Serum Sodium and Cognition in Older Community-Dwelling Men

Study Design

The study methods of the Osteoporotic Fractures in Men (MrOS) study have been described previously (18). Briefly, 5994 men 65 years of age or older were recruited from six geographic areas: Birmingham, Alabama; Minneapolis, Minnesota; Palo Alto, California; the Monongahela Valley near Pittsburgh, Pennsylvania; Portland, Oregon; and San Diego, California, between March of 2000 and April of 2002 (19). To be eligible, participants had to be able to walk unassisted, be free from bilateral hip replacement, and be able to provide self-reported data; they had to reside near a clinical site for the study duration, be free from medical conditions that (in the judgment of the investigator) would result in imminent death, and be able to provide informed consent. Among 5994 individuals who participated in baseline testing, 461 were missing laboratory data. Of the remaining 5533, 25 were excluded due to missing data for serum sodium, any cognitive testing (i.e., missing both modified Mini-Mental Status [3MS] examination score and Trail Making Test Part B [Trails B] time), and/or covariates (described below). An additional 73 participants were missing data for baseline cognitive impairment (i.e., missing 3MS score or Trails B time and not meeting the criteria for cognitive impairment for the opposite test), leaving 5435 participants to be included in the prevalent analyses. Participants were included in the analysis if they met criteria for cognitive impairment for one test (i.e., 3MS or Trails B time, with data for the opposite test missing). This decision was made because if criteria were met for cognitive impairment for one test, classification would not change regardless of performance on the opposite test; however, if criteria were not met for cognitive impairment for one test, it was not possible to classify a person as unimpaired without having information from the opposite test.

Of the 5435 participants, 3647 who did not have cognitive impairment at baseline (741 had baseline cognitive impairment) both attended in the follow-up visit (between January of 2005 and December of 2006) and had measurement of cognitive function, thus were considered for analysis of cognitive decline. Of the 3647, 36 were excluded due to missing data for cognitive impairment (i.e., missing 3MS score or Trails B time and not meeting the criteria for cognitive impairment for the opposite test), and the remaining 3611 were included in the cognitive decline analyses. All participants provided written informed consent, and this study was approved by the investigational review boards at the participating centers in adherence with the Declaration of Helsinki.

Study Variables

Serum sodium was measured in fasting baseline samples stored at −70°C and thawed for analysis at a central laboratory (Oregon Veterans Administration Clinical Laboratory, Portland, OR) using a Roche COBAS Integra 800 automated analyzer (Roche Diagnostics Corp., Indianapolis, IN). The analyzer was calibrated daily and the interassay coefficient of variation for serum sodium was 2.5% (20).

Cognitive function was evaluated at baseline (2000–2002) and follow-up (in 2005) by both the 3MS examination (21) and Trails B time (22). Cognitive impairment was defined as described previously (23–25). Cognitive impairment at baseline (prevalent analyses) was defined as a 3MS score <1.5 SD below the mean (<84) or Trails B time >1.5 SD above the mean (>223 seconds), on the basis of previously established criteria (23–25). Cognitive impairment at follow-up (cognitive decline analyses) was defined as a decrease in 3MS score or increase in Trails B time >1.5 SD of the mean change in completion score (>9) or time (>67 seconds), respectively.

Confounders related to serum sodium and cognitive impairment, all measured at baseline, were selected a priori as potential covariates, and all were measured at baseline. Baseline questionnaires and interviews were administered by trained and certified clinical staff. Race/ethnicity, education, smoking status, and alcohol intake were determined by self-report. Education was classified into five categories (high school or less, some college, college, some graduate school, and graduate school). Smoking was defined as current, former, or never. Alcohol intake was characterized as a dichotomous variable according to whether 12 or more drinks had been consumed in the past 12 months, on the basis of overall low intake in the MrOS cohort (18). Body-mass index was calculated using body weight measured using a balance beam or digital scale and height measured with a Harpenden stadiometer. Physical activity was measured using the Physical Activity Scale for the Elderly questionnaire (26).

Creatinine was measured using fasting baseline serum samples stored at −70°C and thawed for analysis at a central laboratory (Oregon Veterans Administration Clinical Laboratory, Portland, OR), using an enzymatic method calibrated with materials assayed by isotope-dilution mass spectrometry and an interassay coefficient of variation of 5.3% (27). eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (28). Analysis of serum glucose was performed using a hexokinase method at the University of Washington using previously unthawed, fasted samples stored at −70°C. The interassay coefficient of variation on the basis of blind duplicates was <3% (29). Serum glucose was measured as ancillary study and was missing in n=141 and n=81 in the cross-sectional and longitudinal analyses, respectively. History of diabetes, hypertension, stroke, chronic obstructive pulmonary disease, or cardiovascular disease (defined as history of myocardial infarction, angina, or heart failure) was determined by self-report of diagnosis by a doctor or other health care provider. Health-related quality of life was evaluated using a modified Medical Outcomes Study 12-Item Short Form Mental Health Composite Score, with a modification classifying health status according to five categories (excellent, good, fair, poor, or very poor) (30). Participants were asked to bring all of the medications they had taken in the past 30 days to the clinical center for data collection. All medications recorded by the clinics were entered into an electronic medications inventory (San Francisco Coordinating Center, San Francisco, CA). Each medication was matched to its ingredient(s) on the basis of the Iowa Drug Information Service Drug Vocabulary (College of Pharmacy, University of Iowa, Iowa City, IA) (31). BP was measured in rested, seated participants.

Statistical Analyses

The association of serum sodium with prevalent cognitive impairment and cognitive decline was assessed using logistic regression models. Serum sodium levels were evaluated by tertile, using tertile 2 predefined in the analysis to serve as the reference group due to being a mid-normal value, and on the basis of previous studies demonstrating an adverse association of both higher and lower serum sodium levels with clinical outcomes (11–13). It was determined a priori to use serum sodium tertiles to better explore the relation between the exposure and outcomes variables in a nonbiased way. We also examined serum sodium as a continuous predictor variable (per unit decrease).

The initial model was unadjusted, then multivariable adjusted models were performed to include age, race/ethnicity, and education (model 1); model 1 plus smoking status, alcohol intake, body-mass index, and physical activity (model 2); model 2 plus eGFR, cardiovascular disease, diabetes, hypertension, stroke, and chronic obstructive pulmonary disease (model 3); and model 3 plus quality-of-life measures and serum glucose level (model 4). Thiazide diuretic use was missing for n=211 and n=150 for the prevalent cognitive impairment and cognitive decline analyses, respectively, but was added to model 4 for those not missing as an additional sensitivity analysis. Loop diuretics and selective serotonin reuptake inhibitors were also considered for inclusion as covariates, but were omitted due to the low percentage of participants using each in the MrOS study (32,33). Additionally, heart failure was considered for inclusion as a covariate, but was omitted because only 5% of participants had prevalent heart failure.

As secondary analyses, linear regression was used to assess the association of baseline serum sodium level with baseline 3MS score and baseline Trails B time, as well as the longitudinal association of baseline serum sodium level with change in 3MS score and change in Trails B time. To improve the distributional properties of the outcome variables in these models, 3MS score and Trails B time were transformed as follows: log([110−3MS score]^½) and log(Trails B time), respectively. We also performed robust regression analysis considering potential outliers and leverages to justify the use of the linear regression models.

Other sensitivity analyses were performed categorizing serum sodium level on the basis of clinical cut-offs (hyponatremia: <136 mmol/L; normonatremia: 136–145 mmol/L; and hypernatremia: >145 mmol/L), and also by quartiles and more resolute categories of serum sodium level (<136, 136–138, 139–140, 141–142, 143–144, and ≥145 mmol/L). Finally, additional sensitivity analyses were run excluding those with serum sodium <136 mmol/L and excluding those using thiazide diuretics. We also examined death before last observed date of follow-up cognitive testing as a potential competing risk of cognitive decline, as determined by state death certificates (18).

Measures of cognitive function and covariates at baseline were summarized by serum sodium tertiles (as well as by clinical cut-offs and quartiles) and presented as mean (SD) for continuous variables and n (%) for categoric variables. Comparisons across tertiles (as well as across categories according to clinical cut-offs and quartiles) were made using a chi-squared test for categoric data and ANOVA for continuous variables.

Two-tailed values of P<0.05 were considered statistically significant for all analyses. All statistical analyses were performed using SAS version 9.4.