Effects of Sodium Restriction on Blood Pressure, Renin, Aldosterone, Catecholamines, Cholesterols, and Triglyceride: Title and subTitle BreakA Meta-analysis FREE

RECENT large population studies have contributed to the 100-year conflict as to the possible influence of sodium on blood pressure (BP). One study showed a moderate, but significant correlation between sodium intake and BP,¹ whereas another found no correlation.² In 1986, Grobbee and Hofman³ combined 13 randomized intervention studies of hypertensive and normotensive persons in a meta-analysis (12 references) and found a significant hypotensive effect of reduced sodium intake on systolic blood pressure (SBP) of 3.6 mm Hg and a nonsignificant effect on diastolic blood pressure (DBP) of 2.0 mm Hg. In 1991, a second meta-analysis of 28 randomized study populations (24 references, including 11 from Grobbee and Hofman³ ) showed an effect of 4.0/2.5 mm Hg for hypertensive persons and 1.0/0.2 mm Hg for normotensive persons.⁴ Also in 1991, a third meta-analysis⁵ of 30 randomized studies (30 references, including 10 from Grobbee and Hofman³ and 20 from Cutler et al⁴ ) and 48 unrandomized studies (42 references) concluded that a mean reduction in sodium intake as measured by urinary sodium excretion of 100 mmol/24 h reduced BP by 6.6-10/3.6-5 mm Hg, depending on age. A recent meta-analysis of randomized studies⁶ (53 references, including 10 references from Grobbee and Hofman³ , 16 from Cutler et al,⁴ and 21 from Law et al⁵ ) showed an effect of 3.7/0.9 mm Hg in hypertensive persons and 1.0/0.1 mm Hg in normotensive persons.

The recommendation to reduce sodium intake is based on the effect on a surrogate parameter (BP) and on hypothetical consequences concerning the number of strokes and myocardial infarctions.⁵ In recent years, other surrogate variables, such as the renin-angiotensin-aldosterone system, catecholamine levels, and serum lipid levels, have been shown to be influenced by sodium intake. Before advising the public to reduce sodium intake, studies on morbidity and survival should be conducted. In the absence of such studies, at the very least the potentially harmful effects of reduced sodium intake should be estimated in addition to the effect of reduced sodium intake on BP. The purpose of the present study was to quantify the apparent stability of the effect size of reduced sodium intake on blood pressure^{3 - 4 ,6} in a cumulative meta-analysis of randomized studies. (Cumulative meta-analysis means that the meta-analysis is updated every time a new trial appears.) We also estimated the influence of reduced sodium intake on the renin-angiotensin-aldosterone system, catecholamine levels, and serum lipid levels in meta-analyses.

The effect size was defined as the difference between the changes during a low-sodium diet and those during a high-sodium diet. It was calculated for SBP, DBP, and body weight. For hormones and lipids, the ratio, not the difference, between the values obtained during the low-sodium and high-sodium diets was calculated, because different units were used in different studies.

Studies included met all the following criteria: (1) random allocation was to either a low-sodium or a high-sodium diet; (2) the sodium intake was estimated by 24-hour urinary sodium excretion, either measured on the basis of a 24-hour urine collection or estimated from a sample of at least 8 hours; (3) no confounding occurred; ie, studies treating persons with a concomitant intervention, such as an antihypertensive medication, potassium supplementation, or weight reduction, were included only if the concomitant intervention was identical during the low-sodium and high-sodium diets; (4) reporting effects were present on SBP, DBP, and mean blood pressure (MBP); if only MBP was reported, the MBP effect was accepted as both an SBP and a DBP effect; and (5) mean age was older than 15 years.

The first randomized study of the effect of reduced sodium intake on BP was published by Parijs et al in 1973.⁷ An English-language literature search in MEDLINE (1966 through December 1997) was performed using the following combinations of search terms: (1) salt or sodium; (2) restriction or dietary ; (3) blood pressure or hypertension ; (4) randomised or randomized or random. When we combined 1, 2, 3, and 4, we found 291 references. Of these, 76 randomized trials from 60 references fulfilled the inclusion criteria. From the reference lists of these articles and from 4 previous meta-analyses^{3 - 6} an additional 23 references including 38 trials were identified. Thus, 83 references, including 114 randomized study populations, were included.^{7 - 45 46 - 89} When results were reported by subgroup, the subgroup results were used, in contrast to the results reported by Midgley et al,⁶ who combined subgroup results to 1 result for each trial. Fifty-eight hypertensive study samples^{7 - 58} and 56 normotensive study samples^{10 ,17 ,31 ,36 ,38 - 39 ,55 ,59 - 89} were integrated in 2 separate meta-analyses.

Two of us (N.A.G., P.G.) independently recorded the following data from each trial: (1) the sample size (N); (2) the mean age of participants; (3) the fraction of females and males; (4) the duration of the intervention; (5) the sodium reduction measured as the difference between 24-hour urinary sodium excretion during low-sodium and high-sodium diets, and the SD; (6) SBP (SD) and DBP (SD) before and after intervention; (7) difference between changes in SBP and DBP obtained during low-sodium and high-sodium diets. This difference was extracted either between groups (parallel studies) or between treatment phases (crossover studies). The difference between SBP changes is the effect on SBP and the difference between DBP changes is the effect on DBP; (8) the SE of effect on SBP and SE of effect on DBP; (9) P values and t values for differences; and (10) levels of hormones and lipids in the blood and their SEs during low-sodium and high-sodium diets. Furthermore, body weight, the number of persons who had urinary sodium excretions analyzed per trial, the number of urinary sodium excretions analyzed per person per treatment period, and data on the completeness of urine collections were recorded. Finally, we noted whether the main purpose of the study was to estimate effects on blood pressure or to estimate other effects, and whether the method of reduced sodium intake was only dietetic or if it was based on intake of sodium capsules vs placebo.

The obligatory trial quality criterion was randomization. Double-blind, single-blind, or open studies with a parallel or a crossover design were accepted. A study was defined as single-blind if BP was measured by an investigator without knowledge of the diet or by a computerized manometer and as open if precautions to decrease observer bias were not mentioned.

Unweighted and weighted effect size methods were used to estimate the effect size, and a method of summarizing t values was used to combine probabilities.⁹⁰

The effect sizes were calculated as simple means of the extracted data.

The analyses were weighted by the inverse sampling variance as follows:

For parallel studies, N was calculated as follows:

where n₁ is the number of patients receiving a low-sodium diet, and n₂ is the number of patients receiving a high-sodium diet. All studies were ranked according to the year of publication. The SBP effect and DBP effect were multiplied by ω for each study. A running sum was calculated for weighted effect SBP and DBP and divided by a running sum of ω. In this way a weighted average effect size was directly estimated for SBP and DBP for each new study added. For each study, the SD of effect was calculated as SE (effect) √N for both SBP and DBP. The mean SD of effect was calculated for SBP and DBP by means of the following formula:

A running sum of mean SD and 1/√N was calculated and multiplied, the product representing the cumulative SE (effect). From the cumulative SE (effect), the 95% confidence interval (CI) could be calculated by multiplication with 1.96. The same procedure was used to calculate a cumulative 24-hour urinary sodium excretion. In 13 trials there were no SDs for the urinary sodium excretion. In these cases the SD was calculated as sodium excretion times the mean SD of all studies and then divided by the mean sodium excretion of all studies.

For each randomized sample, a t value for the effect on SBP and DBP was searched. No t values could be obtained directly from the 114 investigations. Instead they were (1) extracted from statistical tables⁹¹ on the basis of a P value and df (7 t values); (2) calculated as the difference in BP change between low-sodium and high-sodium diets divided by the SE (effect), with the SE (effect) given (34 t values); or (3) calculated as effect divided by SE (effect), with SE (effect) calculated from the following formula (73 t values):

where SE(H) is the SD of BP during a high-sodium diet divided by √Ν, and SE(L) is the SD of BP during a low-sodium diet divided by √N.

The method used to combine probabilities was the method of adding t values according to the following formula⁹⁰ :

All studies were ranked according to the year of publication. A running sum of Σt was divided by a running sum of

resulting in a column of Z values, each of which represents the combined statistic of the previous Z value plus the effect of the last study added (cumulative meta-analysis).

The test for homogeneity assessed whether the distribution of effect sizes was compatible with the assumption that interstudy differences were attributable to random sampling alone. The test was calculated from the following formula:

To search for sources that might explain differences between the studies and contribute to the effect size, univariate and multiple least squares regression analyses between the effect size estimate for each study and a number of variables were performed.

Eleven meta-analyses were performed. To avoid coincidental significance due to multiple comparisons, the level of significance was reduced using the following formula:

Zero-effect studies have a smaller chance to be published than statistically significant studies, and this may increase the risk of a type I error.⁹⁰ To estimate the significance of this phenomenon, the theoretical number of unpublished studies with zero effect needed to be included to change a significant result into a nonsignificant result (decrease Z to 1.645 in case of a significance level of .05, or to 2.605 in case of the present significance level [.005]) was calculated according to the following formulas:

where K is the number of studies in the meta-analysis.⁹⁰

Koolen and Brummelen,¹⁸ Jula and Karanko,⁵⁰ and Overlack et al⁵³ supplied information on hormone^{92 - 95} and lipid levels.⁹⁶ Measurements of renin,^{8 ,10 ,13 ,15 ,18 - 19 ,23 ,26 ,30 - 32 ,34 ,36 ,39 - 42 ,44 - 45 ,48 ,50 - 51 ,53 - 54 ,56 ,59 ,62 ,66 ,68 - 69 ,71 ,74 ,80 - 81 ,85 ,87 - 88} aldosterone,^{10 ,13 ,18 - 19 ,32 ,34 ,39 ,42 ,44 ,48 ,50 - 51 ,53 - 54 ,56 ,59 ,62 ,66 ,69 ,74 - 75 ,80 - 81 ,85 ,87 - 88} noradrenaline,^{18 - 19 ,26 ,31 ,33 - 34 ,40 ,44 ,53 ,55 - 56 ,59 ,62 ,71 ,74 ,81 ,83 ,88} adrenaline,^{19 ,26 ,44 ,55 - 56 ,59 ,62} cholesterol,^{26 ,40 ,43 ,45 ,55 - 57 ,73 - 74 ,81 ,83 ,86 - 87} high-density lipoprotein (HDL),^{43 ,45 ,56 - 57 ,73 - 74 ,81 ,83 ,86 - 87} low-density lipoprotein (LDL),^{40 ,57 ,73 - 74 ,81 ,83 ,86 - 87} and triglyceride^{43 ,45 ,57 ,73 - 74 ,81 ,83 ,86 - 87} were combined in meta-analyses by means of the method of adding t values⁹⁰ and by regression analyses of effect size (the ratio between the values of the low-sodium and high-sodium diets) vs difference in sodium excretion between the low-sodium and high-sodium diets.

Details of the characteristics of the individual studies and differences between the previous and the present meta-analyses are available on request. The information on the number of males (n=2883), females (n=1411), and blacks (n = 354) was incomplete. The SBP was reported in 91 trials, DBP in 93, effect SBP and DBP in 5, and MBP in 16. Because of the inclusion of the MBP effect as both SBP effect and DBP effect, SBP effects were available in 112 trials and DBP effects in 114 trials. In the 58 studies of hypertensive persons, the median mean age was 49 years (range, 23-73 years), the median study duration was 28 days (range, 4-365 days), and the number of persons included was 2161. Antihypertensive treatment was received by subjects in 13 of these studies. In the 56 studies of normotensive persons, the median mean age was 27 years (range, 15-67 years), the median duration was 8 days (range, 4-1100 days), and the number of persons included was 2581.

In 40 studies, the 24-hour urinary sodium excretion was determined based on one 24-hour collection per person per period, in 26 studies it was determined based on 2 collections, and in 48 studies it was determined based on 3 or more collections (up to 7). There was evidence that all persons had 24-hour urine collections in 68 trials, but an incomplete number of persons in 5 trials. In 41 trials no direct details were given. In 29 of these, 2 or more 24-hour urine samples were collected per person per period. In 35 studies, the mean difference between sodium excreted in two 24-hour urine measurements could be abstracted. During the high-sodium diet this mean difference was 10.7 mmol/24 h (6.1%), and during the low-sodium diet it was 7.8 mmol/24 h (12.7%). In only 12 articles was the completeness of the 24-hour urine collection checked.

Table 1 shows the results of the unweighted meta-analysis. Because there were differences in the effect size estimates of 16 studies between meta-analyses, we have also shown the effects calculated on the basis of all maximal estimates and all minimal estimates of the present and previous meta-analyses. If the studies only giving MBP were excluded, the unweighted effect was 4.8/2.3 mm Hg for hypertensive trials and 1.8/0.3 mm Hg for normotensive trials.

Figure 1 and Figure 2 show the mean effect (95% CI) of the individual trials and the effects of the cumulative meta-analyses. In the trials of hypertensive persons (Figure 1), the final weighted effect of a weighted sodium reduction of 118 mmol/24 h on SBP was 3.9 mm Hg (95% CI, 3.0-4.8 mm Hg) and on DBP was 1.9 mm Hg (95% CI, 1.3-2.5 mm Hg) (Z_SBP=7.7, P<.001; Z_DBP=6.0, P<.001). In the trials of normotensive persons (Figure 2), the final weighted effect of a weighted sodium reduction of 160 mmol/24 h on SBP was 1.2 mm Hg (95% CI, 0.6-1.8 mm Hg) and on DBP 0.26 mm Hg (95% CI, −0.3-0.9 mm Hg) (Z_SBP=4.5, P<.001; Z_DBP=1.2, P=.12). The cumulative meta-analysis shows that the effects have been stable since 1985. The number of unpublished studies with zero effect (of the same sampling size as the mean sampling size of the studies in the meta-analysis) needed to change the significant result into a nonsignificant result in the hypertensive trials was 447 studies for SBP and 247 studies for DBP, and in the normotensive trials the number needed for SBP was 111 studies.

Figure 3 shows an operating characteristic curve of the type II error for the DBP effect in normotensive persons. The risk of not detecting a significant effect of 0.3 mm Hg is large (<0.71), whereas the risk of not detecting an effect of 1.0 mm Hg is small (<0.003).

Design (crossover vs parallel, open vs single-blind vs double-blind), purpose of study (BP estimation vs other), or method of reduced sodium intake (diet vs sodium capsules) did not influence the effect size of SBP or DBP.

The test for homogeneity showed, for hypertensive populations, a χ² for SBP of 164 (critical value, 82.3) and for DBP of 155 (critical value, 84.7). For normotensive populations, χ² for SBP was 155 (critical value, 82.3) and for DBP 197 (critical value, 82.3). Consequently, in none of the meta-analyses were the interstudy differences attributable to random sampling alone; ie, the distribution of studies was heterogenous. A part of the heterogeneity could be ascribed to the fact that some of the trials stratified their samples into salt-sensitive, salt-resistant, and counterregulatory samples.^{53 ,74 ,81 ,89} If the studies with the largest contributions to the χ² value were excluded successively until χ² was less than the critical value, the weighted effect size for hypertensive persons was 3.5/1.9 mm Hg, and for normotensive persons it was 1.2/0.5 mm Hg. A detailed analysis showed that, in heterogenous studies, there was a trend toward a longer duration in normotensive persons (P=.02) and a higher degree of sodium reduction in hypertensive persons (P=.09). All short-term normotensive trials had a large reduction in sodium intake (93-343 mmol/24 h), and all long-term normotensive trials had a low reduction (23-90 mmol/24 h). Therefore, a comparison of studies of long treatment duration (>4 weeks) and large reduction in sodium intake (>100 mmol/24 h) with those of short duration (<4 weeks) and low reduced sodium intake (<100 mmol/24 h) was impossible. Only 8 hypertensive trials had both long duration (28-84 days) and large reduction in sodium intake (100-161 mmol/24 h) (BP effect, 4.2/1.5 mm Hg), whereas 3 had short duration (14-21 days) and low reduced sodium intake (50-97 mmol/24 h) (BP effect, 5.3/4.5 mm Hg). No obvious source of heterogeneity could be detected.

Univariate and multiple least squares regression analyses with effect SBP and DBP as dependent variables and duration of treatment, sample size, reduced sodium intake, body weight, age, and initial BP showed only significant correlations to age and initial BP in the multiple regression analysis if these 2 variables were included one by one, because of covariation.

The effect on body weight could be abstracted in 56 studies. Few of the individual studies showed a significant effect on weight. The mean weight reduction during low sodium intake was 0.961 kg (range, −0.4-3.0 kg) (Z=2.25, P=.01). The association between sodium intake and weight change was confirmed by a significant correlation (r=0.49; P<.001).

Renin was measured in 53 studies, and aldosterone in 38. All individual investigations showed a significant increase in renin and aldosterone in the low-sodium group. Assuming that reduced sodium intake was linearly correlated to renin and aldosterone, a runs test showed that the renin data (P=.47) and aldosterone data (P=.13) did not deviate from a straight line. Figure 4 shows the regression lines forced through 0, 0. In 22 samples in which the sodium excretion was reduced to a level less than 20 mmol/24 h, plasma renin increased by a factor of 5.7 (P<.001) and plasma aldosterone by a factor of 5.0 (P<.001). In 20 populations that had reduced sodium excretion to between 40 and 100 mmol/24 h, plasma renin increased by a factor of 1.8 (P<.001), and aldosterone by a factor of 2.3 (P<.001). In studies with a treatment duration of 4 weeks or more and a mean reduction in sodium excretion of 90 mmol/24 h, there was a significant 1.5 times increase in renin (P<.001)^{13 ,15 ,19 ,26 ,34 ,41 - 42 ,44 ,48 ,50 ,56 ,87} and a 1.7 times increase in aldosterone (P<.05).^{13 ,19 ,34 ,42 ,44 ,48 ,50 ,56 ,87} The studies of hypertensive persons and of normotensive persons did not differ in their renin and aldosterone responses to reduced sodium intake, whereas the noradrenaline response was much stronger in hypertensive persons (r=0.76; P=.002; n=13) than in normotensive persons (r=0.12; P=.64; n=16). There were too few studies of adrenaline and lipids to investigate the separate effects of reduced sodium intake in hypertensive and normotensive populations. The effects observed in the lipid studies were generally observed during excessive sodium reduction. Table 2 summarizes the meta-analyses of the hormones and lipids. There was a significant increase in renin, aldosterone, noradrenaline, cholesterol, and LDL in the low-sodium group, whereas the increase in adrenaline and triglyceride was not significant and HDL was unchanged. The risk of not detecting a significant increase of 10% or more is less than 0.014 for HDL, less than 0.20 for triglyceride, and less than 0.81 for adrenalin.

The cumulative meta-analysis showed stabilization of the treatment effect in the period after 1985 (Figure 1 and Figure 2). Corresponding to this, the results of 1985, 1990, and 1994 meta-analyses are similar to the meta-analyses published in 1986,³ 1991,⁴ and 1996.⁶ Furthermore, the 1994 results are similar to an update by Cutler et al⁴ covering the period through August 1994.⁹⁷ Weighted with the inverse variance, this update showed an effect of 3.8/2.1 mm Hg in hypertensive persons and 1.5/0.8 mm Hg in normotensive persons, whereas the effect weighted with sample size was about 25% larger. The number of imaginary zero-effect trials to be added to the present meta-analyses to change the results from significant to nonsignificant was more than 100. Consequently, it is unlikely that publication bias could be the cause of the significant results, although publication bias may influence the effect size, as found by Midgley et al,⁶ but not by Cutler et al.^{4 ,97} The discrepancy between the meta-analyses of randomized studies^{3 - 4 ,6 ,97} and the one also including unrandomized studies⁵ was caused by a larger effect in the unrandomized studies (10.3 mm Hg for SBP). The different estimates of effect size in 16 of the studies between the meta-analyses had only minor influence on the final average effect size (Table 1). The significance of heterogeneity was minor because exclusion of heterogenous studies had only a small influence on the effect sizes. Like Midgley et al,⁶ we found no source of heterogeneity. Compared with previous meta-analyses, there were differences in the inclusion criteria. For example, Midgley et al⁶ included studies of children and excluded studies with adjuvant antihypertensive therapy and studies reporting only MBP.

The present results depend in part on the accuracy and the reliability of the measurements of 24-hour urinary sodium excretion. Missing urine collections may tend to overestimate the degree of sodium reduction, assuming that those who are not delivering urine are those who do not reduce their sodium intake.⁹⁷ This bias would tend to underestimate the effect on BP, if a dose-response dependency exists. There was evidence that all persons had urine samples in 68 trials. In 29 trials with no direct evidence, it was mentioned that 2 or more 24-hour urine samples were collected per person per period. Consequently it is likely that in at least 97 studies (68+29), all persons (or almost all persons) delivered a 24-hour urine sample per period. Therefore, missing urine data did not seem to be a problem. Insufficient urine collections (occurring when individuals in the study forgot to collect a voiding) would tend to underestimate the sodium intake reduction. It was not possible to quantify the effect of this bias because of insufficient data. The strong correlations between sodium excretion and aldosterone level, renin level, and body weight indicate that the reliability of the 24-hour urinary sodium excretions was acceptable. If the sodium excretions had been determined with major errors, it is unlikely that they would have been sensitive enough to detect these correlations. Furthermore, in 35 studies in which the results of 2 measurements within the same period were reported, the difference between the mean estimates was less than 13%. This is in accordance with the findings of Midgley et al,⁶ who reported a negligible effect of correcting for measurement error.⁶

There was a marked lack of correlation between effect size and other variables, including amount and duration of reduced sodium intake, both in univariate and multiple regression analyses. However, it is difficult to conclude that such relations do not exist, because there were a considerable number of short-term studies of high reduction of sodium intake and long-term studies of low reduction of sodium intake, but very few short-term studies of low reduction of sodium intake and long-term studies of high reduction of sodium intake.

This meta-analysis seems robust concerning renin, aldosterone, and noradrenaline levels (high Z and X values, Table 2). Oliver et al⁹⁸ demonstrated that the Yanomamo Indians, who ingest extremely small amounts of sodium, had a 3 times higher level of renin in the blood and a 10 times higher excretion of aldosterone in the urine than did normal controls. In the present meta-analysis the increase in aldosterone and renin was 5 to 6 times higher in those whose sodium excretion was reduced to less than 20 mmol/24 h, ie, almost as low as the Yanomamo Indians (range, 0.3-6.8 mmol/24 h). In 20 populations with a reduced sodium excretion between 40 and 100 mmol/24 h, renin and aldosterone increased about 2 times, indicating that the renin-angiotensin-aldosterone system is also activated when sodium intake is reduced to a moderate level. Combined with our findings of a significant increase in renin and aldosterone in long-term (>4 weeks) studies with a low reduction (<100 mmol/24 h) in sodium intake, this suggests that the acute increase in renin and aldosterone may become chronic, if the reduced sodium intake is maintained. Thus, the present meta-analysis provides a possible explanation for the relatively small effect of reduced sodium intake on blood pressure: compensatory activation of the renin-aldosterone system is proportional to the degree of sodium reduction. Furthermore, an increase in noradrenaline may contribute to this counterbalance.⁹⁹ At any rate, our findings show that long-term studies of the effect of reduced sodium intake on the renin-angiotensin-aldosterone system and noradrenaline would be interesting.

There are at present 2 major positions concerning the level of the effect of reduced sodium intake on BP. The first is based on a number of smaller population studies performed with different methods without the possibility of correcting for confounders and on a mixture of unrandomized and randomized studies. These are summarized in the meta-analyses by Law et al⁵ and are supported in a recent reanalysis of the Intersalt study of 10079 persons.¹⁰⁰ The second position is expressed in (1) the first analysis of the Intersalt study¹ ; (2) the Scottish Heart Health Study, including 7354 persons, which showed an even smaller effect than the Intersalt study² ; and (3) meta-analyses of randomized studies. The first position indicates an effect of 6-10/3-5 mm Hg after an average daily reduction of sodium intake of 100 mmol/24 h; the second position assumes an effect of 1-4/0-2 mm Hg with similar reduction of sodium intake. Among those who accept that the effect of reduced sodium intake on BP is relatively small, there is disagreement regarding the relevance of the effect size. As pointed out by Stamler,¹⁰¹ even a small reduction in BP may be relevant if it could be applied to the whole population, since a small average reduction in BP could decrease the number of strokes and cardiovascular events substantially. Cutler et al⁹⁷ share that point of view, whereas Midgley et al⁶ do not and emphasize the potential adverse effects of reduced sodium intake. This disagreement exists in spite of similar effect size estimates in the 2 meta-analyses.^{6 ,97} In their regression analysis of reduced sodium intake vs BP effect, Cutler et al⁹⁷ assumed that there was no confounding, and consequently they forced their regression line through 0, 0. This resulted in a significant dose-response relationship between reduced sodium intake and BP effect (SBP in hypertensive persons declined 5.8 mm Hg per 100 mmol/24 h of reduced sodium intake), corresponding to the estimated effect size (SBP in hypertensive persons declined 4.8 mm Hg per 76 mmol/24 h of reduced sodium intake). Midgley et al⁶ did not force their regression line through 0, 0 and therefore found a dose-response relationship (SBP in hypertensive persons declined 3.7 mm Hg per 100 mmol/24 h of reduced sodium intake), which was considerably smaller than the estimated effect size (SBP in hypertensive persons declined 5.9 mm Hg per 95 mmol/24 h of reduced sodium intake). Instead, they concluded that a part of the estimated effect size might be attributed to an unidentified confounder. In conclusion, in spite of similar overall results, different statistical treatment of the available data can lead to different conclusions. Chance may influence the outcome of a meta-analysis and give rise to wrong conclusions, especially when performing subgroup analyses.¹⁰² Perhaps it is better just to focus on the average effect size. Our results indicate that the effect on the normotensive population is small in spite of a considerable reduction in sodium intake (1.2/0.26 mm Hg decline in BP per 160 mmol/24 h of sodium reduction). Therefore, reduction in sodium intake would have to be extreme to obtain measurable effects, and the present meta-analysis shows that this could lead to adverse lipid profile effects. However, it is too early to draw final conclusions because of lack of long-term (>4 weeks) studies with a moderate reduction in sodium intake (about 100 mmol/24 h). Blood lipid levels were only investigated in 2 to 5 long-term studies with a mean sodium reduction of 75 mmol/24 h.^{26 ,43 ,56 - 57 ,87} The evidence from these was not statistically significant. The effect on the lipid profile may be secondary to a shift in fluid balance such as hemoconcentration; in the present study this premise is supported by a significant body weight reduction of about 1 kg in the sodium-reduced group, probably reflecting a decrease in total body water. However, if this was the explanation, a similar percent change should have been expected for all lipids, and this was not the case. There was a significant increase in LDL (which is harmful), but no increase in HDL (which would have been beneficial). Furthermore, the type II error showed that the risk of not having detected a significant increase in HDL was small. Thus, hemoconcentration is probably not the only explanation for the increase in LDL. Furthermore, hemoconcentration may conceal a significant and harmful decrease in HDL. Whether a change in the lipid profile secondary to hemoconcentration or hemoconcentration itself is a risk factor is unclarified.

In our opinion, a BP effect of less than 1 mm Hg in normotensive persons does not support a general recommendation to reduce sodium intake. The effect in hypertensive persons (3.9/1.9 mm Hg) is smaller than the effect of antihypertensive drugs. Recent studies have shown that the risk of myocardial infarction was higher in hypertensive persons with a high renin profile than in persons with a normal or low renin¹⁰³ and in males with low sodium excretion than in males with high sodium excretion.¹⁰⁴ In a study including a sample characterized by a full range of BPs, the relative risk of acute myocardial infarction per 1-SD increase in the level of plasma renin activity was 1.04 (95% CI, 0.84-1.30) in all persons, 1.26 (95% CI, 0.63-2.56) in hypertensive males, and 1.88 (0.78-4.54) comparing the top 12% in the plasma renin distribution with the bottom 32%.¹⁰⁵ It would be hasty to conclude that reduced sodium intake is deleterious on the basis of these studies,^{106 - 107} but they do weaken the attempts to quantify the hypothetical harms of sodium intake.⁵

In summary, the present meta-analysis shows that the effect of reduced sodium intake on BP evaluated on the basis of randomized studies has been consistent since 1985. A mean daily sodium reduction of 160 mmol/24 h for 7 days decreases BP by 1.2/0.3 mm Hg in normotensive persons. This effect size does not justify a general recommendation for reduced sodium intake. A mean daily sodium reduction of 118 mmol/24 h for 28 days decreases BP by 3.9/1.9 mm Hg in hypertensive persons. This effect indicates that reduced sodium intake may be used as a supplementary treatment in hypertension. Whether a long-lasting (>4 weeks), moderate to high daily reduction (>100 mmol/ 24 h) in sodium intake could have bigger effects than those found is not yet clear, because of an insufficient number of such studies. Reduced sodium intake increased renin and aldosterone in studies up to 12 weeks, and noradrenaline in short-term studies. High, short-term reduced sodium intake had small adverse effects on the lipid profile. Whether the short-term effects on hormones and the lipid profile are of clinical significance, and whether the changes persist during long-term reduced sodium intake remain to be proven in longer-term studies. However, the optimum solution to the controversy of the influence of sodium on BP are long-term trials with hard end points, such as stroke, acute myocardial infarction, and survival.