The Rotterdam Study
This case–cohort analysis formed part of the Rotterdam Study, a population-based prospective study among 7,983 men and women aged 55 years and older in the Netherlands [22]. The Medical Ethics Committee of the Erasmus Medical Centre Rotterdam approved the study, and written informed consent was obtained from all participants. From August 1990 until June 1993, a trained research assistant collected data on health, medication use, lifestyle, and risk indicators for chronic diseases during a home interview. Subjects were subsequently invited at the study centre for clinical examination and assessment of diet.
Assessment of diet
Subjects were interviewed at the study centre by a trained dietician, who used a validated, semi-quantitative food frequency questionnaire [23]. The intake of total energy, alcohol, macronutrients, and a large number of micronutrients was computed using Dutch food composition tables [24]. No information on salt use was obtained and therefore data on dietary sodium were considered unreliable for this analysis.
Clinical examination
Height and body weight were measured with the subject wearing indoor clothing without shoes. The body mass index was computed as weight divided by height squared. A trained research assistant measured sitting systolic and diastolic blood pressure twice with a random-zero sphygmomanometer after a 5-min rest, and values were averaged. Hypertension was defined as a systolic blood pressure ≥160 mmHg or diastolic blood pressure ≥95 mmHg or use of antihypertensive medication. Diabetes mellitus was considered present when the subject reported antidiabetic treatment, or when random or post-load plasma glucose levels were 11.1 mmol/l or higher. CVD was considered present in case of a verified history of myocardial infarction, stroke, coronary bypass grafting, or percutaneous transluminal coronary angioplasty. Serum total and HDL cholesterol level (mmol/l) were determined by standard laboratory methods [25].
Assessment of sodium and potassium excretion
Participants collected an overnight urine sample before visiting the research centre and recorded collection times on the jar. They were not aware that samples would be used for estimation of electrolyte intake. At the research centre, volumes were recorded, urines were swirled and 100 ml samples were taken. Samples were stored in plastic tubes at −20°C for future laboratory determinations. Urinary sodium, potassium and creatinine determinations were performed by Vitros® 250 (formerly Ektachem 250) Chemistry System (Johnson & Johnson, Ortho-Clinical Diagnostics Inc., Rochester, New York). Determination of electrolytes and creatinine were based on potentiometry and enzymatic conversion, respectively. Urinary sodium and potassium concentrations (mmol/l) were standardized to 24-h values using recorded collection times and urinary volumes (ml). In addition, urinary sodium/potassium ratio was computed.
Follow-up procedures
The present analysis is based on follow-up data collected from baseline (1990–1993) until 1 January 1998. Informed consent for collection of follow-up data was obtained from 7,802 participants (98%). Information on vital status was obtained at regular intervals from municipal population registries. General practitioners (GPs) used a computerized information system to record fatal and non-fatal events in the research area (covering 85% of the cohort). In the Netherlands, the GP forms the link to all specialized medical care and clinical events are unlikely to be missed by this follow-up procedure. Research physicians verified all information on incident events using GP records and hospital discharge letters. Events were coded independently by two physicians according to the International Classification of Diseases, 10th revision (ICD-10) [26]. Coded events were reviewed by a medical expert in the field, whose judgment was considered definite in case of discrepancies.
Myocardial infarction comprised ICD-10 code I21 and stroke comprised ICD-10 codes I60-I67. Both fatal and non-fatal incident events were recorded. For the present study, only first events were considered. Events followed by death within 28 days were classified as fatal. CVD mortality comprised fatal myocardial infarction, fatal stroke, sudden cardiac death and other forms of fatal CVD (ICD-10 codes I20-I25, I46, I49, I50, I60-I67, I70-I74, and R96).
Study population
Of 7,129 subjects who visited the research centre, 6,605 adequately performed a timed overnight urine collection for which collection times were recorded and volumes exceeded 150 ml. Of those, 5,531 had blood pressure readings and these subjects were eligible for the present analysis. We followed a case–cohort approach for efficiency reasons. Assessment of urinary sodium, potassium and creatinine excretion was performed in all subjects who died (n = 795, including 217 cardiovascular deaths), and in those who experienced a myocardial infarction (n = 206) or stroke (n = 181) during follow-up. A random sample of 1,500 control subjects was taken from the eligible cohort for assessment of electrolyte excretions. Urine samples could not be retrieved for 52 of these subjects, and data on urinary sodium, potassium and creatinine were thus obtained in 1,448 subjects. Dietary data were available for 1,205 subjects (83%) of the random sample, 518 subjects (65%) who died during follow-up, 157 subjects (72%) who died from CVD, 170 subjects (83%) with an incident myocardial infarction and 147 subjects (81%) with an incident stroke. Reasons for missing dietary data were participation in the pilot phase of the Rotterdam Study, low cognitive function, and logistic reasons, as described in more detail elsewhere [23]. Of the random sub-cohort (n = 1,448), 783 subjects (54%) were free of CVD and hypertension at baseline.
Data analysis
Pearson correlations were computed to examine inter-relationships between urinary and dietary measures of electrolyte intake and associations with total energy intake.
The association of urinary and dietary electrolytes with incident myocardial infarction, incident stroke, cardiovascular mortality and all-cause mortality was evaluated in a case–cohort design with standard Cox proportional-hazards models with modification of the standard errors based on robust variance estimates [27, 28]. We used the method according to Barlow in which the random cohort is weighted by the inverse of the sampling fraction from the source population. Members of the random cohort are included from baseline until failure or censoring, whereas cases outside the cohort are included at the time of their event. For the Cox models we used Proc MI and Proc MIanalyze, in conjunction with Proc Phreg (SAS 8.2).
Relative risks (RR) with 95% confidence intervals (95%-CI) were computed per 1 standard deviation increase in urinary sodium (mmol/24 h), urinary potassium (mmol/24 h) and dietary potassium intake (mg/day), and per 1 unit increase in urinary sodium/potassium ratio. Two-sided P-values below 0.05 were considered statistically significant. Adjustment was made for age, sex and, in urinary analyses, for 24-h urinary creatinine excretion (model 1). In a second analysis (model 2), additional adjustment was made for body mass index (kg/m2), smoking status (current, past, or never), diabetes mellitus (yes/no), use of diuretics (yes/no), and highest completed education (three categories). In a third analysis (model 3), dietary confounders were additionally adjusted for, i.e. daily intake of total energy (kJ), alcohol (g), calcium (g), and saturated fat (g). In the analysis for urinary sodium we additionally included urinary potassium in this model, and vice versa.
Analyses were repeated after exclusion of subjects with a history of CVD or hypertension to avoid biased risk estimates due to intentional dietary changes. Within this sub-cohort, a predefined stratified analysis of urinary sodium and urinary sodium/potassium ratio with cardiovascular and all-cause mortality was performed in subjects with a high body mass index (i.e., ≥25 kg/m2), using model 3.
Also in the sub-cohort free of CVD and hypertension, the distribution of 24-h urinary sodium excretion was divided into quartiles to be able to examine the relationship with all-cause mortality at extreme intakes. Quartiles of urinary sodium (cut-off levels: 66, 105 and 151 mmol/24 h) were entered categorically into the fully adjusted model (model 3), using the lower quartile as the reference.