salt and blood pressure: the triangular hypothesis
TRANSCRIPT
SALT ANJ BLOOD PRESSURE: THE TRIANGULAR HYPOTHESIS
D.A. Booth and A.L. Thompson, Department of Psychology, University of Birmingham,Birmingham, B15 2TT, U.K.
R. Shepherd and D.G. Land, AFRC Institute for Food Research, Norwich Laboratory
and R.P. Griffiths, RHM Research Ltd., High Wycombe.
ABSTRACT
If anunselected sample includes individuals whose blood pressure is sensitive to their salt intake and individuals whose blood pressure is not sensitive, then the superposition of these two sub-populations in a scatterplot of individuals' blood pressures against their salt intakes could give a triangular distribution. The non-correlation in the insensitives would obscure the correlation expected in the sensitives. This hypothesis justifies truncation of such data to test for correlation between blood pressure and salt intake among only the individuals in the higher range of blood pressures observed. No criterion of salt sensitivity is needed. The analysis should succeed if salt intake makes a major contribution to hypertension and would be improved if other putative causes were factored out.
INTRODUCTION
Correlations between salt intake (SI) and blood pressure (BP) are evident across populations but are seldom observed within populations (1). We suggest that
one reason for this may well be that the analyses have not been appropriate to current working hypotheses conerning 'the effects of salt on the development of hypertension, for these imply a triangular distribution of SI/BP data in
an unselected sample.
THE HYPOTHESIS
It is generally accepted that only some in a population are likely to be susceptible to chronically hypertensive effects of high salt intake (2). Various
325
rates of increase in BP with age among salt-susceptible individuals are thought to yield the well-established population rise of average BP with age (3), which in turn yields the higher average BP in populations with higher average SI, when general populations with similar age profiles are compared.
It follows from such hypotheses that BP is most unlikely to correlate substantially with SI across a non- clinical sample - especially if young. The lack of correlation hypothesised in the insusceptible will mask any correlation in the susceptible. Nonetheless, high BPS attributable to and correlated with chronically high SIs,
when superimposed on a lower range of BPS unrelated to SI (high or low) could yield a triangular scatterplot of BPS against SIs. That is, the insusceptibles would largely provide a base to the triangle, consisting of the unsloping (non-correlated) distribution of their blood pressures in the normal range across all salt intakes. The susceptibles would provide a right-skewed apex to the triangle, consisting of the partial correlation of the highest pressures with higher intakes.
Of course, if a criterion of susceptibility (such as family history) is known for members of the sample, we can try to correlate BP with SI in putative susceptibles, separated out from putative insusceptibles. However, it follows from the triangular implication of currently accepted theory that we can test the hypothesis of a correlation between BP and SI with individual data, even when we lack any information on which to identify people who may be susceptible to hypertensive action of high salt
intake. Such data can legitimately be truncated, omitting people in the lower range of BPS. If salt is an important
influences on blood pressure, a BP/S1 correlation should be more readily detected amongst the higher blood
pressures than over the whole sample.
Such an outcome is unlikely 'to be a statistical artefact, for elimination of data at an extreme (in this case the lowest BPS) is bound generally to lower the value of the correlation coefficient where there is a
homogeneous bivariate relationship. If, however, there is a relationship that has been masked at one extreme, then truncating the data at that extreme should increase the correlation value. Indeed, if any other information
relevant to BP is available, then a relationship between SI and BP may be tested with greater sensitivity, either by multivariate analysis of the truncated sample or by simple correlation among the people with the higher blood pressures after further elimination of data to allow for the other influence(s) on BP. An improved correlation on still fewer data would be more evidence that an underlying relationship exists.
326
Pilot tests of the triangular analysis ---
This triangular analysis has proved productive in the
two small studies to which we have been able to apply it, one on 30 healthy young men (4) and the other on 36
healthy women. Higher BPS correlated with SI, even though no-one in either sample had especially high blood
pressure.
We averaged the last two of three seated BP measurements by random-zero sphygmomanometer, taking disappearance and appearance of Korotkoff sounds as DBP
and SBP respectively, on one occasion in the men and on
two separate occasions in the women. Salt intake was
estimated by validated questionnaire (5) in the men and by urinary Na over 7 days in the women. Two food-salt
preferences, salting habits, opinions on salt, body
weight, height and smoking were also assessed.
In the 30 men, only the amount of salt preferred in tomato soup was related to total SI in discriminant analysis (similarly to other recent observations), and
only total salt intake predicted BP (SBP only) in multiple
regression. For a SI range of 9.1 - 14.7 g NaCl/day and SBPS and DBPs up to 124 and 92 mm Hg respectively, the simple correlations between SI and BP were, for SBP, r - 0.46 (p < 0.05), DBP r = 0.02, and average BP r - 0.08.
The significant correlation appeared not to be an accidental feature but an attribute of the sample's homogeneity - for, when12 smokers were excluded on the grounds that smoking may decrease BP (6), the SI/SBP r rose to 0.69 (p < 0.05). Then, truncating the base of the
theoretical SI/BP triangle, the 8 non-smokers at or above
SBP of 115 mm Hg gave a still higher correlation coefficient of 0.84 (p < 0.01).
Truncation in the female sample gave results that weigh strongly against those effects in the male sample being fortuitous to that set of data. In a BP range of 95/54 - 169/94 and an intake range of 2.5 NaCl/day,
- 11.4 g an overall SI/SBP correlation of 0.29 rose to
0.54 (p < 0.05) in the 18 women having the higher range of SBPs. Also, the overall SI/DBP r = 0.22 rose to 0.47 (p < 0.06) in the higher half of the DBP range.
CONCLUSION
We therefore suggest that this method of data analysis be applied to healthy and clinical samples, both homogeneous and inhomogeneous. A variety of experience with analysis for a triangular distribution in the scatterplot between salt intake and blood pressure would help ascertain to what extent the puzzling discrepancy between within- and between-population correlations arises
327
\IH B
from failure to direct the statistical testing to the relevant hypothesis.
REFERENCES
Simpson FO. Salt and hypertension: a sceptical review of the evidence. Clin. Sci. 57: 463S-480s.
Dahl LK. Salt and hypertension. Am. J. Clin. Nutr.
25: 231-244, 1979.
Tobian L. The relationship of salt to hypertension. Am. J. Clin. Nutr. 32: 2739-2748, 1979.
Booth DA, Thompson AL, Shahedian B. A robust, brief measure of an individual's most preferred level of
salt in an ordinary foodstuff. Appetite 4: 301-312,
1983:
Shepherd R, Farleigh CA, Land DG. Estimation of salt intake by questionnaire. Appetite 6: 219-233, 1985.
US DHEW. The health consequences of smoking: a
report of the Surgeon-General. US Government
Printing Office, 1972.
328