Mounting Complexities in the Dietary Salt & Health Relationship

Posted on by

Associate Professor Nick Wilson, Professor Tony Blakely, Dr Cristina Cleghorn

salt picA large prospective study on dietary salt and health has recently been reported in the New England Journal of Medicine. While reinforcing extensive past work that a (very) high intake of salt is hazardous to health – an increased hazard was also found for low intakes of salt (i.e., a “J-shaped” or “U-shaped” relationship). This blog post considers this new study in more detail and suggests that we need a high-level international review to clarify the research and policy agenda from here. Our interpretation should be treated as preliminary on what may be an important study; therefore, we welcome and encourage comments on this blog post.  [SEE MULTIPLE COMMENTS ON THIS BLOG BELOW.  AND IN PARTICULAR SEE SUBSEQUENT BLOG WHERE WE UPDATE ANALYSIS BELOW BASED ON DISCUSSIONS WITH COLLEAGUES – IMPORTANT.]

The overall evidence around salt and harm to health (as outlined in two recent systematic reviews (1, 2)) has been described as strong enough to justify public health action to reduce high levels of sodium intake. Indeed, as recently as August 2014 a new study estimated that there were 1.65 million deaths globally from cardiovascular causes in 2010 attributed to sodium consumption above a reference level of 2.0 g per day (3). This study included a new meta-analysis of 107 published trials of sodium reduction and reduced blood pressure (BP).

Furthermore, the risk factor of a “diet high in sodium” has been estimated to be one of the top two dietary risk factors for disease burden identified in the massive Global Burden of Disease Study (4). “Salt reduction” has also been included in the top five priority actions for non-communicable disease (NCD) control internationally (5), and for reducing NCD inequalities (6).

However, the evidence relating to sodium and health has always been controversial to some extent (7). In particular, a recent Institute of Medicine Report (8) highlighted the uncertainty around the health benefits and risks of reducing sodium intakes below the 2300 mg/day level (see the Table below).

Salt table

Most recently, a large study has been reported in the New England Journal of Medicine by O’Donnell et al (12). It reports a “U-shaped” relationship for dietary salt intake and all-cause mortality and also for cardiovascular events (see the Figure below). Of note is that the increased cardiovascular risk was estimated to only start increasing from the lowest estimated risk level (4-5g of sodium – see Table above). Other work has also reported such a J-shaped or “U-shaped” relationship (13-16)) but the last two of these cited studies included people with established cardiovascular disease (so reverse causation is possible – see below). But the O’Donnell et al study is by far the largest that has reported such a pattern. Indeed, it involved over 100,000 subjects from 17 countries. It had other desirable features: 95% follow-up of subjects, 3.7 years of follow-up on average, it considered dietary potassium intake as well, and it undertook various analyses to explore the possibility of reverse causation, sensitivity analyses around potential confounding (array-approach), and propensity-score matched sensitivity analyses.

Figure from O’Donnell et al, NEJM 2014

Figure from O’Donnell et al, NEJM 2014

Possible limitations with the new O’Donnell et al study

But no study is perfect, and the limitations with this new study include:

  1. It was a prospective study and these all suffer from the possibility of residual confounding (unmeasured and mismeasured variables). But the authors did analyses to explore this and point out the level of confounding by a single confounder would need to be large to nullify their findings – and peculiar to generate the U-shape. Nevertheless, if there are multiple confounders on many separate “backdoor paths” (epidemiological jargon for considering other causal pathways) then several modest confounding processes could conceivably sum up to create such a relationship (but, again, we have difficulty seeing how this might result in a “U-shaped” relationship). Unlike with alcohol whereby confounding may generate a J-shaped curve, it is difficult to posit that people with average salt consumption have better profiles on other risk factors than either low and high salt consumers, and that this is not measured or known.
  2. Reverse causation cannot be completely ruled out” as the authors’ state. Reverse causation would arise if early CVD disease caused a lowering of salt intake. The authors excluded people with any pre-existing CVD, and it made no substantive difference to findings. But, it would have been ideal to have even longer follow-up.
  3. They used spot urine measures (and only a single measure) rather than the gold standard of 24-hour urines to measure sodium. There are still uncertainties with the utility of the spot urine measure (see this systematic review of 20 studies (17)). Nevertheless, any such measurement error would tend to weaken the association with sodium intake and health outcomes. Again, it would seem to take a very odd correlation of measurement error processes to generate a “U-shaped” association. Furthermore, the study found the expected relationship with salt and higher BP which suggests that this is unlikely to be problematic (i.e., BP results can be seen as functioning as a sort of internal control on the validity of the study).
  4. There was some dominance of Chinese subjects (42% of the total) – though there was adjustment for Asian vs non-Asian in the analyses.

There are also some limitations with generalising the findings from this study:

  1. It is still just one study (albeit large and apparently well designed).
  2. It is still just an observational study. A large RCT with an intervention to lower sodium intake would be scientifically superior.
  3. There is no obvious mechanism for why low sodium intakes might be hazardous (given the very small amounts that are thought to be required for normal physiological functioning). Nevertheless, some other studies have also suggested this U-shaped relationship (see above) and many nutrients follow J-shaped relationships of both too little and too much being problematic for health (e.g., selenium, vitamin A, and total intake of dietary energy).

Our overall interpretation

The new O’Donnell et al study is large and appears to be fairly robust. As such it probably needs to be considered as a serious challenge to certain aspects of the current prevailing view around the level at which the intake of dietary salt impacts on health outcomes. While there is no doubt that very high salt intakes are hazardous – it is now somewhat uncertain if too little salt is also hazardous and what the threshold is before the hazard starts to appear (for both too little and too much).

This blog post should be treated as our preliminary perspective. This new study will probably generate much academic discussion and probing. We plan to update our perspective in the future (keep checking Public Health Expert blog).

What could be done now?

  1. The World Health Organization (WHO) could establish a high-level review panel and clarify the state of the evidence and the future research agenda. Such a panel may have maximum credibility if it just had scientists with no past publications relating to dietary salt and health. If this is widely seen as the most efficient option – the NZ Government could request that WHO takes on this task.
  2. Health authorities could just wait until a large cluster-randomised trial in China is completed (18) (it includes replacing some sodium chloride [NaCl] intake with potassium chloride [KCl]). But perhaps it will still not be clear if any benefit is due to the higher potassium or the lower sodium (or a mix of both).
  3. Health authorities could proceed with certain sodium reduction actions that still have a high probability of being favourable. That is the reformulation of various high salt products to replace some of the NaCl with KCl in some other processed foods (as has already been happening for some foods in NZ e.g., some soup products). The evidence that increased dietary potassium is good for cardiovascular health is now fairly strong (e.g., these two meta-analyses (19, 20), as well as the O’Donnell et al study for lower risk of death and major cardiovascular events combined (12)). A recent meta-analysis of 5 RCTs using salt substitutes has also reported a benefit for reducing systolic and diastolic blood pressures (21). Furthermore – the health advice to eat more whole foods and to avoid a high intake of processed foods (usually high in all three of: salt, sugar and saturated fat) still stands. Indeed, a junk food tax (as recently introduced by Mexico) could well be justified on numerous health grounds and not just on sodium levels.


The relationship between dietary salt and poorer health is still fairly certain for very high intakes – but given recent research findings there is less confidence about the associations for other levels of intake. Given this uncertainty there appears to be a reasonable case for a WHO-led expert review.


1. He FJ, Li J, Macgregor GA: Effect of longer-term modest salt reduction on blood pressure. Cochrane Database of Systematic Reviews 2013, 4:CD004937.

2.  Aburto NJ, Ziolkovska A, Hooper L, et al: Effect of lower sodium intake on health: systematic review and meta-analyses. BMJ 2013, 346:f1326.

3. Mozaffarian D, Fahimi S, Singh GM, et al: Global sodium consumption and death from cardiovascular causes. The New England Journal of Medicine 2014, 371(7):624-634.

4. Lim SS, Vos T, Flaxman AD, et al: A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990-2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet 2012, 380(9859):2224-2260.

5. Beaglehole R, Bonita R, Horton R, et al: Priority actions for the non-communicable disease crisis. Lancet 2011, 377(9775):1438-1447.

6. Di Cesare M, Khang YH, Asaria P, et al: Inequalities in non-communicable diseases and effective responses. Lancet 2013, 381(9866):585-597.

7. Bayer R, Johns DM, Galea S: Salt and public health: contested science and the challenge of evidence-based decision making. Health Affairs 2012, 31(12):2738-2746.

8. Institute of Medicine: Sodium Intake in Populations: Assessment of Evidence. Washington, DC: The National Academies Press, 2013.

9. WHO: Guideline: Sodium intake for adults and children. Geneva, World Health Organization (WHO), 2012.

10. NHMRC/MoH: Nutrient Reference Values for Australia and New Zealand. Canberra, ACT: National Health and Medical Research Council (NHMRC); New Zealand Ministry of Health (MoH), 2006.;

11. McLean R, Williams S, Mann J, Parnell W: How much salt are we eating? Estimates of New Zealand population sodium from the 2008/2009 Adult Nutrition Survey [Presentation on 2 December 2011]. Joint Annual Scientific Meeting of the Australian and New Zealand Nutrition Societies. Queenstown, New Zealand (29 November – 2 December); 2011.

12. O’Donnell M, Mente A, Rangarajan S, et al: Urinary sodium and potassium excretion, mortality, and cardiovascular events. The New England Journal of Medicine 2014, 371(7):612-623.

13. Graudal N, Jurgens G, Baslund B, Alderman MH: Compared with usual sodium intake, low- and excessive-sodium diets are associated with increased mortality: a meta-analysis. Am J Hypertens 2014, 27(9):1129-1137.

14. Pfister R, Michels G, Sharp SJ, Luben R, Wareham NJ, Khaw KT: Estimated urinary sodium excretion and risk of heart failure in men and women in the EPIC-Norfolk study. European Journal of Heart Failure 2014 [E-publication].

15. O’Donnell MJ, Yusuf S, Mente A, et al: Urinary sodium and potassium excretion and risk of cardiovascular events. JAMA 2011, 306(20):2229-2238.

16. Thomas MC, Moran J, Forsblom C, et al: The association between dietary sodium intake, ESRD, and all-cause mortality in patients with type 1 diabetes. Diabetes Care 2011, 34(4):861-866.

17. Ji C, Sykes L, Paul C, et al: Systematic review of studies comparing 24-hour and spot urine collections for estimating population salt intake. Pan American Journal of Public Health 2012, 32(4):307-315.

18. Li N, Yan LL, Niu W, et al: A large-scale cluster randomized trial to determine the effects of community-based dietary sodium reduction–the China Rural Health Initiative Sodium Reduction Study. American Heart Journal 2013, 166(5):815-822.

19. Aburto NJ, Hanson S, Gutierrez H, et al: Effect of increased potassium intake on cardiovascular risk factors and disease: systematic review and meta-analyses. BMJ 2013, 346:f1378.

20. D’Elia L, Iannotta C, Sabino P, Ippolito R: Potassium-rich diet and risk of stroke: updated meta-analysis. Nutrition, Metabolism, and Cardiovascular Diseases 2014, 24(6):585-587.

21. Peng Y-G, Li W, Wen X-X, et al: Effects of salt substitutes on blood pressure: a meta-analysis of randomized controlled trials1-3. The American Journal of Clinical Nutrition 2014, [E-publication 15 October]

This entry was posted in Uncategorized and tagged , , , , , , , by Kate Sloane. Bookmark the permalink.

9 thoughts on “Mounting Complexities in the Dietary Salt & Health Relationship

  1. Excellent summary of the evidence and very practical recommendations thank you. With all dietary advice context is so important – excess salt from processed foods vs. a little salt added to a whole food diet.

    • This is an excellent point, Helen. Context matters. I have always viewed the evidence against salt as evidence against processed foods (given these foods are always the predominant source of salt). This, of course, could be applied to multiple markers of health in nutrition, accounting for much of the confusion that exist in nutrition recommendations and the evidence base. Salt, saturated fat, carbohydrates, sugars, et al… all seemingly in issue when derived from highly processed foods. But a murkier picture appears when we start looking at these very same markers within a whole, real foods pattern/framework.

      Sodium as a marker is highlighted by this paper, suggesting that recommendations for sodium and potassium cannot be simultaneously met (though both likely could within a real food framework):

  2. A few additional comments you might (or might not) be interested in.

    An issue you have not addressed is the choice of estimating equation used by the PURE investigators – the Kawasaki equation. And the effect that might have had upon estimating the absolute level of consumption for each group of participants. You may have noticed that the levels of consumption reported in the PURE paper are remarkably high.

    We have recently completed (and are preparing for submission for publication) a systematic review of the association between measured (based upon 24hr urine samples) and estimated (based upon spot urine samples) daily salt consumption. We have done this by examining data for about 60 populations where the same group of individuals had estimates of daily intake made using the 24h and spot methods. While the data are imperfect, the Kawasaki equation stands out as substantially over-estimating mean population salt intake compared to the estimates based upon 24hr urine samples (by about 3g). And much more so than other equations. This would suggest the nadir indicated by the PURE data is shifted substantively to the right of where it should be (if indeed it exists). I believe you should build that into your assumptions because it will have significant implications upon where you set the limits for your uncertainty around the effects of lower levels of salt.

    The other point I would make is that the PURE assertion about the validity of the estimated measures of 24hr urine is somewhat uncertain. You only have to look at the variability around repeat measures of daily salt intake estimated from repeat 24hr collections compared to repeat estimates based upon spot urine samples to appreciate the fact that the other factors in the estimating equations can play a large role. The measure to measure variability is greatly reduced for some estimating equations compared to repeat 24hr collections suggesting, I think, that it is the less variable components of the estimating equation (perhaps age, BMI and the like) that are driving the salt intake estimate obtained for each individual. There must be substantial uncertainty about what a salt intake estimated from an equation based upon a spot sample actually really means.

    I also note on the PURE spot vs. 24hr plots that the intercept from the Kawasaki equation crosses at 2g sodium suggesting, I think, a 5g over-estimation of intake compared to 24hr urine samples. And further supporting a shift of the proposed nadir in risk substantially to the left.

    1. There are plenty of things that go into the estimating equations based upon spot samples that predict BP and death other than salt. Some (like BMI even have their own inverse associations with CVD) The metric that comes out of the estimating equation is an estimate of something (not just 24hr salt intake) driven by everything that goes into the equation. It’s theoretically not really any different to what comes out of a Framingham risk equation. Why we should believe that it is primarily the spot urine driving the observed association with CVD in PURE is unclear to me when you have other such incredibly powerful predictors of risk (and BP) as age, sex and renal function also in there. As such I remain entirely unconvinced by the PURE CVD data.
    2. Equations based upon spot urines seem fine for estimating mean population salt intake, but unsuited to the purpose of causal inference that PURE are proposing here.
    3. The PURE data almost certainly massively over-estimate the salt intake of individuals and the nadir (if indeed there is one) should not be anywhere near as far to the right as you are suggesting.
    4. You should perhaps consider weighting the cohort study findings in your deliberations differently because of the significantly greater risk of bias inherent in the study design. Nutritional epidemiology, and observational epidemiology in general, is ridden with J and U-shaped associations that turn out to be the consequence of confounding. Why you would assert so strongly that this time its real is very unclear.



    • Dear Bruce,
      Thanks for your comment – it is very useful. I had assumed that any measurement error of 24 hour urine from spot urine would be of the classic measurement error variety, or ‘random noise’ that would only serve to weaken the predicted association of sodium intake with CVD. Your comment has prompted me to look more closely at that actual Kawasaki formula that uses the ratio of sodium to creatinine (UNa/UCr) in the spot urine to predict 24 hour sodium. Here it is:
      24-h urinary Na excretion (mEq/day) = 16.3x {(UNa/UCr)x 24h-UCr }^ 0.5
      Where “24-UCr” is the 24 hour urinary creatinine which, as you say, has to be estimated with more formulas:
      male: estimated 24h-UCr(mg/day) = 15.1x BW + 7.4 x height – 12.4 x age – 80
      female: estimated 24h-UCr(mg/day) = 8.6 x BW + 5.1 x height – 4.7 x age – 75
      Where BW = body weight (in Kg).
      So body weight, height and age are ‘feeding in’ to the prediction. I can now see your concern, in that the predicted 24 hour urine is actually a composite measure that folds in these other predictors of CVD which – if not carefully adjusted for in subsequent analyses – may cause an odd (i.e. spurious) associations of spot urine predicted sodium with CVD. I am still struggling to see how it would be J-shaped, but it is possible.
      Put another way, spot urines may be good enough to predict population averages, and even individual sodium intake with some error, but when using these predicted values to unravel the association of sodium with CVD it may all come unstuck.
      Bruce – what quality papers/research would you suggest looking at on the potential errors in using spot urine predicted sodium intake as a predictor of CVD? I am assuming there is some ….
      Best, and thanks for your contribution.
      Tony Blakely

  3. We are not persuaded that a WHO high-level review in response to the PURE study publication is needed. In fact a comprehensive review of the literature was commissioned by the Nutrition Guidance Expert Advisory Group (NUGAG) of WHO and published in 2013 in the BMJ (1, 2). This led to a strong recommendation from WHO to lower intakes of dietary sodium to less than 2 grams per day (equivalent to 5 grams of salt) for adults(3). This is consistent with the totality of the evidence, which shows adverse effects on blood pressure and cardiovascular disease outcomes from well-conducted studies, including randomised controlled trials(4). The usual mechanism proposed for potential adverse cardiovascular disease outcomes associated with a low sodium diet is the activation of renin-angiotensin-aldosterone system, and elevation of LDL cholesterol. The recent WHO review found no evidence to support these mechanisms.(1)
    The limitations outlined by Bruce Neal and acknowledged by Tony Blakely above are widely acknowledged. We further add that estimates based on a single spot urine bear little resemblance to the subjects own 24 hour urine which, although considered the gold standard of sodium intake, also only approximates an individuals usual intake due to considerable day to day variability of sodium intake. The Kawasaki formula used in the PURE study has been shown to be particularly inaccurate, and in particular over-estimates intake at lower estimated intake levels.(5) Estimates of intake based on spot urine samples are only suitable for population level estimates, and do not provide sufficient accuracy to apply this method in epidemiological studies such as the PURE study.

    Other limitations of the PURE study are potential for confounding and reverse causation. The authors of this study made no attempt to control for other dietary factors apart from a similarly inappropriate estimate of usual dietary potassium intake. The DASH trial illustrates the importance of total diet quality in blood pressure control, rather than examining single nutrients in isolation(6).

    Given the above, and an examination of the totality of the evidence around the adverse effects of a high dietary sodium intake, the results of the PURE study hardly pose a “serious challenge” to current dietary guidelines about sodium intake. Apparent disagreement among experts can undermine important public health messages. Salt restriction is an important component of the WHO approach to reduction of non –communicable diseases(7). It is important that this be supported by public health opinion leaders. The PURE study should in no way undermine our confidence in this cornerstone of Public Health Nutrition.
    Rachael McLean & Jim Mann

    1. Aburto NJ, Ziolkovska A, Hooper L, Elliott P, Cappuccio FP, Meerpohl JJ. Effect of lower sodium intake on health: systematic review and meta-analyses. BMJ. 2013 doi: 10.1136/bmj.f1326
    2. He FJ, Li J, MacGregor GA. Effect of longer term modest salt reduction on blood pressure: Cochrane systematic review and meta-analysis of randomised trials. BMJ. 2013 doi: 10.1136/bmj.f1325
    3. World Health Organization. Guideline: Sodium intake for adults and children. Geneva: World Health Organization (WHO), 2012.
    4. He FJ, MacGregor GA. Salt reduction lowers cardiovascular risk: meta-analysis of outcome trials. The Lancet. 2011;378(9789):380.
    5. McLean R, Williams S, Mann J. Monitoring population sodium intake using spot urine samples: validation in a New Zealand population. Journal of Human Hypertension. 2014. doi:10.1038/jhh.2014.10
    6. Sacks FM, Svetkey LP, Vollmer WM, Appel LJ, Bray GA, Harsha D, et al. Effects on Blood Pressure of Reduced Dietary Sodium and the Dietary Approaches to Stop Hypertension (DASH) Diet. The New England Journal of Medicine. 2001;344(1):3-10.
    7. United Nations General Assembly. Political Declaration of the High-level meeting of the General Assembly on the Prevention and Control of Non-communicable Diseases. New York: United Nations, 2012.

    • Dear Rachael and Jim,

      Thanks for the expert comments – this is exactly the discussion we are wanting to see on this issue.

      My comment to Bruce Neal still stands though. Yes, spot urines (using formulas such as Kawasaki) misclassify 24 hour sodium intake of individuals. But can anyone point me to a methodological piece of work that has shown how that misclassification (within plausible bounds) could lead to a J-shape or U-shape association? I think this is the critical issue. And is necessary to completely rebut the O’Donnell study published in the NEJM. (And I suspect there will be several groups internationally examining just this issue at the moment if it has not already been done.)

      Best, and thanks again for your comments.

      Tony Blakely

      • Hi Tony
        I am not aware of anything published. The other problem with using spot urines (apart from potential problems inherent in the conversion formula) is that a spot urine only reflects sodium intake over the past few hours, and bears little resemblance to a person’s ‘usual intake’ as day-to-day (and within day) variability of intake is so high.

  4. So, the advice I need to give as a joining GP is not much changed – goldilocks stuff really. Not too much, not too little, just right…
    Though given we all pretty much have too much already cutting back probably the best thing!

    • Agreed – given we mostly have too much salt, cutting back is likely to be beneficial.
      I also note in passing that we will be posting another blog on this topic next week, summarizing the useful discussions we have had with many colleagues in the last week on this issue of salt and J- or U-shape associations. Personally, my position is moving to be less concerned about measurement error in the spot urine method (but still somewhat concerned), and more concerened about residual reverse causation – see blog early next week.
      Tony Blakely

Leave a Reply

Your email address will not be published. Required fields are marked *

* *