Dietary sodium, potassium, sodium–potassium ratio and lifestyle factors

Table 1 provides baseline characteristics of the cohort according to sex and categories of estimated dietary intakes of sodium, potassium and sodium–potassium ratio. The median daily intakes of sodium and potassium were 2,812 mg and 3,364 mg for men, and 2,192 mg and 2,916 mg for women, respectively. Participants with greater sodium intake in both sexes were more likely to have diabetes, higher BMI and intake of total energy, be physically active, be current smokers or report poor or fair health, and to have lower educational level, lower HEI-2015 score, and higher intake of fiber, fruit and vegetable.

Table 1 Baseline characteristics of cohort participants according to categories of sodium and potassium intake and the sodium–potassium ratioa, bDietary sodium and overall and cause-specific mortality

Participants in this cohort had a median age of 62 years (interquartile range [IQR]: 57–66) at baseline and a median follow-up of 15.5 years (IQR: 15.5–15.8). During a total of 6,009,748 person-years of follow-up, there were 77,614 deaths (49,297 men and 28,317 women), including 22,228 from CVD (14,647 men and 7,581 women), of which 17,901 were from heart disease (12,178 men and 5,723 women) and 3,748 from stroke (2,153 men and 1,595 women), 28,099 from cancer (17,816 men and 10,283 women), 5,606 from respiratory disease (3,135 men and 2,471 women), 2,923 from infectious disease (1,796 men and 1,127 women), 2,614 from injuries and accidents (1,878 men and 736 women), and 16,144 from other causes combined (10,025 men and 6,119 women).

Figure 1a demonstrates the multivariable-adjusted associations between dietary sodium and mortality, with the HRs and corresponding 95% CIs of the cubic-restricted splines shown in the supplemental data (Additional file 1: Tables S3 and S4). Our results show substantially the same associations based on minimally adjusted and parsimoniously adjusted models (Additional file 1: Figures S1 and S4). Table 2 and supplemental data (Additional file 1: Table S5) present the estimated HRs and their 95% CIs based on quintiles, per 500 mg or per SD change in sodium intake for the associations with the risk of overall mortality and cause-specific mortality. We found statistically significant non-linear associations between sodium intake and overall mortality in men and women (Pnonlinearity < 0.0001), with the nadir of mortality occurring at intake levels of approximately 2,300 mg for men and 1,700 mg for women (Fig. 1a). Among participants with a daily sodium intake of 2,000 mg and above, a higher intake of sodium was associated with an increased risk of overall mortality for both men and women (fifth versus first quintile, HR for men = 1.06, 95% CI: 1.02, 1.09; HR for women = 1.10, 95% CI: 1.04, 1.16; both Ptrend ≤ 0.0002; Table 2). With the daily intake range of 2,000- < 3,000 mg as the reference level for men (category 2), an intake of ≥ 5,000 mg sodium (category 5) was associated with a 12% increased risk of overall mortality (HR = 1.12, 95% CIs: 1.06, 1.18) with adjusted ARDs at 16 years of follow-up of 0.84% (95% CI, 0.40% to 1.37%) (Additional file 1: Table S3). An intake of < 2,000 mg (category 1) was also associated with a 4% increased mortality (HR = 1.04, 95% CI: 1.01, 1.06). When compared with the category 2 reference level in women of 1,400- < 2,200 mg, an intake of ≥ 4,000 mg sodium (category 5) was associated with a 14% increased risk of overall mortality (HR = 1.14, 95% CI: 1.06, 1.23), with adjusted ARDs at 16 years of follow-up of 0.73% (95% CI: 0.20% to 1.24%) (Additional file 1: Table S4).

Fig. 1

a Sex Stratified Associations Between Sodium Intake and Overall and Cause-specific Mortality in Multivariable-Adjusted Cubic Spline Regression Models. Analyses were adjusted for age at baseline, BMI, alcohol consumption, smoking status (never, former, current or missing), physical activity, race or ethnic group, education, marital status, diabetes (yes vs. no), health status, vitamin supplement use, total energy intake, and the Healthy Eating Index 2015 (HEI-2015) score excluding the sodium component. For women, the risk estimates were additionally adjusted for postmenopausal hormone therapy (yes vs. no). The solid line denotes the HR of overall mortality according to sodium intake with a four-knot cubic spline selected at the 5th, 25th, 75th, and 95th percentiles of intake, dashed lines and shaded areas represent the 95% confidence intervals, blue indicates men and red indicates women. b Sex Stratified Associations Between Potassium Intake and Overall and Cause-specific Mortality in Multivariable-Adjusted Cubic Spline Regression Models. Analyses were adjusted for age at baseline, BMI, alcohol consumption, smoking status (never, former, current or missing), physical activity, race or ethnic group, education, marital status, diabetes (yes vs. no), health status, vitamin supplement use, total energy intake, and the Healthy Eating Index 2015 (HEI-2015) score components of sodium, seafood and plant protein, saturated fat, fatty acids and refined grains. For women, the risk estimates were additionally adjusted for postmenopausal hormone therapy (yes vs. no). The solid line denotes the HR of overall mortality according to potassium intake with a four-knot cubic spline selected at the 5th, 25th, 75th, and 95th percentiles of intake, dashed lines and shaded areas represent the 95% confidence intervals, blue indicates men and red indicates women. c Sex Stratified Associations Between Sodium–Potassium Ratio and Overall and Cause-specific Mortality in Multivariable-Adjusted Cubic Spline Regression Models. Analyses were adjusted for age at baseline, BMI, alcohol consumption, smoking status (never, former, current or missing), physical activity, race or ethnic group, education, marital status, diabetes (yes vs. no), health status, vitamin supplement use, total energy intake, and the Healthy Eating Index 2015 (HEI-2015) score components of seafood and plant protein, saturated fat, fatty acids and refined grains. For women, the risk estimates were additionally adjusted for postmenopausal hormone therapy (yes vs. no). The solid line denotes the HR of overall mortality according to sodium–potassium ratio with a four-knot cubic spline selected at the 5th, 25th, 75th, and 95th percentiles of intake, dashed lines and shaded areas represent the 95% confidence intervals, blue indicates men and red indicates women

Table 2 Risk of overall and cause-specific mortality associated with daily sodium intake in the NIH-AARP studya

Although with wider confidence intervals, there was also clear evidence of non-linear associations of dietary sodium intake with risk of mortality from CVD and heart disease in men (both Pnonlinearity = 0.0002), and mortality from other causes combined in both men and women (Pnonlinearity ≤ 0.001) (Fig. 1a). In men, the HRs for CVD mortality was 1.07 (95% CI: 1.00, 1.14) in the fifth versus the lowest quintile; 1.08 (95% CI: 1.01, 1.15) for heart disease mortality (Table 2). Evidence to suggest possible nonlinearity was observed between dietary sodium and mortality from CVD, heart disease mortality in women (Pnonlinearity = 0.01 and 0.03, respectively; all P value < 0.0001; Fig. 1a). For CVD and heart disease mortality in women, the HRs were 1.21 (95% CI: 1.09, 1.34) and 1.27 (95% CI: 1.13, 1.43) for sodium intake in the fifth category versus the first quintile, respectively (Table 2).

Dietary potassium and overall and cause-specific mortality

From the multivariable-adjusted models, the associations between dietary potassium and mortality outcomes are demonstrated in Fig. 1b, and minimally adjusted and parsimoniously adjusted associations are presented in the supplemental data (Additional file 1: Figures S2 and S5). The supplemental data showed the HRs and their 95% CIs of the multivariable-adjusted cubic-restricted splines (Additional file 1: Tables S3 and S4). Table 3 and the supplemental data (Additional file 1: Table S6) present the estimated HRs and their corresponding 95% CIs based on quintiles, per 500 mg or per SD change in potassium intake for the associations with the risk of overall mortality and cause-specific mortality. Restricted cubic splines showed statistically significant curvilinear inverse associations between potassium intake and overall mortality in men and women (Pnonlinearity < 0.0001). The overall mortality association for men was initially steeper and plateaued at approximately 4000 mg/d, while the association for women appeared more linear (Fig. 1b). In both men and women, participants in quintiles 2–5 experienced a significantly lower risk of overall mortality than those in the lowest quintile, and a stronger inverse association was noted in women (fifth versus first quintile, HR for men = 0.96, 95% CI: 0.92, 0.99; HR for women = 0.82, 95% CI: 0.79, 0.86; all Ptrend ≤ 0.01; P for interaction = 0.0006) (Table 3). Adjusted ARDs at 16 years of follow-up were -0.39% (95% CI: -0.65% to -0.14%) and -0.49% (95% CI: -0.75% to -0.48%) for men and women, respectively (fifth versus first category) (Additional file 1: Tables S3 and S4).

Table 3 Risk of overall and cause-specific mortality associated with daily potassium intake in the NIH-AARP studya

For mortality from heart disease, and other causes combined, similar patterns of non-linear associations of potassium intake were observed for both men and women, but with wider confidence intervals (Pnonlinearity ≤ 0.0006; Fig. 1b). Higher intakes of dietary potassium were associated with a greater significant risk reduction for mortality from CVD and heart disease for women (fifth versus first quintile, HR for CVD mortality = 0.79, 95% CI: 0.72, 0.87; HR for heart disease mortality = 0.80, 95% CI: 0.73, 0.89), but not for men (fifth versus first quintile, HR for CVD mortality = 1.02, 95% CI: 0.95, 1.09; HR for heart disease mortality = 1.03, 95% CI: 0.96, 1.10) (P for interaction = 0.0028; Table 3).

A U-shaped association of potassium intake with stroke mortality was suggested for men, with a minimum risk found at approximately 4,000 mg/d (Pnonlinearity = 0.002, Fig. 1b). There were linear inverse associations for respiratory disease mortality in men and women and infectious disease mortality in women (all P-values ≤ 0.04). Participants in quintiles 2–5 had a lower risk of respiratory disease mortality than those in the lowest quintile (fifth versus first quintile, HR for men = 0.82, 95% CI: 0.71, 0.94; HR for women = 0.64, 95% CI: 0.55, 0.75; Table 3). Higher potassium intake was associated with a lower risk of infectious disease mortality for women (fifth versus first quintile, HR = 0.77, 95% CI: 0.61, 0.97), but not for men (HR = 1.01, 95% CI: 0.84, 1.22) (Table 3).

Dietary sodium–potassium ratio and overall and cause-specific mortality

Figure 1c shows the dose–response multivariable-adjusted associations between sodium–potassium ratio and mortality outcomes, and the HRs and their corresponding 95% CIs of the cubic-restricted splines are shown in the supplemental data (Additional file 1: Tables S3 and S4); minimally adjusted and parsimoniously adjusted associations are reported in Additional file 1: Figures S3 and S6. Table 4 and the supplemental data (Additional file 1: Table S7) present the estimated HRs and their corresponding 95% CIs based on quintiles and per SD change in sodium–potassium ratio for the associations with the risk of overall mortality and cause-specific mortality. Among men, curvilinear associations for the dietary sodium–potassium ratio were observed for overall and CVD mortality (Pnonlinearity values ≤ 0.0003), and heart disease mortality (Pnonlinearity = 0.002). With the lowest quintile as the reference group, only the fifth quintile of the sodium–potassium ratio showed significantly increased risk of overall and CVD mortality for men (HR for overall mortality = 1.09, 95% CI: 1.05, 1.13; HR for CVD mortality = 1.08, 95% CI: 1.00, 1.15; both Ptrend < 0.05) (Table 5).

Table 4 Risk of overall and cause-specific mortality associated with daily intake of the sodium–potassium ratio in the NIH-AARP studyaTable 5 Risk of overall mortality associated with intakes of sodium, potassium and sodium–potassium ratio stratified across racial/ethnic groups among men and womena

We observed linear associations of dietary sodium–potassium ratio with overall mortality in women (fifth versus first quintile, HR = 1.23, 95% CI: 1.16, 1.31), and with mortality from respiratory disease in women (fifth versus first quintile, HR = 1.50, 95% CI: 1.24, 1.83) and other causes combined in men and women (Table 4). We also found that the linear associations of the sodium–potassium ratio were steeper for CVD and heart disease mortality in women than in men (fifth versus first quintile, HR for CVD mortality = 1.34, 95% CI: 1.19, 1.50; HR for heart disease mortality = 1.38, 95% CI: 1.22, 1.58) (both Ptrend < 0.0001; both P for interaction < 0.0001; Table 4).

Dietary sodium, potassium, sodium–potassium ratio and overall mortality in cohort subgroups

Based on the multivariable-adjusted models in men and women, Fig. 2a, b provides findings from stratified analyses of cohort subgroups. The overall mortality associations with dietary intake of sodium, potassium and sodium–potassium ratio were generally similar across population subgroups of age at baseline, smoking status, diabetes, alcohol intake, vegetable consumption, supplemental vitamin use, HEI-2015, self-reported health status and years of follow-up. However, the association between potassium intake and overall mortality was stronger among men with lower BMI (P for interaction < 0.0001, Fig. 2a). Additionally, the association between potassium intake and overall mortality in men was modified by fruit consumption, with the association being significant in lower fruit consumers (fifth versus first category, HR = 0.88, 95% CI: 0.83, 0.92), but not in high fruit consumers (HR = 1.03, 95% CI: 0.97, 1.08) (P for interaction = 0.00036). Regarding sex differences, the associations between potassium intake, the sodium–potassium ratio and risk of overall and CVD mortality were statistically significantly different (P for interaction ≤ 0.0006 for overall and CVD mortality, Tables 3 and 4).

Fig. 2

a Associations Between Intakes of Sodium, Potassium, Sodium–Potassium Ratio and Overall and Cause-specific Mortality in Multivariable-Adjusted Models Stratified by Selected Characteristics Among Men. Hazard ratios (HRs) and their 95% confidence intervals (CIs) for overall mortality comparing the highest versus the lowest quintile. Multivariable analyses were adjusted for age at baseline, BMI, alcohol consumption, smoking status (never, former, current or missing), physical activity, race or ethnic group, education, marital status, diabetes (yes vs. no), health status, vitamin supplement use, and total energy intake. For sodium intake, models were further adjusted for Healthy Eating Index 2015 (HEI-2015) score excluding the sodium component; for potassium intake and the sodium–potassium ratio, models were additionally adjusted for HEI-2015 components for sodium (potassium model only), seafood and plant protein, saturated fat, fatty acids and refined grains. P for interaction was examined by the likelihood ratio test, entering the cross-product term of exposure factors (categories) and the stratification variables (categorized as shown), all as ordinal variables, to the Cox proportional hazard regression model. b Associations Between Intakes of Sodium, Potassium, Sodium–Potassium Ratio and Overall and Cause-specific Mortality in Multivariable-Adjusted Models Stratified by Selected Characteristics Among Women. Hazard ratios (HRs) and their 95% confidence intervals (CIs) for overall mortality comparing the highest versus the lowest quintile. Multivariable analyses were adjusted for age at baseline, BMI, alcohol consumption, smoking status (never, former, current or missing), physical activity, race or ethnic group, education, marital status, diabetes (yes vs. no), health status, postmenopausal hormone therapy (yes vs. no), vitamin supplement use, and total energy intake. For sodium intake, models were additionally adjusted for Healthy Eating Index 2015 (HEI-2015) score excluding the sodium component; for potassium intake and the sodium–potassium ratio, models were additionally adjusted for HEI-2015 components for sodium (potassium model only), seafood and plant protein, saturated fat, fatty acids and refined grains. P for interaction was examined by the likelihood ratio test, entering the cross-product term of exposure factors (categories) and the stratification variables (categorized as shown), all as ordinal variables, to the Cox proportional hazard regression model

In the stratified analyses for both men and women, the associations between intakes of sodium, potassium and sodium–potassium ratio and risk of overall mortality were not statistically significantly different across racial/ethnic groups (Table 5). Compared with the referent (lowest quintile) category, a higher intake of daily sodium was associated with a 5% and 14% increased risk of overall mortality in men, and a 9% and 21% increased risk in women, among non-Hispanic white individuals and the combined racial/ethnic participants, respectively; although stronger associations were seen for the combined racial/ethnic group, they did not reach a statistically significant threshold (Pinteraction = 0.092 and 0.59, respectively). We observed stronger potassium-mortality associations among non-Hispanic white participants in women, with a 19% and 8% decreased risk (fifth versus first quintile) in the non-Hispanic white participants than the combined racial/ethnic group (Pinteraction = 0.017, respectively). Participants in a higher category of sodium–potassium intake ratio experienced a 9% and 15% increased risk of overall mortality in men, and a 27% and 11% increased risk in women (fifth versus first quintile), among the non-Hispanic white individuals and the combined racial/ethnic participants, respectively (Pinteraction = 0.89 and 0.25, respectively).

Dietary sodium, potassium, sodium–potassium ratio and overall mortality in sensitivity analyses

The primary findings of associations between dietary intake of sodium, potassium, sodium–potassium ratio and risk of overall and cause-specific mortality did not change materially in the sensitivity analyses which excluded the first five years of follow-up (Additional file 1: Figures S7-S9), or participants who reported a history of diabetes at study entry (Additional file 1: Figures S10-S12), or participants who reported poor to fair health status or unknown health status at study entry (Additional file 1: Figures S13-S15), or additionally adjusted for family income or dietary potassium intake in the sodium-mortality models (Additional file 1: Figures S16-S19).

Systematic review and meta-analysis

Overall, our initial search yielded 3,318 articles after removing duplicate studies. After evaluating the titles and abstracts, 93 articles were reviewed in full, and 33 studies (34 with the current study) were included in the meta-analysis (Additional file 1: Tables S8-S10). The supplemental data depicts the study selection process (Additional file 1: Figure S20), and illustrates characteristics of the individual studies included in the meta-analysis (Additional file 1: Table S10). Based on the Newcastle–Ottawa quality assessment scale, 16 studies (including the current study) reached a score of at least seven, reflecting a low risk of bias (Additional file 1: Table S11).

A total of 34 studies of sodium intake and risk of incident CVD and CVD mortality comprised a total of 42 risk estimates, 2,085,904 participants (dietary questionnaires: 1,014,282 participants; urinary measurement: 1,071,622) and 80,085 CVD events (dietary questionnaires: 44,530 participants; urinary measurement: 35,555). The pooled RR (95% CI) was 1.13 (1.06, 1.20) for the highest versus lowest category of sodium intake. We found no indication of publication bias for the association between sodium intake and risk of CVD (including CVD mortality) (P value for Egger’s test = 0.10, Additional file 1: Fig. S21), but observed considerable between-study heterogeneity (I2 = 72.9%, Fig. 3), although no single study was overly influential in this regard (Additional file 1: Fig. S22). In the meta-analysis of predefined subgroups, there were no statistically significant interactions, with the exception of dietary assessment (Pinteraction = 0.027), where the positive association between sodium intake and CVD risk was stronger among studies with repeated assessments of sodium intake in the follow-up period (pooled RR = 1.36, 95% CI: 1.09, 1.69) than among those with only one baseline assessment estimate of sodium intake (pooled RR = 1.08, 95% CI: 1.02, 1.14) (Fig. 3 and Additional file 1: Table S12). Additionally, all subgroup data were provided in greater than 10 studies/risk estimates in accordance with the Cochrane recommendations (https://training.cochrane.org/handbook/current/chapter-10), with the exception of sex-specific analyses: 9 risk estimates for men and 7 risk estimates for women.

Fig. 3

Association of Sodium Intake with Cardiovascular Disease Risk for the Highest Versus Lowest Intake Category Using Random-Effects Meta-Analysis According to Baseline Only and Repeated Assessment of Sodium Intake (P for interaction = 0.027). Squares represent study-specific relative risks, with their areas being proportional to the specific-study weight in the meta-analysis. Horizontal lines denote 95% Cis. I2 refers to the proportion of heterogeneity among studies. Abbreviations: AF = atrial fibrillation; CHD = coronary heart disease; CI = confidence interval; CVD = cardiovascular disease; HF = heart failure

In the meta-analysis of 29 cohort studies with 2,056,043 participants and 66,526 CVD events, we conducted a 3-knot restricted cubic spline regression analysis which yielded similar findings, with the nadir of CVD risk (including CVD mortality) in the daily sodium intake range of 1,700 mg to 2,300 mg, and a striking CVD risk excess for individuals with sodium intake greater than 5,000 mg/day (Pnonlinearity < 0.001, Fig. 4).

Fig. 4

Association Between Sodium Intake and Risk of Cardiovascular Disease Using a Cubic-Restricted Spline Model in the Dose–Response Meta-Analysis Based on 29 Published Studies. The solid line represents the relative risk of cardiovascular disease according to sodium intake, and the dashed lines denote the 95% confidence intervals (Pnonlinearity < 0.001). The spline was computed on the basis of three knots selected at 5th, 50th, 95th percentiles of sodium intake, and the red line depicts the sodium intake of 2,300 mg/day. The analysis included 2,056,043 participants from 29 cohort studies with 66,526 CVD events