Lacey R F (Water Research Centre, Medmenham Laboratory, Henley Road, Medmenham, Marlow SL7 2HD, UK) and Shaper A G. Changes in water hardness and cardiovascular death rates. International Journal of Epidemiology 1984, 13: 18-24.
Ten years ago Crawford, Gardner and Morris reported that changes in cardiovascular death rates between 1951 and 1961 had been associated with changes in the hardness of water supplies. That study has been critically reviewed but after amendments to method and data its overall conclusion is upheld.
New results based on changes that have taken place in water hardness and in cardiovascular death rates between 1961 and 1971 in the county boroughs of England and Wales indicate a significant trend for men, in the direction of decreasing cardiovascular mortality with increasing hardness, but no trend for women. The trend in male mortality appears to be specific to cardiovascular disease. The results are similar to those of the earlier study and support the hypothesis of a weak causal relationship between the hardness of drinking water and mortality from cardiovascular disease.
Of the many investigations into the link between cardiovascular disease and the hardness of drinking water, perhaps one of the most convincing has been the ‘changes’ study by Crawford, Gardner and Morris which examined the changes in death rates that had taken place between 1951 and 1961 in the county boroughs of England and Wales.1 Cardiovascular death rates, in general, showed a favourable change in the towns where water had become harder and an unfavourable change in the towns where water had become softer. No such effect was apparent for non-cardiovascular rates. Because these findings gave evidence suggesting a causal relationship, they have been widely regarded as strong support for the ‘water story’ in relation to cardiovascular disease. Subsequent comments2,3 on the study however suggested that the results might not be statistically significant but did not provide the detailed reasoning for such a conclusion. The purpose of this paper is to review the earlier study and to present new results based on data from more recent changes in water hardness in Great Britain. Further technical details and comments on other studies have been given in a longer report.4
In keeping with the original publication1 the term ‘cardiovascular’ is taken to cover all disorders of the heart and circulation. More precisely it is defined by the International Classification of Diseases codes 330-334, 410-468 (6th/7th Revisions) and 393-458 (8th Revision).
The study by Crawford and her colleagues was based on a comparison between the changes in death rates three groups of towns:
The criterion for assigning a county borough to Group A or B was that there had been a change of at least 50 mg CaCO3 equivalent per litre (mg 1-1 in the total hardness of its water supply in the 30 or so years up to 1960. This definition allowed some changes that were almost complete before the first mortality period 1948-54, while others had only just qualified by the time of the second mortality period, 1958-64. The grouping was therefore very unspecific with regard to any hypothesized time lag between cause and effect Although it would have been desirable to tighten the definition of ‘change’ in hardness, this was probably not possible with the data which were then available while retaining numbers of towns in Groups A and B large enough to permit a worthwhile statistical analysis.
The crucial results of the earlier study are summarized in Table 1. The differences in mean change in death rates between groups are generally consistent with an inverse relationship between water hardness and cardiovascular disease. The authors claimed that the overall statistical significance of these differences was high (p < 0.02), based on the combination of results from the four population classes. The apparently anomalous contrast between Groups A and C for women aged 65-74 was noted but not pursued.
The method of statistical analysis, described by Gardner, was as follows. For each town the change in cardiovascular death rate between 1951 and 1961 for a particular population class was measured by the log ratio,
For each of the four population classes the effect of changing water hardness was assessed by contrasting the mean values of y for Groups A, B and C. These group means will be noted by respectively and are the basis of the results shown in Table 1. The systematic effect of a positive change in hardness can be judged statistically in terms of the group differences,
For each of the four population classes the significances of these two differences were tested using the variance of y estimated from within the three town groups. All of this is valid. In order to combine the eight results a further assumption was, however, made that the changes in death rates in the four population classes were independent.5 This assumption does not now appear to be quite correct, as shown by the correlation coefficients in Table 2. The joint distribution of changes in death rates for the two most highly correlated classes is illustrated in Figure 1.
Although the correlations between age-sex classes were not very strong, they are not negligible. Making allowance for all of these positive correlations in testing the sum of the eight contrasts would have the effect of increasing the critical probability to about 0.06, a more equivocal result than the 0.02 that was reported.
From Figure 1 it is apparent that in Canterbury, one of the towns which had experienced an increase in water hardness, there was an exceptional combination of death rate changes, with the largest death rate decrease for men but the largest increase for women. This behaviour seemed suspect, particularly as it could help to account for the anomaly earlier noted for the older women in town Group A (Table 1). The Canterbury data were therefore checked for clerical or arithmetical errors, but none were found. An investigation of the population structure in Canterbury in 1951 and 1961, however, indicated that there had been an unusually large influx of women born before 1900, and this suggested that institutional populations might be involved. Information kindly supplied by the Kent Area Health Authority pointed to St Martin’s, a psychiatric hospital which had changed from having about 30 male beds and no female beds in 1951 to 200 female beds and no male beds in 1961. The records for this hospital during 1958-64 were still in existence, and during that period there had been 64 cardiovascular deaths of women aged 65-74, which was one-third of the number for the whole of Canterbury. Of the 64 patients, only four had previously been resident in Canterbury itself. These findings justify an amendment to the mortality data for Canterbury, shown in Figure 1. The numerical outcome is to alter the estimate for women aged 65-74 in Group A in Table 1 from -5.4 to -12.2.
Retaining the original grouping of towns (A, B and C), the numbers of towns in Groups A and B are roughly balanced. If the effect of water hardness on cardiovascular disease is believed to be reversible then a powerful straightforward test of the water hardness effect in this design of study is to examine the contrast . The value of this contrast, multiplied by 100, is approximately interpretable as the average difference in the per cent change in cardiovascular mortality, between Groups A and B. The results are presented in this way in Table 3, which includes the amendment of data for Canterbury. The method used here for combining results from different age-sex classes is simply to average the respective y’s, but the positive correlations between classes have been taken into account in formulating the standard errors of the combined results.
The hardness-related effect is significant for men but not for women, although it is of the same sign. While it appears that there might be a difference between the size of effect for men and women, the difference is not significant and so an overall average can be estimated if desired. This overall average is significantly different from zero, at a significance level almost identical to that originally claimed. The amendment of the data for Canterbury thus improves the consistency of the results to an extent which just outweighs the weakening effect of allowing for the inter-class correlations. The conclusion of Crawford, Gardner and Morris can therefore be upheld, provided that one accepts the criteria for water hardness changes on which it is based.
Since 1960 there have been changes in the water supplies to a number of the county boroughs. An investigation of the corresponding changes in death rates has been based on population numbers from the Censuses of 1961 and 1971 and numbers of deaths for the periods 1958-64 and 1969-73 respectively. This investigation followed broadly the design of the earlier study but with several modifications. The evolution of the method from that used in the earlier research has been described elsewhere but its principal features are as follows.
(i) Same cohorts
One possible disadvantage of analysing intercensal changes in mortality rates for a fixed age group, say 45 to 64 years, is that the actual membership of the population class whose health is being measured also changes. A way round this is to base the analysis of changes in death rates not on fixed age groups, but on cohorts whose membership would remain constant, apart from geographical migration or death. The most convenient cohorts for use in this study were those of men and women who were between 45 and 64 years old around 1961 and who became between 55 and 74 years old around 1971. Although cardiovascular death rates increase with age, the question can be put whether they have increased faster or slower in towns which have changed their water hardness than in ones which have not. In order to analyse this, y was redefined, for each county borough and for men and women separately, to be the logarithm of the proportional increase in death rate of the relevant cohort between 1961 and 1971 or, equivalently,
(ii) Timing of changes in water hardness
In order to be more precise about timing, the changes in water hardness were assessed over the same time interval as the changes in mortality rates. The following analyses are therefore testing for a response which happens instantaneously or within a year or two of the change in water, and not one which lags by more than a decade.
(iii) Sizes of changes in water hardness
For each of the county boroughs information was collected from a number of sources to try to ascertain what the hardness had been in 1961 and 1971 (using five-year means where possible). In the earlier study, because of lack of precision of timing, it was sensible to describe the sizes of changes in a categorical way by grouping the towns. For the more recent data it is however possible to represent the sizes of water hardness changes on a continuous scale and to take this into account in comparing the values of y for towns which have changed their hardnesses by different amounts.
The method of quantifying the changes in water hardness depends on the exact form of hypothesis to be tested. In this work we were particularly interested in the extent to which the response to changes in water hardness might confirm the results of the latest large-scale geographical analysis.6,7 That research suggested that the effect of water hardness might be non-linear with greater potency in the range up to 170 mg 1-1. For each town the ‘effective size of change’ was therefore defined as the extent of change within the hardness range 0 to 170 mg 1-l.
Between 1961 and 1971 substantial changes in this range took place in 14 county boroughs whose data are shown in Table 4. This table includes all those county boroughs for which the effective size of change exceeded 20 mg 1-1 Most are well above that threshold. Of the further 62 county boroughs available for this study some nine had changed hardness by between 10 and 20 mg 1-1, the remaining 53 by less than 10 mg 1-1. These 62 were assigned the same plotting position as of zero change. Canterbury and those county boroughs which were newly constituted or re-constituted with different boundaries between 1961 and 1971 have been omitted.
(iv) Statistical method
The method of statistical analysis was linear regression of y, the logarithm of death rate increase, on x, the effective size of change of hardness. For ease of interpretation the regression slope b is tabulated as 104b which is approximately equal to 100 (exp(100b)-1), the per cent change in death rate associated with an increase of hardness of 100 mg 1-1 in the hardness range below 170 mg 1-1.
(v) Social factors
The possibility of socioeconomic conditions influencing the relative changes in death rates in different towns was checked by introducing two of the factors that had appeared significant in the British Regional Heart Study.6,7 These were ‘% manual workers’ and ‘% unemployed’. The values of these were available for each county borough for 1961 and for 1971 so that either their absolute levels or their changes could be used as covariates.
The changes in cardiovascular death rates for the male and female cohorts are shown in Figures 2 and 3 and the results for cardiovascular and non-cardiovascular deaths are summarized in Table 5. For changes in cardiovascular mortality of the male cohorts, the trend in the direction predicted by the ‘water story’ is statistically significant. For the female cohorts there is no detectable effect, nor is there for non-cardiovascular mortality for men or women. The contrast between the -8.1% for men and the 1.2% for women is statistically significant (p=0.019). Although the results in Table 5 suggest that the effect of water hardness is specific to cardiovascular disease, the outcome of a formal test contrasting the -8.1% with the -2.8 is unconvincing (p=0.21).
The above results were modified only slightly by the introduction of the social factors. None of the correlations between y and ‘% manual’ or ‘% unemployed’ were significant at the 5% level. The largest size of correlation coefficient, that between male cardiovascular y and ‘% unemployed’ in 1961, was -0.22. which was in reverse of the expected direction. There was however a slight tendency for the towns whose hardness increased to contain fewer than average manual workers (r=-0.18) and a slight tendency for ‘% manual’ in 1961 to be associated with high increases in cohort mortality (r=+0.18). When adjustment was made for this, the regression coefficient for cardiovascular deaths in men (Table 5) became -7.5 and its significance was corresponding somewhat lower (p=0.020). These results however do more to confirm than detract from the previous findings.
Although this paper has underlined the reservations about the criteria for changes in hardness used in the study by Crawford, Gardner and Morris,1 the results of re-analysing their data and the results of the more recent work are remarkably similar. Both sets of results show a significant effect for men, in the direction consistent with an inverse relationship between water hardness and cardiovascular mortality but no effect for women. This difference between the sexes does not appear to be fortuitous, particularly in the 1961-71 analysis.
It is of interest to compare the size of effect emerging from this work with that suggested from the results of Phase I of the Regional Heart Study.6,7 In that study a difference of one standard deviation in water hardness (109 mg 1-1), occurring in the hardness range up to 170 mg 1-1, was associated with a difference of 7 in SMR for all cardiovascular disease. In the present study an increase in water hardness of 100 mg 1-1 in that range was estimated to decrease male cohort mortality by about 8%. These estimates of size of effect are well within the bounds of statistical agreement. In the Regional Heart Study men and women showed similar relationships between water hardness and cardiovascular mortality and to that extent the results of the Regional Heart Study and the present analysis do not agree.
The importance to be attached to the results of this study depends on the extent to which a natural experiment can be regarded as if it were a controlled experiment. One aspect outside of our control was the allocation of hardness changes to towns, and in the new data six of the change towns occur in geographically adjacent pairs. The interpretation rests on the assumption that factors other than water quality are not seriously confounded with the changes shown in Table 4. This assumption has been checked for the two most obvious socioeconomic variables and found to hold, but there may be other factors operating and the safeguard against this cannot be complete.
All conclusions based on this study need to recognize that the design and the method of analysis are relatively crude. On the whole the findings add some support to the idea that the association between cardiovascular disease and the hardness of drinking water is real and causal. The size of the effect appears however to be rather small and neither from this nor from other epidemiological work has it been possible to identify the specific water constituent that is responsible.
We are grateful to Mr W A Stovell of the Kent Area Health Authority for investigating the hospital data for Canterbury.
The work at the Water Research Centre (RFL) was carried out under contracts to the Department of the Environment and the Commission of the European Communities (CEC Contract 246-77-1 ENV UK), whose permissions to publish have been given. The Regional Heart Study (AGS) is supported by a programme grant from the Medical Research Council.
1 Crawford M D, Gardner M J and Morris J N. Changes in water hardness and local death-rates. The Lancet 1971; 2: 327-9.
2 Clayton D G. Water hardness and cardiovascular mortality in England and Wales. In: Amavis R, Hunter W J and Smeets J G M P (eds) Hardness of Drinking Water and Public Health: 323-340, Pergamon, Oxford, 1975.
3* Shaper A G. Water hardness and cardiovascular mortality. In Drinking Water Quality and Public Health: 19-38, Water Research Centre, Medmenham, 1976.
4* Lacey R F. Changes in Water Hardness and Cardiovascular Death rates. Technical Report TR 171. Water Research Centre, Medmenham, 1981.
5 Gardner M J. Using the environment to explain and predict mortality. Journal of the Royal Statistical Society, Series A 1973; 136(3):421-40.
6 Pocock S J, Shaper A G, Cook D G, Packham R F, Lacey R F, Powell P and Russell P F. British Regional Heart Study: Geographic variations in cardiovascular mortality, and the role of water quality. British Medical Journal 1980; 2(280): 1243-9.
7 Pocock S J, Cook D G and Shaper A G. Analysing geographic variation in cardiovascular mortality: methods and results. Journal of the Royal Statistical Society, Series A, 1983; 145(3): 313-41.
*References 3 and 4 are obtainable from the Water Research Center
(Revised version received February 1983)
This page was first uploaded to The Magnesium Web Site on September 20, 2002