Proper hydration is necessary for the body to sustain normal physiological functioning such as the transportation of nutrients, excretion and the regulation of body temperature. Chronic mild to moderate dehydration has been associated with several disease states, such as fatal CHD and CVD(Reference Chan, Knutsen and Blix1). The association between chronic mild dehydration and CHD may be explained by the elevation in haemorheological factors(Reference Dvilansky, Bar-Am and Nathan2, Reference Kristal-Boneh, Glusman and Chaemovitz3). Haemorheological factors that are independently positively associated with CHD are the whole-blood viscosity, plasma viscosity, fibrinogen and haematocrit levels(Reference Lowe, Lee and Rumley4–Reference Erikssen, Thaulow and Sandvik8). These factors respond acutely to changes in hydration, which can occur, e.g. during normal daily activities and the use of medications such as diuretics(Reference Kurabayashi, Kubota and Tamura9–Reference Yasaka, Yamaguchi and Oita12). Furthermore, it has been suggested that water consumption at night can attenuate nocturnal increase in blood viscosity, and therefore, can prevent morning occurrence of cerebral infarction(Reference Kurabayashi, Kubota and Tamura9). Some investigators reported a decrease in blood viscosity with increased water intake(Reference Kurabayashi, Kubota and Tamura9, Reference Okamura, Washimi and Endo13, Reference Wood14), while others reported no changes in blood haemorheology and chemistry with increased water intake(Reference Tonstad, Klemsdal and Landaas15).
Only one cohort study analysed the effect of water intake and other fluid intake on fatal CHD(Reference Chan, Knutsen and Blix1). The above-mentioned study, which was done by Chan et al., found a protective effect of water intake on mortality related to CHD in men but not in women. In contrast, a high intake of ‘fluids other than water’ was associated with an increased mortality risk in women but not in men. The authors explained these findings by the typically lower blood viscosity (lower haematocrit levels, erythrocyte rigidity and aggregability at all shear rates) in women(Reference de Simone, Devereux and Chien7). Because of the lower blood viscosity, women experience less reduction in plasma volume with dehydration, and body temperature regulation occurs at lower sweat rates(Reference Byrd, Stewart and Torranin16, Reference Rocker, Kirsch and Stoboy17). Thus, in women, reduced water intake does not produce high levels of blood viscosity as in men, which would explain why water intake is not protective in women. Because of the fact that men have larger blood volumes than women, the same quantities of hyperosmolar fluids or caffeinated beverages produce a lower proportionate decrease in intravascular volume and subsequently smaller increase in blood viscosity, explaining why ‘fluids other than water’ have smaller effects on CHD mortality in men(Reference Neuhauser, Beine and Verwied18–Reference Bhalla, Sankaralingam and Dundas20).
The effect of other beverages such as coffee and tea on IHD or stroke mortality has also been reported previously in the literature, but inconsistent results were obtained. A meta-analysis of twenty-one prospective cohort studies found a positive association between coffee and IHD in men, while coffee drinking was associated with a lower risk of IHD in women(Reference Wu, Ho and Zhou21).
Also, the effect of tea intake on IHD mortality has been reported to be either inverse, no association, or even positive(Reference Geleijnse, Launer and Hofman22–Reference Woodward and Tunstall-Pedoe27).
The lack of sufficient evidence on the relationship between total fluid and other beverage intake and mortality related to IHD or stroke led to the investigation of this association in the Netherlands Cohort Study (NLCS). Due to the detailed collection on dietary and drinking habits in the NLCS, we were able to analyse the association between total fluid (and specific beverage) intake and IHD and stroke mortality, with the possibility of adjustment for potential confounding factors.
Methods
Study population
In September 1986, the population-based prospective NLCS on diet and cancer was started. In this cohort study, 58 279 men and 62 573 women aged between 55 and 69 years living throughout the Netherlands were included(Reference van den Brandt, Goldbohm and van't Veer28). Because of efficiency reasons, a case–cohort approach was used for data processing and statistical analysis(Reference Prentice29, Reference Wacholder, Gail and Pee30). In a case–cohort design, cases are collected from the entire cohort, while detailed follow-up information of a random subcohort provides an estimate of the person-time experience of the cohort. In the NLCS, this subcohort consisted of 5000 cohort members extracted randomly from the entire study population immediately after baseline. The subcohort was followed up biennially for vital status information. Only two male subcohort members were lost to follow-up during the 10-year follow-up period. The study design has been described in detail previously(Reference van den Brandt, Goldbohm and van't Veer28). The NLCS was approved by the institutional review boards of the Maastricht University and the Netherlands Organization for Applied Scientific Research. All cohort members consented to participate in the study by completing a mailed, self-administered questionnaire.
Data collection
All the 120 852 cohort members completed a self-administered questionnaire on dietary habits, beverage consumption, anthropometry and other risk factors related to cancer and CVD at baseline(Reference van den Brandt, Goldbohm and van't Veer28). Habitual consumption of food and beverages during the year preceding the start of the study was assessed using a 150-item semi-quantitative food frequency section of the questionnaire. Validation of this questionnaire was done against a 9 d diet record(Reference Goldbohm, van den Brandt and Brants31). The Spearman correlation coefficient between mean daily intake of non-alcoholic beverages assessed by the questionnaire and that estimated from the 9 d diet record was 0·63. The Spearman correlation coefficient between mean daily intake of alcoholic beverages assessed by the questionnaire and that estimated from the 9 d diet record was 0·89(Reference Goldbohm, van den Brandt and Brants31). Specific nutrient intakes were calculated from this section of the questionnaire using the computerised Dutch food composition table(32). Nutrient intake was adjusted for energy intake using the residual method(Reference Willett and Stampfer33).
Assessment of total fluid and specific beverage intake
Of the 150 FFQ items, twenty-four addressed beverage consumption. The questionnaire covered almost all beverages consumed regularly, namely water (tap or bottled), milk (i.e. milk, buttermilk, cocoa), juice (i.e. tomato, orange, others), soda and lemonade (i.e. cola, others), alcoholic beverages (i.e. beer, red wine, white wine, sherry, port, liqueur, spirits), soup, coffee and tea. The amount of beverages consumed was measured in household units (glasses, cups, soup plates, etc.). The participants could indicate the frequency of beverage consumption and the portion size. During the validation study, the capacity of the household units was measured by a dietitian who visited the subjects(Reference Goldbohm, van den Brandt and Brants31). The standardised size of a glass measured 175 ml, a cup (coffee or tea) 125 ml and a soup plate 250 ml. Based on the questionnaire data, the total fluid consumption was calculated using information on frequency and beverage-specific serving size of all the specific beverages. When subjects reported of drinking a specific beverage but did not report the number of units, a default number was used based on the mean number reported by the other drinkers.
Ascertainment of cases
Follow-up for mortality data was established by linkage of the NLCS to the central bureau of genealogy. Using this record linkage, 18 091 deaths were identified between January 1987 and December 1996. On the basis of the collected subcohort data on vital status, it could be estimated that the Central Bureau of Genealogy linkage was complete for 99 %. The causes of death, coded in international classification of diseases (ICD)-9 until December 1995 and in ICD-10 from January 1996, were obtained by the linkage to Statistics Netherlands. Using this linkage, the cause of death was obtained for 18 045 (99·7 %) of the 18 091 deceased. The primary cause of death was related to CHD (ICD-9 code 390-459/ICD-10 code I00-I99) in 37·2 % (n 6735) of all deaths. In the present analysis, CHD-related mortality was analysed as death due to IHD (ICD-9 code 410-414/ICD-10 code I20-I25, n 3610) or CVD (stroke) (ICD-9 code 430-435/ICD-10 code I60-I69, n 1204).
Statistical analysis
Analysis was confined to participants who did not report physician-diagnosed heart disease, stroke or diabetes as indicated in the baseline questionnaire (exclusion of 3066 subjects). Additionally, subjects with incomplete or inconsistent dietary data (n 1010) were excluded from the analysis, leaving 1789 IHD mortality cases, 708 stroke cases and 3970 subcohort members available for the analysis. For categorical analysis, the lowest consumption stratum was chosen as the reference category. Because different associations between consumption of water and fluids other than water and cardiovascular mortality for males and females have been reported in the literature, we conducted all the analyses separately for men and women(Reference Chan, Knutsen and Blix1). For each sex, the hazard ratios (HR) and the corresponding 95 % CI were calculated by means of a Cox proportional hazards model. The proportional hazards assumption was tested using the scaled Schoenfeld residuals(Reference Schoenfeld34). When this test indicated that the assumption was violated, the − ln( − ln) survival plots were used to assess the proportional hazards assumption. Standard errors were estimated using the robust Huber–White sandwich estimator to account for additional variance introduced by sampling a subcohort from the total cohort(Reference Lin and Wei35).
The multivariable regression models included the following variables: age at baseline (years), sex (if applicable), cigarette smoking (current, frequency and duration) and energy intake (kcal/d). The following variables were considered as potential confounders (other known IHD risk factors or factors influencing total fluid consumption), but were not included in the models because they did not change the HR by >5 %: hypertension (yes/no), BMI (kg/m2), physical inactivity (coded in four categories: very little, average, active or very active), educational level (coded as primary school, lower vocational, high school (junior and senior) and higher vocational/university), fat consumption (g/d), fruit and vegetable consumption (g/d), hormone-replacement therapy (yes/no) (for women only), use of diuretics (yes/no) and multivitamin intake (yes/no)).
When separate beverages were analysed, a second model was used with additional correction for the intake of fluids other than the analysed beverage (e.g. analysis for coffee was corrected for the intake of beverages other than coffee). In none of the beverage-specific analyses did this additional correction with other beverage intake change the HR of the first model. Also, additional adjustment for the ethanol compound of alcoholic beverages (coded in categories 0, 0·1–4, 5–14, 15–29 g/d, ≥ 30 g/d) did not change any of the reported HR substantially. Therefore, only the results of the first model are presented in this report.
When the intake of beverages was analysed as a continuous variable, the HR was calculated per one standard deviation increment (calculated for men and women together). Tests for trends were assessed by fitting the ordinal exposure variables as continuous terms. A two-sided P value of 0·05 or less was considered as statistically significant.
Results
During the 10-year follow-up period, 1789 deaths primarily related to IHD have been reported. Of these deaths, 1252 (70·0 %) occurred in the male population, and 537 (30·0 %) occurred in the female population. Stroke mortality occurred in 708 subjects, of which, 417 (58·9 %) were males and 291 (41·1 %) were females.
The mean daily total fluid consumption from beverages was 1462 ml/d in the male subcohort, and 1341 ml/d in the female subcohort (Table 1). The major beverages consumed in the male and female subcohorts were coffee (mean 581 and 498 ml/d, respectively), tea (mean 319 and 384 ml/d, respectively) and milk (mean 187 and 181 ml/d, respectively). Only a very little amount of the daily total fluid intake was in the form of water consumption (mean 82 ml/d in men and 109 ml/d in women). Potential confounders in the present analyses are also described in Table 1. Compared with the subcohort members, subjects who suffered from fatal IHD or stroke were more often smokers, hypertensive, and less physically active.
(Mean values with their standard errors, number, percentages and ranges)
NA, not applicable.
Total fluid consumption
When analysing total fluid consumption on a continuous scale with an increment of 431 (1 sd) ml/d, no association between total fluid and water intake and IHD or stroke mortality was found in either men or women (see Tables 2 and 3).
(Hazard ratios (HR) and 95 % confidence intervals)
* Adjusted for age.
† Adjusted for age, current smoking, number of cigarettes smoked, years of active smoking and total energy intake (kcal).
‡ This P or HR has to be interpreted with care, because the categorical analysis shows a strong U-shaped association.
§ One cup = 125 ml.
(Hazard ratios (HR) and 95 % confidence intervals)
* Adjusted for age.
† Adjusted for age, current smoking, number of cigarettes smoked, years of active smoking and total energy intake (kcal).
‡ One cup = 125 ml.
§ This P or HR has to be interpreted with care, because the categorical analysis shows a strong U-shaped association.
Categorisation of total fluid consumption revealed that 41 % of the male and 30 % of the female subcohort members consumed the recommended daily dose of total fluids (at least 1·5 litres total fluid per d). It was revealed that 38 % of the male subjects and 31 % of the female subjects who suffered from fatal IHD, and 39 % males and 28 % females who died of stroke had drunk at least 1·5 litres of total fluids. Drinking at least 1·5 litres of total fluids was not related to IHD or stroke mortality compared with drinking less than the recommended daily dose in men or women. Also, analysis over five total fluid categories did not reveal an association between total fluid intake and IHD mortality in men or women. HR for the association between total fluid intake and stroke mortality were all below 1, but were not statistically significant.
Total fluid intake (consumption of at least 1·5 v. < 1·5 litres) was not associated with IHD and stroke mortality across different subgroups of the population defined by risk factors for CVD (sex, smoking, hypertension and BMI) (Table 4).
(Hazard ratios (HR) and 95 % confidence intervals)*
* Hazard rates of mortality related to IHD and stroke in groupings of different CVD risk factors among men and women who reported no physician-diagnosed heart disease, angina pectoris, stroke or diabetes, and where the exposure is consumption of the daily recommended dose (1·5 litres fluids) in the Netherlands Cohort Study, 1987–96
† Adjusted for sex, age, current smoking, number of cigarettes smoked, years of active smoking and total energy intake (kcal).
Specific beverage consumption
In men, a positive association between increasing coffee intake and IHD mortality was observed (HR per increment of 1 sd: 1·09, 95 % CI 1·00, 1·18), while an inverse relationship between increasing tea consumption (increment of 1 sd) and IHD mortality was observed (HR: 0·91, 95 % CI 0·83, 1·00) (Table 2). In women, an inverse association between coffee consumption (increment of 1 sd) and IHD mortality was observed (HR: 0·88, 95 % CI 0·78, 1·00), while a positive (not statistically significant) relationship between increasing tea consumption (increment of 1 sd) and IHD mortality was observed (HR: 1·08, 95 % CI 0·97, 1·20). No relationship was observed between milk, soft drink, juice, soup and alcoholic beverage consumption and IHD mortality. When the endpoint was stroke mortality, no associations with separate beverages were observed (Table 3), except for the intake of alcoholic beverages in females. In females, intake of alcoholic beverages (per 1 sd or 166 ml/d) was inversely associated with stroke mortality (HR: 0·66, 95 % CI 0·44, 0·97).
We analysed the observed relationship between coffee and tea intake and IHD mortality more detailedly using categorical variables for these beverages. In the categorical analysis, the positive association between coffee consumption and IHD mortality in men was not statistically significant. For tea, we observed that all categories of tea intakes (moderate (1–2 cups/d) and high tea consumption (>3 cups/d)) were inversely related to IHD mortality in men (HR: 0·75; 95 % CI 0·61, 0·93 and HR: 0·71, 95 % CI 0·57, 0·88, respectively (P-trend = 0·007)) compared with 0–1 cup/d of tea consumption (Table 2).
In women, a more detailed analysis of coffee consumption revealed an inverse dose–response relationship with IHD mortality (P-trend = 0·01). In the categorical analysis, the positive association between tea consumption and IHD mortality in women was no longer significant.
The association between coffee and tea consumption categories and IHD and stroke mortality was relatively consistent across different subgroups of the population defined by risk factors for CVD (sex, smoking, hypertension and BMI) (data not shown).
Water as such was not consumed by 53 % of the male and 43 % of the female subcohort members, 57 % of the males and 45 % of the females who died of IHD and 56 % of males and 46 % of the females who died of stroke. In men, drinking any water compared with drinking no water at all was not statistically significantly associated with a reduced risk of IHD mortality (HR: 0·87, 95 % CI 0·74, 1·03) in the multivariate analysis. In women, drinking any water compared with drinking no water at all was not associated with IHD mortality (HR: 0·92, 95 % CI 0·75, 1·11). When stroke mortality was the endpoint, no significant association with water consumption was observed in men or women. When analysing water in five categories, we observed a decreased HR of IHD mortality for male subjects who consumed between >0 and 100 ml water a day compared with those consuming no water (HR: 0·69, 95 % CI 0·55, 0·87). With the other water categories also, no relationship was observed (P-trend = 0·68). Also, when stroke mortality was the outcome measure, no relationship between any of the five water categories was observed in men. In women, no relationship between any of the five water categories and IHD or stroke mortality was observed.
Discussion
We found no overall association between total fluid intake and IHD or stroke mortality in either men or women. However, a positive association between coffee consumption and IHD mortality was observed in men, while the opposite (inverse association between coffee intake and IHD mortality) was observed in women. For tea consumption, we observed an inverse relationship with IHD mortality in the male subjects, but not in the female subjects. In women, except for the intake of alcoholic beverages, no associations were observed when stroke mortality was the outcome measure.
We observed that a higher coffee consumption was statistically significantly associated with lower IHD mortality in women, while the opposite was observed in men. These opposite results are in line with the results of a meta-analysis of twenty-one prospective cohort studies(Reference LeGrady, Dyer and Shekelle36). The reason for this sex difference, however, is unknown. Apart from a direct effect of coffee consumption on IHD mortality, it is also known that coffee consumption is strongly associated with smoking. Smoking is often associated with unhealthier lifestyle habits (e.g. physical inactivity, alcohol abuse) compared with non-smoking(Reference LeGrady, Dyer and Shekelle36). We adjusted for inter alia smoking in our multivariate analysis – and this changed the HR estimates in women substantially – but adjustment for these factors may not have completely eliminated their influence on the coffee–IHD association, and therefore, may partially explain the observed sex differences. The positive association between IHD and coffee consumption in men has also been reported in previous studies(Reference Rosengren and Wilhelmsen37–Reference Urgert and Katan40). In the meta-analysis done by Wu et al., the 1 pooled HR for males with heavy (5–6 cups/d) and very heavy ( ≥ 7 cups/d) coffee consumption (HR: 1·07, 95 % CI 0·91, 1·26 and HR: 1·15, 95 % CI 0·91, 1·46) were slightly elevated, but they did not reach statistical significance(Reference LeGrady, Dyer and Shekelle36). The positive association between coffee consumption and IHD can be explained by diterpenes cafestol and kahweol that are present in unfiltered coffee(Reference Weusten-Van der Wouw, Katan and Viani41–Reference Tverdal, Stensvold and Solvoll43). Cafestol and kahweol are potent cholesterol-elevating compounds present in coffee, which are independently associated with the risk of CVD(Reference Hammar, Andersson and Alfredsson44, Reference Coats45). However, in the Netherlands, mainly filtered coffee is consumed, in which the concentration of these diterpenes is very low and not associated with the elevated risk of IHD.
Tea consumption was inversely associated with IHD mortality in the male study population. The potential beneficial effect of tea on IHD has been reported previously(Reference Klag, Mead and LaCroix39–Reference Weusten-Van der Wouw, Katan and Viani41). However, other studies found no association or even a positive association between tea intake and IHD mortality(Reference Klatsky, Armstrong and Friedman25–Reference Woodward and Tunstall-Pedoe27) Our analysis also suggests a sex difference for the association between tea intake and IHD mortality, but not as clearly as has been suggested for coffee. As mentioned in the previous section, the observed opposite effect could partially be explained by the fact that we have not been able to remove all the confounding effects(Reference Woodward and Tunstall-Pedoe27). We observed that when coffee and tea consumptions were included together in the multivariate analyses, the HR did not change substantially. The intriguing differences between the sexes observed for both coffee and tea indicate that further investigation of the responsible compounds and the biological mechanisms involved has to be done.
The effect of water intake on fatal CHD was analysed in only one other study(Reference Chan, Knutsen and Blix1). In contrast to the study done by Chan et al., we found no association between the intake of water and fluids other than water and mortality related to IHD. The reason for us not finding an association could be the marked differences in drinking habits between the present study population and the population analysed by Chan et al. The average water consumption in the Adventist population was approximately 1000 ml/d, while it was approximately 95 ml/d in the NLCS population. The reported total fluid intake from beverages was around 2 litres/d in the Adventist population, and was 1·4 litres/d in the present study population. Another possible reason for the different results is that in these observational studies, the results have been biased by confounding. Subjects who drink more water could be more health conscious, e.g. drinking more water may be a marker of higher physical activity(Reference Chan, Knutsen and Blix1). We corrected for potential confounders such as physical activity in the multivariate analyses. Diabetics, who may drink more fluids, and are at increased risk of IHD and stroke, were excluded from these analyses. Furthermore, the null association was relatively consistent across subgroups of the population.
The strengths of the present study include the relatively large number of cases in a prospective cohort; the ability to examine the intake of total fluid and specific beverages in relation to the risk of IHD and stoke mortality due to the broad data collection on commonly consumed beverages (including water); the possibility to control for a broad spectrum of confounders such as age, cigarette smoking, hypertension and the intake of other fluids besides the one under investigation.
One disadvantage of the present study is the low consumption of fresh water. Of the male and female subcohort members, 53 and 43 %, respectively, reported that they did not consume any fresh water at all. Another disadvantage is that when subjects reported of drinking a specific beverage but did not report the number of units, a default number was used based on the mean number reported by the other drinkers. This could lead to reduction of the variance of the variable. However, this occurred in only a limited number of participants.
In conclusion, total fluid intake was not associated with IHD- or stroke-related mortality in either men or women in the NLCS. Coffee consumption was inversely associated with IHD mortality in women and it was positively associated with it in men, while tea consumption was inversely associated with IHD mortality in men, but not in women.
Acknowledgements
None of the authors has a financial or personal conflict of interest. This manuscript has been read by all the authors who accept responsibility for its contents. L. J. L. participated in all the processes of the present research topic. L. J. S. assisted with the statistical analysis. L. J. S. and P. A. v. d. B. contributed to writing of the manuscript and R. A. G. contributed to the refining of the manuscript. The present study was funded by KWR, Watercycle Research Institute. This project was conducted within the Joint Research Programme of the Dutch water companies by researchers from the Maastricht University and Kiwa Water Research. External review of the design and results was conducted by a Scientific Advisory Committee consisting of Professor P. Hunter (University of East Anglia, Norwich, United Kingdom), Professor R. Rylander (Emeritus Professor, Sweden) and M.I. Sinclair (Water Quality Research Australia Limited, Monash University, Australia). The authors are much indebted for their valuable comments and, in addition, would like to thank Margreet Mons for initiating this project and for her involvement in the design and advice. They also thank A. Volovics and A. Kester for statistical advice; S. van de Crommert, H. Brants, J. Nelissen, C. de Zwart, M. Moll, W. van Dijk and A. Pisters for assistance; and H. van Montfort, T. van Moergastel, L. van den Bosch, R. Schmeitz and J. Berben for programming assistance. Furthermore, they are indebted to all the participants of the present study and the Dutch Central Bureau of Genealogy and Statistics Netherlands for data provision.
