and black (African American) males (n = 371). Prevalence was calculated as % of people with the disease in each of four middle age sodium groups.

The analyses were performed using the SAS (SAS Institute Inc., Cary, NC, USA) and SigmaPlot (Systat Software, San Jose, CA, USA). For Fine–Gray proportional hazard and logistic regression models, we used PHREG and LOGISTIC functions in SAS. The 3D mesh graphs were constructed using the SigmaPlot software with LOESS smoothing and rejection of outliers.

Results

Study population

Time-to-event analysis for HF was performed on 11 814 participants who remained after exclusions of people whose serum sodium concentration could be shifted by factors affecting water balance regulation and therefore would not correctly represent level of hydration (see methods for details). The characteristics of the participants are presented in Table 1. Average age at baseline Visit 2 was 57 years. During 25 years of follow-up, 1369 participants developed HF. The cohort consisted of 52% female, 78% white, and 22% black participants. Twenty-three percentage were smokers, 66% had normal, and 44% had high BP, 21% maintained their BP under control with BP medications and 6% had high BP in spite of taking BP medications. The participants had average BMI of 26.5 kg/m², total cholesterol of 209 mg/dL, and eGFR of 96 mL/min/1.73 m².

Table 1

Demographic and Visit 2 clinical characteristics of the ARIC study participants included in the time-to-event analysis for heart failure presented in Figures 2–4

Characteristic	All participants (N = 11 814)	Serum sodium groups, mmol/L
		135–139.5 (N = 2869)	140–142 (N = 6088)	142.5–143 (N = 1631)	143.5–146 (N = 1226)
Age, years—mean ± SD	57.0 ± 5.7	56.4 ± 5.9	57.0 ± 5.7	57.6 ± 5.6	57.7 ± 5.6
Incident HF—no. (%)	1369 (11.6)	294 (10.2)	686 (11.3)	206 (12.6)	183 (14.9)
Serum Na+, mmol/L—mean ± SD
Visit 1	141.1 ± 2.2	138.6 ± 1.5	141.1 ± 1.4	142.9 ± 1.3	144.4 ± 1.6
Visit 2	140.9 ± 2.1	138.7 ± 1.6	141.0 ± 1.4	142.5 ± 1.3	143.9 ± 1.6
Average of Visits 1 and 2	141.0 ± 1.8	138.6 ± 0.9	141.0 ± 0.7	142.7 ± 0.3	144.2 ± 0.7
Exposure variables related to hydration and serum sodium—mean ± SD
Water deficita, % BW	0.41 ± 0.78	−0.61 ± 0.41	0.42 ± 0.30	1.15 ± 0.10	1.72 ± 0.28
Tonicityb, mosmol/kg	287.6 ± 3.7	282.8 ± 1.9	287.6 ± 1.5	291.1 ± 0.7	293.9 ± 1.5
Haematocrit, %	42.1 ± 3.1	42.2 ± 3.1	42.2 ± 3.1	42.0 ± 3.1	41.7 ± 3.0
Haematocrit/haemoglobin	2.98 ± 0.06	2.97 ± 0.06	2.98 ± 0.06	2.99 ± 0.07	2.99 ± 0.06
Potassium, mmol/L	4.3 ± 0.4	4.3 ± 0.6	4.3 ± 0.4	4.3 ± 0.4	4.3 ± 0.4
Adding salt to food at the table—no. (%)
Yes	4552 (39)	1222 (43)	2331 (38)	595 (37)	404 (33)
No	7262 (61)	1647 (57)	3757 (62)	1036 (63)	822 (67)
Sex—no. (%)
Female	6130 (52)	1390 (48)	3102 (51)	888 (54)	750 (61)
Male	5684 (48)	1479 (52)	2986 (49)	743 (46)	476 (39)
Race—no. (%)
White	9270 (78)	2428 (85)	4846 (80)	1184 (73)	812 (66)
Black	2544 (22)	441 (15)	1242 (20)	447 (27)	414 (34)
Current smoker—no. (%)
Yes	2551 (23)	665 (25)	1287 (23)	343 (23)	256 (23)
No	8318 (77)	1976 (75)	4313 (77)	1171 (77)	858 (77)
Groups by BP status—no. (%)
Normal BP	7186 (66)	1748 (66)	3804 (68)	956(63)	678 (60)
High BPc	803 (7)	181 (7)	393 (7)	137 (9)	92 (8)
Normal BP/BP meds	2216 (21)	575 (22)	1078 (19)	322 (21)	241 (22)
High BP/BP meds	681(6)	141 (5)	333 (6)	99 (7)	108 (10)
BMI, kg/m2—mean ± SD	26.5 ± 3.7	26.3 ± 3.7	26.4 ± 3.7	26.6 ± 3.8	26.8 ± 3.8
Total cholesterol, mg/dL—mean ± SD	209 ± 39	205 ± 38	209 ± 38	214 ± 40	215 ± 40
eGFR, mL/min/1.73 m2—mean ± SD	95.8 ± 14.8	96.1 ± 14.6	95.7 ± 14.6	95.8 ± 14.6	95.6 ± 16.6
CKD (eGFR < 60) —no. (%)
Yes	181 (1.7)	54 (2)	78 (1.4)	21 (1.4)	28 (2.5)
No	10 686 (98.3)	2590 (98)	5521 (99)	1493 (98.6)	1082 (97.5)

For serum sodium, concentrations from Visits 1 and 2 as well as averaged concentration are shown. The table also provides number of HF events that occurred during follow-up period in each serum sodium group. See Supplementary material online, Table S1 for more information about exposure variables related to haematocrit.

BMI, body mass index; BP, blood pressure; BW, body weight; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; HF, heart failure, SD, standard deviation.

a

Water deficit is calculated as (60 × (1 − (140 ÷ [Na⁺]))

b

Tonicity is calculated as (2[Na⁺] + [Glucose]/18).

c

High BP is defined as systolic BP > 140 mmHg or diastolic BP > 90 mmHg.

Assessing baseline hydration status in ARIC study participants

As our purpose was to examine association between middle age hydration with HF risk, we analysed several exposure variables related to the participants hydration status. Our main exposure variable was serum sodium. It represents 95–98% of the serum tonicity—the main activator of ADH release during hypohydration leading to activation of water conservation mechanism. To convert sodium into a metric that more directly reflects hydration status, we also calculated water deficit from sodium values for each participant as % of body weight (60 × (1 − (140/[Na⁺])). Such calculation is used in clinical practice to estimate amounts of water needed to be infused to normalize elevated serum sodium in case of dehydration.²¹ We created two exposure variables for tonicity. Tonicity-1 was calculated based on Worthley equation (2[Na⁺] + Glucose (mg/dL)/18).²² For tonicity-2 variable, potassium was added (2[Na⁺] + Glucose (mg/dL)/18 + 2[K⁺]) to test for possible contribution of interrelationships between sodium and potassium.²³ We also tested potassium as a separate exposure variable because hyperkalemia often accompanies dehydration and therefore has a potential to be a measure for hydration status. To asses hypohydration based on decreased plasma volume rather that increased electrolytes,²⁴ we also tested haematocrit and haematocrit to haemoglobin ratio for the association with HF risk (see Supplementary material online, Results and Discussion).

Analysis of association with HF risk of serum sodium and other hydration-related exposure variables: continuous exposure variables

In this study, we used average serum sodium from Visits 1 and 2 as a measure of participants hydration habits. Main assumption that was made is that two measurements of serum sodium performed at Visits 1 and 2 represent an average level of serum sodium that each person would have for a long time and therefore can affect long-term health outcomes. This assumption was done based on reported analysis of clinical records demonstrating that for each particular person serum sodium concentration remained stable within a narrow range for 10 years.²⁵ It is also supported by our previous analysis of serum sodium in participants of the ARIC study showing that it was stable within 2–3 mmol/L between Visits 1 and 2.²⁶ Similar to serum sodium, average values from Visits 1 and 2 were calculated for all other tested hydration-related exposure variables: water deficit, tonicity, potassium, haematocrit, and haematocrit to haemoglobin ratio.

In Fine–Gray subdistribution proportional hazard model adjusted for major HF risk factors, serum sodium showed significant association with HF (Figure 2B). There was 5% increase in HF risk for each mmol/L sodium increase within 135–146 mmol/L range. The model was adjusted for age, sex, race, BP status, smoking, BMI, total cholesterol, and kidney function (eGFR). Other hydration-related exposure variables also showed significant associations with HF risk: water deficit (11% increased risk per 1% increase withing −2% to 2% range), tonicity-1 and tonicity-2 (20% increased risk per 10 mosmol/kg), and potassium (21% increase per 1 mmol/L within 3.5–5.5 range) (Figure 2C, Supplementary material online, Figure S1, Tables S2 and S4). Haematocrit within normal range, analysed separately for male and female, was not associated with risk of HF. However, haematocrit to haemoglobin ratio that is used to identify dehydration when its value exceeds 3, showed significant association with HF risk (28% increase per 0.1 unit within 2.8–3.2 range) (Figure 2C, Supplementary material online, Figure S1, Tables S3 and S4 and Results and Discussion).

Having found association of serum sodium and several hydration metrics with risk of HF, we then assessed possible involvement of salt intake in the observed associations. Salt intake was not specifically measured in the ARIC study. However, as a part of dietary survey, the participants were asked whether they normally add salt to the dishes at the table. According to the survey, 39% of the participants needed to add more salt to already prepared meals (Table 1). We reasoned that the positive response would identify individuals with higher salt intake. When used as exposure variable, ‘Adding salt at the table’ was associated with 15% increased risk of HF (HR 1.15, 95% CI 1.02–1.30) (Figure 2C and Supplementary material online, Table S4), that is consistent with negative cardiovascular consequences of high salt intake.²⁷ However, after adjustment of main model (Figure 2B) for ‘Adding salt at the table’ as additional covariate, serum sodium remained significantly associated with risk of HF with the same 5% increase in HF risk for each 1 mmol/L serum sodium increase (Supplementary material online, Table S5). In addition, higher proportion of people who were adding salt to prepared meals had their serum sodium in lower part of serum sodium range (Table 1). This analysis demonstrated that higher serum sodium is associated with HF risk independently of salt intake.

In summary, our analysis identified significant association of middle age serum sodium and several other hydration measures with increased risk of HF, estimating about 50% increase along sodium range of 135–146 mmol/L. The analysis also demonstrated that the association with serum sodium is independent of salt intake in the analysed cohort of participants without water/salt regulation abnormalities. Such independence supports our main assumption that serum sodium in this cohort can be used to estimate hydration status. This assumption is also supported by reported results from intervention trials demonstrating very small changes in serum sodium concentrations in response to sustained changes in salt intake within ranges identified for different regions by worldwide population surveys (from 6 to 14 g of salt per day, 8–9 g in the USA).^28,29 In the interventional trials, there was only about 0.4 mmol/L difference in serum sodium between groups with salt intake of 5–6 and 10–12 g/day.²⁸ Such a small effect of salt intake on serum sodium is explained by a very efficient renal excretion of excessive sodium³⁰ and by an ability to temporarily move sodium into tissue storages in osmotically inactive state.³¹ In comparison to salt intake, population differences in water intakes has substantially larger effects on serum sodium. According to worldwide surveys, the average amount of fluids consumed by different people on a regular basis varies from 0.7 to 3–4 L/day.^32,33 Adroque–Madias formula used for calculation of amount of fluids needed for sodium correction during hypernatremia,²¹ estimates that 1 L of water would decrease serum sodium in 80 kg men by about 3 mmol/L. These considerations imply that within the ranges that are observed in worldwide populations, differences in fluids intake have much more profound effects on serum sodium concentration than differences in salt intake.

Thresholds for serum sodium and other hydration-related exposure variables that identify people with increased risk of HF

After establishing association of elevated serum sodium and other metrics of hypohydration with increased risk of HF, we aimed to identify thresholds that could be used in clinical practice as prevention targets aimed at improving hydration in order to minimize HF risks. We applied the CART algorithm in order to split continuous ranges of sodium and other hydration-related variables into maximally distinct groups based on cumulative HF rates (Figure 3A and B). Figure 3B shows distributions of the participants among the groups created by the CART analysis for sodium, water deficit, tonicity, haematocrit to haemoglobin ratio and potassium. Heart failure rates for each group are shown above the histograms. We then constructed separate cumulative incidence functions (CIFs) for the groups for each exposure variable (Figure 3C). This analysis demonstrated significant separation of CIFs for sodium/water deficit and for tonicity but not for potassium and haematocrit to haemoglobin ratio.

Figure 3

Study participants can be divided into groups with significantly different cumulative incidence functions for heart failure if divisions are based on sodium, water deficit or tonicity. CART algorithm was used to split study participants into groups based on hydration-related variables that showed significant association with HF in Fine–Gray models shown in Figure 2. Cumulative heart failure incidence at the end of follow-up was used as outcome variable for the splitting algorithms. (A) Examples of CART algorithm outcome for group splitting based on sodium and tonicity 1 (2[Na] + [Glucose]/18). (B) Histograms showing distributions of the study participants according to tested variables. Groups identified by CART algorithm are shown in different colours. % of heart failure cases in each group are shown above each histogram. (C) Cumulative incidence functions for heart failure accounting for competing mortality plotted separately for each group. P-values from Gray’s test for equality of cumulative incidence functions are shown on the plots.

In the next step of our analysis, we run Fine–Gray subdistribution proportional hazard models with the categorical hydration-related exposure variables to identify the groups with significant increases in HF risk after adjustment for the same covariates that were used in models in Figure 2. Distributions of the covariates among the study participants according to 4 sodium groups are shown on Table 1. The analysis shows that age of participants did not differ between the sodium groups. Higher sodium groups had slightly higher BMI, higher levels of total cholesterol and more hypertensive participants that is consistent with previous analysis of crossectional associations of serum sodium with these measures in ARIC study participants.²⁶ Higher sodium groups also contained a larger proportion of female and black participants. All analyses presented in this study were adjusted for these covariates. Table 1 also provides information about serum sodium concentrations in four sodium groups, averaged as well as calculated separately for Visits 1 and 2, confirming that it was stable between these two visits. Adjusted Fine–Gray subdistribution hazard models run with the hydration-related categorical exposure variables showed 39% increase in HF risk for 143.5–146 mmol/L sodium group (HR 1.39, 95% CI 1.14–1.70), 31% increase in HF risk for water deficit of 1–2.4% body weight (HR 1.31, 95% CI 1.00–1.72) and 23% increase in HF risk for 290–300 mosmol/kg tonicity-1 group (HR 1.23, 95% CI 1.06–1.43) (Figure 4, Supplementary material online, Table S6).

Figure 4

High normal sodium (>143 mmol/L), water deficit (>1% body weight) and tonicity-1 (>290 mosmol/kg) are associated with significant increase in risk of heart failure in adjusted Fine–Gray subdistribution proportional hazard model accounting for competing mortality. The models are adjusted for the same heart failure risk factors described in legend for Figure 2B. Left panel: Full results for the model with sodium as exposure variable. Right panels: Overview of results from Fine–Gray subdistribution PH model shown for sodium on left panel but run with other exposure variables related to hydration. See Supplementary material online, Figures S6 and S7 for full models results.

In summary, time-to-event analyses with categorical variables allowed to find clear thresholds for significant increases of HF risk (Figure 4) that can be used to identify individuals at risk: >143 mmol/L for sodium, >1% body weight for water deficit and >290 mosmol/kg for tonicity (2[Na⁺] + [Glucose]/18). Haematocrit to haemoglobin ratio and potassium also showed significant association with risk of HF in adjusted models when used as continuous variable (Figure 2C and Supplementary material online, Table S4) supporting conclusion about association of hypohydration with increased risk of HF. However, they did not separate well into significantly different groups based on CIFs (Figure 3C) and did not provide thresholds for significant increase of HF risk (Figure 4 and Supplementary material online, Table S7). Therefore, potassium and haematocrit to haemoglobin ratio would not be very practical measures for identification of people at higher risk of HF due to mild hypohydration. See also Supplementary material online, results and discussion about effects of dehydration on haematocrit.

Time-to-event analysis of the association of serum sodium with HF risk in 70- to 90-year-old participants evaluated at Visit 5

Prevalence of the HF increases sharply with age with most deaths and prevalent cases in the USA and Europe occurring in individuals older than 65 years.^2,4 Therefore, in next analysis we aimed to find out if identified associations of midlife serum sodium with risk of HF persist in sub-cohort of participants who lived until older age. We repeated time-to-event analysis for HF on 5162 70- to 90-year-old study participants who attended Visit 5. Table 2 presents demographic and clinical characteristics of the Visit 5 cohort that includes information for whole cohort as well as for each sodium group. Comparing with main cohort, Visit 5 cohort contained larger proportion of female and white participants and smaller proportion of participants who have already had high BP at Visit 2. There were no major shifts in distribution of variables among sodium groups. During follow-up period, 480 of Visit 5 attendees developed HF (Table 2). Plotting CIF for HF events demonstrated that, similar to results from main cohort, larger proportion of people in higher sodium and tonicity-1 groups were developing HF (Supplementary material online, Figure S2A) in spite of similar average ages of participants in these groups (Table 2). In Fine–Gray subdistribution hazard models, there was 53% increase in HF risk for 143.5–146 mmol/L sodium group (HR 1.53, 95% CI 1.09–2.15) and 44% increase in HF risk for 290–300 mosmol/kg tonicity-1 group (HR 1.44, 95% CI 1.12–1.85) (Supplementary material online, Figure S2B).

Table 2

Demographic and clinical characteristics of the ARIC study participants who attended Visit 5 and had left ventricular hypertrophy evaluation

Characteristic	All Visit 5 attendees (N = 5162)	Serum sodium groups, mmol/L
		135–139.5 (N = 1278)	140–142 (N = 2701)	142.5–143 (N = 707)	143.5–146 (N = 476)
Age (Visit 5), years—mean ± SD	75.8 ± 5.3	74.8 ± 5.2	75.9 ± 5.3	76.6 ± 5.1	76.7 ± 5.2
HF—Visit 5—no. (%)	480 (9.3)	92 (7.2)	250 (9.3)	78 (11.0)	60 (12.6)
LVH—Visit 5—no. (%)	311 (6.5)	50 (4.1)	148 (5.9)	59 (9.1)	54 (12.5)
Serum Na+, mmol/L—mean ± SD
Visit 1	141.0 ± 2.2	138.6 ± 1.5	141.0 ± 1.4	142.9 ± 1.3	144.3 ± 1.7
Visit 2	140.9 ± 2.1	138.7 ± 1.5	141.0 ± 1.4	142.5 ± 1.3	143.8 ± 1.6
Average of Visits 1 and 2	140.9 ± 1.8	138.6 ± 0.9	141.0 ± 0.7	142.7 ± 0.2	144.1 ± 0.7
Water deficit, % BW	0.39 ± 0.75	−0.59 ± 0.39	0.43 ± 0.30	1.14 ± 0.10	1.68 ± 0.27
Tonicity, mosmol/kg	287.4 ± 3.6	282.8 ± 1.9	287.6 ± 1.5	291.0 ± 0.7	293.7 ± 1.4
Adding salt to food at the table—no. (%)
Yes	1968 (38)	542 (42)	1024 (38)	243 (34)	159 (33)
No	3194 (62)	736 (58)	1677 (62)	464 (66)	317 (67)
Sex—no. (%)
Female	2873 (56)	697 (55)	1470 (54)	408 (58)	298 (63)
Male	2289 (44)	581 (45)	1231 (46)	299 (42)	178 (37)
Race—no. (%)
White	4212 (82)	1127 (88)	2230 (83)	539 (76)	316 (66)
Black	950 (18)	151 (12)	471 (17)	168 (24)	160 (34)
Current smoker—no. (%)
Yes	834 (16)	222 (18)	420 (16)	114 (16)	78 (17)
No	4221 (77)	1032 (82)	2214 (84)	586 (84)	389 (83)
Groups by BP status—no. (%)
Normal BP	3721 (73)	951 (76)	1973 (75)	478 (68)	319 (68)
High BPa	293 (6)	59 (5)	149 (6)	55 (8)	30 (6)
Normal BP/BP meds	871 (17)	211 (17)	432 (16)	132 (19)	96 (21)
High BP/BP meds	178 (4)	35 (3)	84 (3)	35 (5)	24 (5)
BMI, kg/m2—mean ± SD	26.5 ± 3.6	26.2 ± 3.6	26.5 ± 3.5	26.5 ± 3.7	26.8 ± 3.7
Total cholesterol, mg/dL—mean ± SD	208 ± 37	203 ± 36	208 ± 37	212 ± 39	213 ± 39
eGFR, mL/min/1.73 m2—mean ± SD	97.2 ± 13.3	97.7 ± 12.7	96.9 ± 13.4	97.6 ± 12.6	97.3 ± 14.7
CKD—no. (%)
Yes	32 (0.6)	9 (0.7)	17 (0.6)	0 (0)	6 (1.3)
No	5025 (99.4)	1247 (99.3)	2615 (99.4)	700 (100)	463 (98.7)

The data from this cohort are included in the analysis presented in Figures 5 and 6. Covariates are from Visit 2 if not indicated otherwise.

BMI, body mass index; BP, blood pressure; BW, body weight; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; HF, heart failure, LVH, left ventricular hypertrophy; SD, standard deviation.

a

High BP is defined as systolic BP > 140 mmHg or diastolic BP > 90 mmHg.

Middle age serum sodium and risk of LVH and HF assessed in Visit 5 cohort by retrospective case-control analysis

To furthers explore associations between middle age serum sodium with long-term cardiac outcomes, we performed retrospective case-control analysis²⁰ that evaluated association of serum sodium with LVH in 4961 70- to 90-year-old participants who attended Visit 5 and had LVH evaluation. Left ventricular hypertrophy is the thickening of the walls of the heart’s main pumping chamber (left ventricle) that slowly develops over time to compensate for the decreased ability of the heart to pump blood and precedes HF diagnosis.³⁴ At Visit 5, LVH was assessed electrocardiographically by Cornell voltage. In adjusted multivariable logistic regression analyses, midlife serum sodium in 142.5–143 mmol/L range was associated with 62% increase in odds to have LVH diagnosis at Visit 5 (OR 1.62, 95% CI 1.03–2.55, P = 0.037) and serum sodium in 143.5–146 mmol/L with 107% increase (OR 2.07, 95% CI 1.30–3.28, P < 0.001) (Figure 5A). Serum sodium within 143.5–146 mmol range was also associated with 54% increase in odds of HF in Visit 5 cohort (OR 1.54, 95% CI 1.06–2.23) (Figure 5A). Consistently, Cornell voltage used for LVH diagnosis at Visit 5 was higher in participants who had higher serum sodium at baseline and was increasing faster with age indicating accelerated hypertrophic left ventricular remodelling (Figure 5B). Similar acceleration of left ventricular hypertrophic changes upon elevation of serum sodium is seen when Cornell voltage change between Visits 1 and 5 is calculated for each person and plotted against age at Visit 5 (Figure 5C). These associations are reflected in increased prevalence of HF and LVH in 70- to 90-year-old participants in higher middle age serum sodium groups when all study participants are analysed together as well as when they are separated into groups based on sex and race (Figure 6).

Figure 5

Higher serum sodium at Visits 1 and 2 is associated with increased risk of left ventricular hypertrophy and heart failure at Visit 5 at ages 70–90 years. Only participants from original analysis cohort who attended Visit 5 and did not have left ventricular hypertrophy at Visit 2 were included in the analysis (n = 4961). (A) Retrospective case-control analysis of factors associated with increased odds to develop left ventricular hypertrophy and heart failure: multivariable logistic regression. The model is adjusted for same covariates described in Figure 2B legend. (B) 3D Mesh Plot, visualizing Cornell voltage criteria values obtained at Visit 5 for left ventricular hypertrophy diagnosis as functions of serum sodium concentration at Visits 1 and 2 and age at Visit 5. Participants with serum sodium in upper half of normal range have higher Cornel voltage that increases with age at higher rate. (C) 3D Mesh Plot, visualizing change in Cornell voltage values between Visits 1 and 5 for each study participant as functions of serum sodium at Visits 1 and 2 and age at Visit 5. Size of left ventricle increases more and at younger age in participants whose middle age serum sodium was higher.

Figure 6

Prevalence of left ventricular hypertrophy and heart failure in ARIC study participants at Visit 5 depending on average serum sodium concentration measured at Visits 1 and 2. Shown is percent of people (70–90 years) who had left ventricular hypertrophy and heart failure diagnosis at the time of Visit 5 examination depending on their serum sodium measured at Visits 1 and 2 (45–66 years). Higher sodium is associated with higher prevalence of left ventricular hypertrophy and heart failure in all Visit 5 participants (left panels) and in groups created based on gender and race (right panels): W/F, white females; W/M, white males; B/F, black (African American) females; B/M, black (African American) males.

Sensitivity analysis

For sensitivity analysis, we repeated the analyses performed in currents study for middle age serum sodium as exposure variable for participants who were not taking any anti-hypertensive medications at baseline (n = 7954). Cumulative incidence functions showed significantly higher rate of incident HF for participants whose serum sodium exceeded 142 mmol/L (P = 0.0085) (Supplementary material online, Figure S3A). In time-to-event analysis for HF, serum sodium >143 mmol/L was associated with 29% increased risk for whole cohort (HR 1.29, 95% CI 0.98–1.69, P = 0.069), and with 54% increased risk for 3999 Visit 5 attendees (HR 1.54, 95% CI 1.02–2.34, P = 0.041) (Supplementary material online, Figures S3B and C). In logistic regression analysis of Visit 5 attendees, serum sodium >143 mmol/L was associated with 210% increased odds of LVH (OR 2.10, 95% CI 1.16–3.44, P = 0.012) and 65% increased odds of HF (OR 1.65, 95% CI 1.05–2.58, P = 0.029) (Supplementary material online, Figure S4).

Discussion

The main finding of this study is that it identifies middle age serum sodium and other measures of hydration in the upper part of normal reference ranges as a new risk factor for future HF and LVH. The analysis was performed on data from ARIC study, a large population-based prospective cohort study composed of four community-based US cohorts that enrolled 15 792 middle age participants (45–64 years) and followed them for 25 years (70–90 years).¹⁴ Hydration-related measures that were analysed included serum sodium and water deficit, serum potassium, tonicity, haematocrit, and haematocrit to haemoglobin ratio. The analysis presented in the current study not only revealed significant associations of hydration-related variables with risks of HF and LVH, but, more importantly, it identified thresholds that can be used in clinical practice to identify individuals at increased risk. Among the variables that showed significant associations with HF, serum sodium represents a most convenient measure for use in clinical practice due to its wide availability and regular use. In Fine–Gray subdistribution proportional hazard models, risk of incident HF was increased by 39% if serum sodium exceeded 143 mmol/L, corresponding to water deficit of 1% body weight and tonicity of 290 mosmol/kg. In retrospective case-control analysis performed on 70- to 90-year-old attendees of Visit 5, middle life serum sodium of 142.5–143 mmol/L was associated with 62% increase in odds to develop LVH, and serum sodium above 143 mmol/L was associated with 107% increase in odds to develop LVH. These associations are reflected in increased prevalence of HF and LVH in 70- to 90-year-old participants in higher middle age serum sodium groups (Structured Graphical abstract, Figure 1). The findings suggest chronic subclinical hypohydration as a new modifiable risk factor for HF and LVH. Future studies are needed to confirm this hypothesis in randomized controlled clinical trials.

Strengths of the current study are that analysis was performed on a large, well-characterized, randomly sampled population with lengthy (25 years) follow-up in a prospective study designed to examine cardiovascular disease incidence and risk factors with very carefully documented outcomes.¹⁴ Our study identified habitual hypohydration as a new risk factor for HF that was not considered before. The analyses found clear serum sodium threshold of 142 mmol/L for the increased risk, which would allow easy identification of people at risk even at early age before any other risk factors develop. Worldwide surveys report that the average amount of fluids consumed by different people on a regular basis varies from 0.7 to 3–4 L and that a large proportion of people are chronically hypohydrated.^32,33 Differences between people could originate from varying drinking habits due to family traditions or different thirst perceptions that could be related to genetics. Recommendations on daily fluid intake vary from 1.6 to 2.1 L for women and 2 to 3 L for men.^35,36 According to the worldwide surveys, many people do not meet even the lower ends of these ranges and are chronically hypohydrated.^32,33 From the public health perspective, adding recommendations for optimal fluid intake in the same portfolio with recommendations for salt intake ^27,37 to optimize overall water and salt balance, could attract more public attention for good hydration as a preventive measure for long-term cardiovascular health outcomes.³⁸

The main limitation of our study is its observational nature resulting in the possibility of residual confounding. Therefore, the causative relationships between serum sodium (hydration) and HF need to be confirmed in randomized controlled clinical trials. This limitation of the observational studies is reduced to some degree in our case, because the idea for this analysis originated from a well-controlled mouse study in which lifelong water restriction promoted cardiac fibrosis.⁵ Another limitation is availability of serum sodium measurements only for Visits 1 and 2. Additionally, lack of detailed data, did not allow us to assess effects of different classes of anti-hypertensive medications on the analyses outcomes.

The substantial strength of our study is that preventive measures based on these findings can be implemented immediately by reinforcing already existent recommendations for optimal water intake.^35,36 These preventive measures could include evaluation of patients hydration habits during regular physical exams and providing information to the people about recommended amounts of fluids to consume on a daily basis. Serum sodium concentration (ordered regularly as part of basic metabolic panel) above 142 mmol/L could serve as a reference to identify patients who would benefit the most from such evaluation.

Supplementary material

Supplementary material is available at European Heart Journal online.

Acknowledgements

The authors thank the staff and participants of the ARIC study for their important contributions. Data from the ARIC study were obtained from the National Heart, Lung, and Blood Institute Biologic Specimen, and Data Repository Information Coordinating Center. We thank Dr. Douglas Rosing (National Heart, Lung, and Blood Institute) for helpful comments and suggestions that improved the manuscript.

Funding

This work was supported by Intramural Research programme of the National Heart, Lung, and Blood Institute (NHLBI): the National Institutes of Health grant ZIA-HL006077-10. The ARIC study has been funded in whole or in part with federal funds from the NHLBI; the National Institutes of Health (NIH); and the Department of Health and Human Services, under contract numbers HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700005I, and HHSN268201700004I.

Data Availability

Anonymized data from the ARIC study are available through the National Heart, Lung, and Blood Institute Biologic Specimen, and Data Repository Information Coordinating Center. Interested researchers may additionally contact the ARIC study Coordinating Center to access the study data.

Author notes

Conflict of interest: The authors declare that there is no conflict of interest