Joint Effects of Sodium and Potassium Intake on Subsequent Cardiovascular Disease: The Trials of Hypertension Prevention Follow-up Study

Background  Previous studies of dose-response effects of usual sodium and potassium intake on subsequent cardiovascular disease (CVD) have largely relied on suboptimal measures of intake.

Methods  Two trials of sodium reduction and other interventions collected 24-hour urinary excretions intermittently during 18 months from September 17, 1987, to January 12, 1990 (Trials of Hypertension Prevention [TOHP] I), and during 36 months from December 18, 1990, to April 7, 1995 (TOHP II), among adults with prehypertension aged 30 to 54 years. Among adults not assigned to an active sodium reduction intervention, we assessed the relationship of a mean of 3 to 7 twenty-four–hour urinary excretions of sodium and potassium and their ratio with subsequent CVD (stroke, myocardial infarction, coronary revascularization, or CVD mortality) through 10 to 15 years of posttrial follow-up.

Results  Among 2974 participants, follow-up information was obtained on 2275 participants (76.5%), with 193 CVD events. After adjustment for baseline variables and lifestyle changes, there was a nonsignificant trend in CVD risk across sex-specific quartiles of urinary sodium excretion (rate ratio [RR] from lowest to highest, 1.00, 0.99, 1.16, and 1.20; P = .38 for trend) and potassium excretion (RR, 1.00, 0.94, 0.91, and 0.64; P = .08 for trend) but a significant trend across quartiles of the sodium to potassium excretion ratio (RR, 1.00, 0.84, 1.18, and 1.50; P = .04 for trend). In models containing both measures simultaneously, linear effects were as follows: RR, 1.42; 95% confidence interval (CI), 0.99 to 2.04 per 100 mmol/24 h of urinary sodium excretion (P = .05); and 0.67; 0.41 to 1.10 per 50 mmol/24 h of urinary potassium excretion (P = .12). A model containing the sodium to potassium excretion ratio (RR, 1.24; 95% CI, 1.05-1.46; P = .01) had the lowest Bayes information criterion (best fit).

Conclusion  A higher sodium to potassium excretion ratio is associated with increased risk of subsequent CVD, with an effect stronger than that of sodium or potassium alone.

Observational data and randomized trials have identified decreased blood pressure (BP) and reduced risk of hypertension among subjects with lower levels of sodium intake and higher levels of potassium intake.^1-8 Recent evidence suggests that long-term interventions aimed at sodium reduction or potassium substitution may lead to a reduced risk of cardiovascular disease (CVD).^9,10 However, dose-response relationships of sodium and potassium intake, as well as the sodium to potassium excretion ratio, with CVD have not been definitively determined.¹¹ The biologic interaction of sodium and potassium is of particular interest because it may have a dominant role in the pathogenesis of hypertension and in the development of CVD.¹² Findings from several studies^13-17 suggest that a high sodium to potassium excretion ratio is associated with increased BP and with CVD.

Observational studies^15,18 of subsequent CVD have used imperfect measures of sodium and potassium intake, namely, a single dietary recall or a single urinary excretion. Such imprecise measures may attenuate the estimated effect on subsequent risk. During the Trials of Hypertension Prevention (TOHP),^4,19 repeated measures of 24-hour urinary electrolyte excretion were carefully collected intermittently during an 18-month period (TOHP I) or during a 3-year period (TOHP II). Therefore, we were able to assess the relationship of usual long-term mean urinary sodium and potassium excretion, as well as the sodium to potassium excretion ratio, with subsequent CVD. A previous study⁹ compared subjects randomized to an active sodium reduction intervention or to a usual care control group in the same trials. To focus on long-term usual intake, the present analysis includes only those subjects who were not assigned to an active sodium reduction intervention.

TOHP I was designed to test the feasibility and efficacy of 7 nonpharmacologic interventions in reducing BP among persons with high-normal BP (prehypertension).²⁰ These included lifestyle interventions of weight loss, sodium reduction, and stress management, as well as nutritional supplement interventions of calcium, magnesium, potassium, and fish oil. Eligible participants were aged 30 to 54 years, had a mean diastolic BP between 80 and 89 mm Hg, and were not taking antihypertensive medication. Randomization occurred at 10 clinic sites from September 17, 1987, through October 27, 1988. Of 2182 randomized participants, 327 were assigned to a sodium reduction intervention and were excluded, leaving 1855 eligible for these analyses.

The follow-up period for the lifestyle interventions was 18 months. The 4 nutritional supplement interventions, including potassium, were conducted in 2 stages of 6 months' duration each with an intervening washout period. Follow-up visits occurred at 3 and 6 months for all interventions, as well as at 12 and 18 months for lifestyle interventions. Twenty-four–hour urinary sodium excretions were collected at baseline (2 collections) for all interventions; at 6, 12, and 18 months for the lifestyle interventions; and at each of 5 follow-up visits for the nutritional supplement interventions. For 178 participants randomized to the short-term active potassium supplement intervention, urinary excretions obtained during this 6-month intervention were not included in the mean. Collection of final visit data ended January 12, 1990.

TOHP II tested the effect of weight loss and sodium reduction interventions during a 3-year follow-up.²¹ The 2 × 2 factorial design included intervention groups of weight loss alone, sodium reduction alone, a combination of weight loss and sodium reduction, and a usual care comparison group. Participants were aged 30 to 54 years with a body mass index (BMI [calculated as weight in kilograms divided by height in meters squared]) representing 110% to 165% of desirable body weight. Eligible participants had a baseline diastolic BP of 83 to 89 mm Hg and a systolic BP of less than 140 mm Hg and were not taking antihypertensive medication. A total of 2382 participants were randomized into the trial from December 18, 1990, to March 31, 1992, at 9 clinic sites, with 1191 participants assigned to an active sodium reduction intervention. These were excluded, leaving 1191 participants eligible for this analysis.

Participants in each of the 4 study arms were seen at clinic visits every 6 months for 36 to 48 months, with 24-hour urinary excretions collected at baseline, 18 months, and 36 months. Additional urinary excretions were collected from selected participants at 6, 42, and 48 months. Final visits concluded April 7, 1995.

Observational follow-up for CVD began in 2000, approximately 10 years after the end of TOHP I and 5 years after the end of TOHP II. The cohort comprised 3009 participants who had not been randomized to an active sodium reduction intervention. Of these, 20 participants had CVD events during the trial periods, and 15 participants had no valid urinary excretion measures. These were excluded from the analyses, leaving 2974 participants, of whom 37 took part in both trials. For these 37 participants, follow-up time was assigned to TOHP I until the start of TOHP II, when follow-up time thereafter was assigned to TOHP II. Follow-up was conducted by the TOHP Coordinating Center at the Brigham and Women's Hospital, Boston, Massachusetts, and was approved by institutional review boards there and at participating clinic centers.

Follow-up was conducted centrally by mail and by telephone. Beginning in January 2000, initial questionnaires were sent out, with up to 4 additional requests, followed by telephone calls as needed. Questionnaires sought information about cardiovascular and other health outcomes, as well as limited information about weight, BP, and several health behaviors. Additional questionnaires were sent at 2-year intervals through early 2005, with interim annual postcards mailed to collect information about address changes and study end points.

The primary end point for the follow-up study was CVD, including stroke, myocardial infarction (MI), coronary artery bypass graft, percutaneous transluminal coronary angioplasty, and death from cardiovascular causes. On notification of the occurrence of a primary nonfatal end point, consent was sought to obtain medical records. The medical records were reviewed by a study physician (K.M.R.) blinded to treatment assignment to confirm reported events using standardized end point criteria. Of 301 nonfatal end points reported, including multiple end points per subject, consent for medical records was obtained for 239 (79.4%), and medical records were obtained for 213 (89.1%) of these. Of the reported end points with medical records, 193 (90.6%) were confirmed. Because of the high confirmation rate, all reported end points (193 subjects) were included in these analyses to enhance statistical power, excluding those that were disconfirmed following medical record review. A search of the National Death Index identified deaths through December 2003 among nonrespondents to the questionnaires.

We conducted an observational analysis of the relationship of usual sodium and potassium intake ascertained during the trial periods with subsequent CVD or mortality end points. Exposure was defined as the mean urinary excretion during the trial period, with outcomes ascertained following the trial. The primary outcome for this analysis was CVD following the end of the trial. We also examined the composite outcome of CVD or death from any cause, total mortality, and other combinations of the components of the primary end point, including stroke, MI, and coronary heart disease (CHD).

Usual intake of sodium or potassium or their ratio was calculated as the mean of available urinary excretion measures at 5 (lifestyle interventions) or 7 (nutritional supplement interventions) scheduled collections during 18 months in TOHP I and at 3 or up to 5 scheduled collections during 3 years in TOHP II. The mean urinary excretion levels were examined across categories of baseline characteristics and were tested using linear regression analysis. Because of large differences by sex, sex-specific quartile categories of urinary sodium and potassium excretion and their ratio were formed.

Cox proportional hazards regression models were used to assess the rate ratio (RR) (hazard ratio) for quartile categories of urinary excretion from the end of the trial period to the end of the follow-up period. Models were stratified by trial and were first adjusted for clinic, age, sex, race/ethnicity, and treatment assignment (model 1). Additional models adjusted for education status, family history, baseline weight, alcohol use, smoking, and exercise (model 2) and for changes in weight, smoking, and exercise during the trial periods (model 3). Other analyses controlled for BP, change in BP, and antihypertensive medication use during the trials.

To examine the shape of the relationship with CVD, we first fit linear terms for each urinary excretion measure and used penalized splines,²² or flexible nonlinear functions, to assess linearity. Models were fit for sodium and potassium separately, together, and as a ratio. No improvement was found when a log transformation for the urinary excretion variables was included or when an interaction term for sodium and potassium was included. The additional predictive information of the sodium to potassium excretion ratio was examined by comparing standardized coefficients for sodium, potassium, and the ratio^23,24 and by assessing model fit using the Bayes information criterion.²⁵ Possible modification of the linear effect by age, sex, race/ethnicity, baseline BMI, smoking, trial, and participation in an active weight loss intervention was examined and tested using interaction terms.

Despite averaging several repeated measures of sodium and potassium, the estimated results could be affected by remaining within-person variability and by the correlation between measures.^15,26,27 We estimated the joint between-individual and within-individual variance components using a mixed model for measures over time,^28,29 and we corrected for within-person variability using multivariate regression calibration.^15,30 Statistical analyses were performed using commercially available software (splines were fit in SPlus; Insightful Corporation, Seattle, Washington; all other analyses were conducted in SAS 9.1; SAS Institute, Cary, North Carolina).

Of 2974 eligible participants, 2207 provided follow-up information about study questionnaires, and 68 deaths occurred before the start of follow-up. Therefore, end point information was obtained for 2275 participants (2306 including duplicates in both trial phases), for an overall response rate of 76.5%. Response was higher among TOHP II participants (78.4%) than among TOHP I participants (75.3%). Among 2974 participants, name or other contact information was unavailable for 65 participants (2.2%), and 226 participants (7.6%) had addresses that could not be forwarded. One hundred two participants (3.4%) were unwilling to provide study information, and 306 participants (10.3%) did not respond. Of participants with a known address, 84.8% responded. Response varied considerably by study clinic site, but there was no difference by quartile of the mean urinary sodium excretion after adjustment for trial phase and clinic.

Among 2275 subjects with follow-up information, 193 (8.5%) experienced study end points, with a higher proportion among those participating in TOHP I (9.3%) vs TOHP II (7.3%) because of the longer follow-up period. First events included 68 MIs, 77 coronary revascularizations, 22 strokes (1 participant reported MI and stroke), 27 CVD deaths, and 51 non-CVD deaths. Eighty deaths occurred during the follow-up period, 2 of which occurred after a nonfatal event.

Participants had a mean of 4.8 urinary excretion measures (median, 5; range, 1-7) in TOHP I and a mean of 3.6 measures (median, 4; range, 1-5) in TOHP II. The overall median urinary sodium excretion was 158 mmol/24 h (interquartile range [IQR], 127-194 mmol/24 h). The median was higher among men (171 mmol/24 h) than women (134 mmol/24 h) and was higher among participants in TOHP II (172 mmol/24 h) than in TOHP I (149 mmol/24 h), likely associated with the higher weight criterion. The median urinary potassium excretion was 60 mmol/24 h (IQR, 48-73 mmol/24 h), and the median sodium to potassium excretion ratio was 2.8 mmol/24 h (IQR, 2.2-3.4 mmol/24 h), again with higher levels in TOHP II.

We compared the mean urinary excretion levels by categories of baseline characteristics among men and women separately because of the large sex differences (Table 1). The means varied significantly with several characteristics, including age, race/ethnicity, education status, baseline systolic BP, smoking, alcohol use, exercise, and baseline BMI. Differences were modest except for the positive relationship of BMI with urinary sodium excretion.

In Cox proportional hazards regression models, there was some evidence of a trend in CVD risk across quartiles of urinary sodium excretion, but this was not statistically significant and was attenuated after adjustment for baseline characteristics (Table 2). Extreme quartiles represent a difference in medians of 122 mmol/24 h in men and 95 mmol/24 h in women. The RR for participants in the highest vs the lowest urinary sodium excretion quartile was 1.20 (95% confidence interval [CI], 0.73-1.97; P = .38 for trend) after full adjustment. There was a significant inverse trend across quartiles of urinary potassium excretion, with a 45% reduction in risk among participants in the highest vs the lowest quartile (RR, 0.55; 95% CI, 0.35-0.87; P = .01 for trend) after adjustment for baseline characteristics and a 36% reduction after additional adjustment for lifestyle changes during the trials (P = .08 for trend). The trend across quartiles of the sodium to potassium excretion ratio was the strongest and was statistically significant, with an RR of 1.50 (95% CI, 0.94-2.39; P = .04 for trend) across extreme quartiles in fully adjusted analyses.

In analyses treating the mean urinary sodium excretion as a linear term, there was a trend toward increasing risk of CVD with increasing urinary sodium excretion that was attenuated in fully adjusted analyses (Table 3). A 100 mmol/24 h higher level of urinary sodium excretion was associated with a 25% increase in risk (RR, 1.25; 95% CI, 0.91-1.72; P = .18). The RR was similar after additional adjustment for baseline BP, change in BP, and use of antihypertensive medications during the trials (RR, 1.25; 95% CI, 0.90-1.72; P = .18). Spline plots suggested increasing risk with higher levels of urinary sodium excretion (P = .15, test for linearity), and tests for nonlinearity were nonsignificant (P = .82).

A linear inverse relationship of urinary potassium excretion with CVD risk was suggested but was not statistically significant in fully adjusted models (RR, 0.83; 95% CI, 0.53-1.29 per 50 mmol/24 h; P =.40) (Table 3). Spline plots did not show a significant linear trend (P =.62). Some curvature in the relationship was suggested, but CIs were wide, and the test for nonlinearity was not significant (P =.22).

The correlation between the mean urinary sodium and potassium excretion was 0.49, with slight variation by trial phase and sex. When urinary sodium and potassium excretion were included in the same model, effects strengthened for both measures. There was a 42% estimated increase in CVD per 100 mmol/24 h increase in urinary sodium excretion and a 33% risk reduction in CVD per 50 mmol/24 h increase in urinary potassium excretion (Table 3). Expressed per 1.00-SD unit, the RRs were 1.22 (95% CI, 1.00-1.48 per 55.5 mmol/24 h increase; P = .05) for sodium and 0.82 (95% CI, 0.64-1.05 per 25.0 mmol/24 h increase; P = .12) for potassium, with regression coefficients that were equal but opposite in direction. An interaction term for sodium and potassium was not statistically significant (P = .24).

However, the sodium to potassium excretion ratio demonstrated a statistically significant linear relationship to risk of CVD, with an RR of 1.24 (95% CI, 1.05-1.46; P = .01) (Table 3). Adjustment for BP, change in BP, and BP medication use during the trials produced little change in the estimated effect (RR, 1.25; 95% CI, 1.06-1.48; P = .009). The relationship seemed to be linear (Figure) (P = .01, test for linearity), with no evidence of nonlinearity in the relationship (P = .47, test for nonlinearity). Expressed per SD unit, the RR for CVD was 1.24 (95% CI, 1.05-1.46 per 1.01 U; P = .01). The model using the ratio demonstrated better fit according to the Bayes information criterion²⁵ than the model using both urinary excretion measures.

The repeated measures obtained during the trial allowed us to examine variation over time and to correct for measurement error in urinary sodium and potassium excretion during the trials. The correlations over time (reliability) of the urinary excretions were 0.35 for sodium, 0.40 for potassium, and 0.32 for the ratio. The within-person correlation of urinary sodium and potassium excretion was 0.47, and the between-person correlation was 0.52. When correcting for remaining measurement error in both urinary excretion measures, the RR for sodium increased to 1.84 per 100 mmol/24 h, and the RR for potassium decreased to 0.59 per 50 mmol/24 h. The corrected RR for the sodium to potassium excretion ratio increased to 1.35.

In fully adjusted analyses, estimated effects of the sodium to potassium excretion ratio were similar for CHD (including 154 MIs, coronary revascularizations, or CHD deaths; RR, 1.26; 95% CI, 1.06-1.50) and for stroke (including 21 strokes OR cerebrovascular deaths; 1.25; 0.86-1.82) but not total mortality (66 deaths; 1.01; 0.76-1.32). There was little evidence of modification of the RR in subgroup analyses (Table 4). All interactions were nonsignificant, and effects were consistent across subgroups.

In observational analyses of the mean urinary excretion during 1½ to 3 years, we found a suggested positive relationship of urinary sodium excretion and a suggested inverse relationship of urinary potassium excretion with risk of CVD, but neither was statistically significant when considered separately. Both measures strengthened when modeled jointly, with opposite but similar effects on risk. However, the sodium to potassium excretion ratio displayed the strongest and statistically significant association, with a 24% increase in risk per unit of the ratio that was similar for CHD and stroke and was consistent across subgroups.

Few epidemiologic studies have jointly examined the relationship of urinary sodium and potassium excretion or their ratio with subsequent CVD. Most large longitudinal studies^31,32 of potassium intake have used food frequency questionnaires, which generally do not fully capture sodium intake. However, the sodium to potassium excretion ratio has been found to be a somewhat stronger predictor of BP than urinary sodium excretion alone in data from TOHP I,¹⁵ INTERSALT,^13,14 and southern California.¹⁶ In a meta-analysis⁵ of trials of potassium supplementation, the effect on BP was modified by the mean sodium intake. In our study, we found that the sodium to potassium excretion ratio was the strongest of the 3 measures in predicting CVD and that the effect of urinary sodium or potassium excretion was enhanced when the other was included in the model, supporting the notion that the joint activity of these 2 electrolytes may have an important biologic role.¹²

Liew et al²⁴ demonstrate that using a ratio makes the implicit assumption that the regression coefficients for the 2 variables are equal in magnitude but opposite in direction. This was true in our data when urinary sodium and potassium excretion was expressed per 1.00-SD unit. Furthermore, the sodium to potassium excretion ratio provided a better fit to the data, and the addition of a multiplicative interaction term was not statistically significant.²³ It is possible that the ratio offers a correction for characteristics of the urine sample collection such as completeness during the 24-hour period and correlated measurement errors.²⁷ Whether the improved fit of the ratio is due to an inherent biologic synergism¹² or to an artifact of the urinary excretion measurements is unclear.

Previous studies^33-37 have largely relied on dietary records or recalls for estimating sodium intake, which tend to be less precise than urinary measures, and results have been inconsistent. Four studies have examined a single measure of urinary sodium excretion and CHD end points. One found an inverse relationship with MI in men with hypertension but not in women with hypertension,^38-40 another found a positive association with CHD in women only,⁴¹ and a third found a positive relationship with CHD and total mortality that seemed stronger in overweight men.⁴² Most recently, the Rotterdam Study⁴³ found no effect of sodium or potassium intake (or their ratio) on CVD or mortality except for an effect of the ratio on mortality among subjects with a BMI of 25 or higher. However, that study used a single overnight urine sample, which may have included measurement error and attenuated any effects. Exposure in the TOHP was calculated as the mean of 3 to 7 twenty-four–hour urinary excretions obtained during 1 to 3 years, reducing measurement error and within-person variability.^15,18

Fewer studies have examined potassium intake and CVD, with most attention paid to stroke. Higher dietary potassium intake has generally been associated with decreased stroke incidence or mortality,^23,31,44-46 although results are sometimes marginal.^32,47 Few investigations have examined urinary potassium excretion and risk of CHD.⁴¹ We did not find an inverse association of urinary potassium excretion with stroke in adjusted analyses, but this was limited by the small number of occurrences of strokes among our data. A stronger, although nonsignificant, association was found with total CVD.

Some observational studies have found effects of sodium and potassium intake on stroke or CVD that may be independent of effects on BP^36,42,44 or hypertension,^31,32 as in the present study. Urinary sodium excretion has been found to be positively related to urinary albumin excretion,^48-50 suggesting that the effect of sodium on CVD may be partially mediated by endothelial damage. Other proposed mechanisms for a cardiovascular effect of sodium independent of BP include a direct effect on left ventricular mass⁵¹ or increased blood flow and vascular reactivity with higher sodium exposure.^52,53 The present study found an effect of sodium on CVD outcomes even after controlling for BP. However, we did not measure BP during the follow-up period, so we cannot determine if the sodium observation is fully independent of BP.

Other limitations of the present study include a lack of complete follow-up among participants for nonfatal events. However, response did not seem to be associated with urinary excretion measures. We also did not have a full complement of CVD risk factors (such as lipids) for adjustment. In addition, we used urinary excretion as our exposure, which is assumed to adequately reflect dietary intake. We obtained no urinary sodium excretion measures during the follow-up period and cannot account for possible changes during this period.

To better represent long-term usual intake, the present analysis of the TOHP Follow-up Study data included only participants who were not in an active sodium reduction intervention. Previous independent analyses of the TOHP trials found a significant reduction in total CVD, as well as a suggested reduction in total mortality, among participants assigned to a sodium reduction intervention.⁹ A cluster randomized trial conducted among older residents of a veterans retirement home in Taiwan compared the effects of consumption of potassium-enriched salt, containing less sodium, with the effects of regular salt consumption among a control group and found a significant 41% reduction in CVD mortality among the experimental group.¹⁰ These data support our findings of reduced CVD risk among subjects with lower sodium intake, higher potassium intake, or both. The 2005 US dietary guidelines⁵⁴ recommend consumption of potassium-rich foods such as fruits and vegetables, as well as consumption of little salt. The totality of evidence suggests that lowering dietary sodium intake, while increasing potassium consumption, at the population level might reduce the incidence of CVD.

Correspondence: Nancy R. Cook, ScD, Division of Preventive Medicine, Brigham and Women’s Hospital, Harvard Medical School, 900 Commonwealth Ave East, Boston, MA 02215-1204 (ncook@rics.bwh.harvard.edu).

Accepted for Publication: June 8, 2008.

Group Information: A list of the Trials of Hypertension Prevention Collaborative Research Group was published in Arch Intern Med. 1997;157(6):661.

Author Contributions: Dr Cook had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Cook, Obarzanek, Cutler, Kumanyika, and Whelton. Acquisition of data: Cook, Buring, Rexrode, Appel, and Whelton. Analysis and interpretation of data: Cook, Obarzanek, Cutler, Buring, Rexrode, Kumanyika, Appel, and Whelton. Drafting of the manuscript: Cook. Critical revision of the manuscript for important intellectual content: Cook, Obarzanek, Cutler, Buring, Rexrode, Kumanyika, Appel, and Whelton. Statistical analysis: Cook. Obtained funding: Cook and Appel. Administrative, technical, and material support: Cook, Obarzanek, Cutler, Rexrode, Appel, and Whelton. Study supervision: Cook.

Financial Disclosure: None reported.

Funding/Support: TOHP I and II were supported by cooperative agreements HL37849, HL37852, HL37853, HL37854, HL37872, HL37884, HL37899, HL37904, HL37906, HL37907, and HL37924, and the TOHP Follow-up Study was supported by grant HL57915, all from the National Heart, Lung, and Blood Institute, National Institutes of Health.