Milk drinking, ischaemic heart disease and ischaemic stroke I. Evidence from the Caerphilly cohort

Article metrics


Objective: To test the hypothesis that milk drinking increases the risk of ischaemic heart disease (IHD) and ischaemic stroke in a prospective study.

Design: In the Caerphilly Cohort Study dietary data, including milk consumption, were collected by a semiquantitative food frequency questionnaire in 1979–1983. The cohort has been followed for 20–24 y and incident IHD and stroke events identified.

Subjects: A representative population sample in South Wales, of 2512 men, aged 45–59 y at recruitment.

Main outcome measures: In total, 493 men had an IHD event and 185 an ischaemic stroke during follow-up.

Results: After adjustment, the hazard ratio in men with a milk consumption of one pint (0.57 l) or more per day, relative to men who stated that they consumed no milk, is 0.71 (0.40–1.26) for IHD and 0.66 (0.24–1.81) for ischaemic stroke. At baseline, 606 men had had clinical or ECG evidence of vascular disease, and in these the vascular risk was even lower (0.37; 0.15–0.90). The hazard ratio for IHD and ischaemic stroke combined is 0.64 (0.39–1.06) in all men and 0.37 (0.15–0.90) in those who had had a prior vascular event.

Conclusion: The data provide no convincing evidence that milk consumption is associated with an increase in vascular disease risk. Evidence from an overview of all published cohort studies on this topic should be informative.

Sponsorship: The Medical Research Council, the University of Wales College of Medicine and Bristol University. Current support is from the Food Standards Agency.


Dietary advice which is given both to individual subjects, and to communities in the hope of reducing the risk of vascular disease, usually includes the reduction of dietary saturated fats. The sources of these fats include milk, and so it is usually advised that milk consumption is limited and a low-fat milk is used rather than whole milk (Committee on Medical Aspects of Food Policy, 1994; Naidoo & Willis, 1994).

Advice to limit milk consumption is supported by the results of ecological studies, which show positive relationships between estimates of milk consumption by a community and vascular disease rates in that community (Segall, 1977). Further support is derived from relationships between milk drinking and risk factors for vascular disease (Elwood, 2001), and from the effects of short-term intervention studies of milk consumption on blood lipid levels (Roberts et al, 1982; Steinmetz et al, 1994).

In ecological studies, however, it is not possible to take adequate account of possible confounding by lifestyle and other factors in the individual subjects. Evidence from studies of the relationship between milk and possible vascular disease mechanisms is somewhat confused and the results of short-term feeding studies are inconsistent (Barr et al, 2000; St Onge et al, 2000). It would be unwise, therefore, to judge the overall effect of milk drinking on vascular disease from its effects in any such studies. Conclusions must be drawn from relationships with disease incidence within a representative population cohort of subjects.

A number of community-based prospective studies have been published and these are listed in an accompanying paper (Elwood et al, 2004). In this paper, we report prospective evidence on milk drinking, ischaemic heart disease and ischaemic stroke in the Caerphilly Cohort Study of older men.


The Caerphilly cohort is a representative population sample of 2512 men born between 1920 and 1935 (The Caerphilly and Speedwell Collaborative Group, 1984). Of the subjects in the original defined cohort, 88% agreed to co-operate. The men were seen first in 1979–1983 and they have subsequently been seen at 5-y intervals, that is they have been followed for 20–24 y. Details of the methods and reports on prediction of vascular disease by platelet, haemostatic, rheological, lipid and psychosocial factors have been reported elsewhere (Gallacher et al, 1999; Elwood et al, 2001a, 2001b; Baker et al, 2002; Yarnell et al, 2001).

Briefly, a detailed semiquantitative food frequency questionnaire (Fehily et al, 1988) was given to the men at the time of their first examination and they were asked to complete this with the help of their partner. This included questions about milk consumption in terms of: none; less than half a pint per day; between half and one pint per day and more than one pint per day. The questionnaire from each subject was carefully reviewed with a dietitian and further questions asked as necessary to ensure that the answers were complete and unambiguous. A 30% sample of the men were also asked to keep a 7-day weighed dietary intake record and data from these were used both to derive average portion sizes for use in the analysis of the questionnaires and to test the validity of the questionnaires (Fehily et al, 1987; Fehily & Hopkinson, 1993).

Data were also collected at baseline in special clinics on a wide range of items including social class (based upon the most recent occupation, grouped into nonmanual and manual), current cigarette consumption (grouped at five levels: never smoked, ex-smoker, 0–4, 5–14, 15–24 and 25+ cigarettes per day) and usual alcohol consumption (calculated as average cc/day). Height (m), measured on a staidiometer and weight (kg) on a beam balance were used to derive body mass index (BMI: weight over height2). Blood pressure was measured at room temperature after a prolonged rest, using a random zero sphygmomanometer. Evidence on prevalent vascular disease, ischaemic heart disease and/or stroke, including a 12-lead ECG, was recorded. Samples of fasting blood were taken for estimation of platelet (Elwood et al, 2001a), haemostatic (Baker et al, 2002), rheological (Elwood et al, 2001b), lipids (Yarnell et al, 2001), psychosocial (Gallacher et al, 1999) and other factors of possible relevance to vascular disease.

At approximately 5-y intervals thereafter, the men were questioned and ECGs repeated in order to identify possible incident vascular events as completely as possible. Clinical details of each possible event were collected from the men, from their general practitioner and hospital records. Standard diagnostic criteria were applied (Bainton et al, 1988; Greenwood et al, 2001) to identify new ischaemic heart disease events (deaths attributed to ischaemic heart disease plus nonfatal myocardial infarction: 121-5 ICD 10th Revision) and ischaemic strokes (163-4 ICD). Incident events up to and including the year 2000 have been included in this report.


Cox proportional hazards model was used to estimate ratios of ischaemic heart disease events (myocardial infarction or ischaemic heart disease death) and ischaemic strokes (fatal or nonfatal) between subgroups of subjects defined by reported milk consumption at baseline. Hazard ratios, which estimate risk, taking account of time, were derived with adjustment for the possible effects of confounding by lifestyle and other factors. The choice of possible confounding factors for which adjustment was made was based upon our understanding of the mechanisms likely to be involved in the relationships examined. They are listed below in the tables that follow.


During 20–24 y follow-up was ascertained for 2403 (96%) of the men ever enlisted into the cohort. There were 439 incident IHD events (235 fatal and 258 nonfatal), and 185 ischaemic strokes (46 fatal and 139 nonfatal). In all, 811 men died during the follow-up.

The reliability of the questions about milk consumption was examined in detail. The questionnaire asked only about the consumption of liquid milk, and not about milk used in food preparation. Nevertheless, 7-day records kept by 665 of the men recorded all milk consumed from any identifiable source and the correlation (‘r’) between estimates from these and those from the questionnaires was 0.61 (P<0.001). It was not possible to take account of milk that may have been present in special breads and other manufactured food items.

Table 1 shows that differences in factors of relevance to vascular disease between subgroups of men, defined by the amount of milk consumed, are relatively small. Compared with the men who drank no milk, those who stated that they drank milk were very slightly younger, rather more of them were cigarette smokers and their consumption of alcohol was slightly lower. A higher calorie intake by the milk drinkers suggests that they were more active, and their slightly higher intakes of fat and protein suggest that they were perhaps more health conscious than the men who drank no milk. There were very small differences in blood pressure, but the fasting lipid levels, and the mean homocysteine levels, were very similar across the groups.

Table 1 Baseline characteristics of subjects in the Caerphilly cohort, subdivided by their self-reported daily milk consumption

It is difficult to judge the overall relevance of these differences to vascular disease risk, as some suggest a favourable association with milk drinking, but others, such as heavier smoking, less alcohol and higher fat consumption are certainly unfavourable. An important finding is the absence of differences in lipid levels. Nevertheless, statistical adjustments were made in the analyses that follow for lipids and a wide range of other possible confounding factors.

Table 2 shows the numbers of ischaemic heart disease events that occurred during follow-up, and estimated hazard ratios for an ischaemic heart disease event in groups defined by milk consumption. Incident data are shown both for the total cohort and for men who, at baseline, had had questionnaire or ECG evidence of a previous myocardial infarct, or stroke. There is no evidence suggestive of any increase in risk in the men who had had the highest consumption of milk. Although the data suggest a lower risk in the men with the highest consumption of milk at baseline, there is some attenuation of effect on further adjustments for confounding, but there is no suggestion of any increased risk from milk drinking.

Table 2 Hazard ratios for fatal and nonfatal ischaemic heart disease in all the men and in men with evidence of prior heart disease or stroke at baseline

Table 3 shows the same data for ischaemic stroke. Again, there is no evidence of any increase in risk and a suggestion that ischaemic stroke incidence may have been lower in the men with the greatest consumption of milk. Numbers of events are, however, small and neither is the trend significant, nor does any individual difference in hazard ratios differ significantly from unity once adjustments have been made for possible confounding.

Table 3 Hazard ratios for fatal and nonfatal ischaemic stroke in all the men and in men with evidence of prior heart disease or stroke at baseline

Table 4 shows the numbers of men who experienced an incident ischaemic event, either heart disease or stroke. The evidence is suggestive of a possible reduced risk, greatest in the men who drank the most milk. After adjustments for confounding, the risk in the total cohort is 0.64 (0.39–1.06) and 0.37 (0.15–0.90) in the men who, at baseline, had had evidence of a vascular disease.

Table 4 Hazard ratios for a vascular event (fatal and nonfatal IHD and/or ischaemic stroke) in all the men and in men with evidence of heart disease or stroke prior to baseline

Finally, Table 5 shows all-cause mortality in the cohort during the 23 y of follow-up. There is no convincing evidence of any association with milk consumption. Had the associations shown with milk consumption in Tables 2, 3 and 4 been due to confounding, then one would expect a similar pattern with total mortality. The absence of any association suggests, therefore, that there is no serious residual confounding of the relationships between milk consumption and either ischaemic heart disease or ischaemic stroke.

Table 5 Hazard ratios for death from any cause


Whole milk contains around 4% total fat and 2.4% of saturated fat by weight. In an average diet, however, milk makes a relatively small contribution to the total fat intake. At baseline, the men in the present cohort who drank milk obtained on average about 11% of their total fat, and about 18% of their saturated fat intake from milk. There is, however, no evidence that the risk of either ischaemic heart disease events, or ischaemic stroke, is related to milk consumption, either within the total cohort, or in the men who had already had a vascular event at baseline.

The data on lipid levels (Table 1) show only trivial differences between men grouped by their milk consumption. This is consistent with the findings in many other cohort studies (Ness et al, 2001; Abbott et al, 1996; Iso et al, 1999; Vijver et al, 1992). Blood pressure, on the other hand, is slightly lower in subjects who reported drinking milk (Table 1) and again this effect has been shown consistently in both observational and intervention studies (Ackley et al, 1983; Pryer et al, 1995). There is evidence that it is the calcium in milk and dairy products that is responsible for this effect (Hajjar et al, 2003), and although small, the effect upon vascular disease within a community could be considerable (Jorde & Bonna, 2001). It is, however, impossible to judge the likely overall effect of milk consumption on vascular disease from effects on risk factors. Conclusions must be drawn from relationships with disease incidence.

The present study is based upon a representative population sample with a high compliance rate and intensive and repeated efforts were made to identify vascular events as completely as possible. This last applies both to prevalent disease at baseline, and incident disease events during the 20–24 y of follow-up. We judge, therefore, that the results we have presented are likely to be representative of associations in the general population.

The study has, however, a number of limitations. First, the estimates of milk intake used in the analyses were derived from a semiquantitative food frequency questionnaire. In the interpretation of this, however, portion sizes were derived from 7-day weighed intake records kept in the cohort (Fehily et al, 1987; Fehily & Hopkinson, 1993). Moreover, the correlation between milk consumption as recorded in the 7-day weighed intake records provided by a subsample of 665 of the men, and as stated by those men in their questionnaires at baseline, was high (0.61; P<0.001).

An important omission in our data is whether the milk drunk by the men was whole, or fat reduced. According to the dairy company which is the main distributor of milk to the Caerphilly area, fat-reduced milks were introduced in 1982, and by 2000 about 8% of the milk distributed was skimmed, about 50% semiskimmed and about 42% whole milk. In 2002, we contacted a random sample of 167 survivors in the Caerphilly Cohort. Just under half of those who had had a vascular event, and just over half of the other men said that they now drink fat-reduced milk. The average time, when they said they had changed from full-fat milk, was 12 y ago. This means that for the major duration of the follow-up of this cohort, most of the milk being drunk was probably whole milk.

There are differences in the lifestyle, and in the general nutrition of the men in the different milk drinking groups in the present cohort (Table 1). The few other studies, which give such data, show similar differences. Differences in some, but not all the factors we examined favour the men who drank the most milk with respect to vascular risk. However, the differences are small, and once adjustments were made for age and total energy intake, there were only very small further changes in the hazard ratios when other possible confounding factors were adjusted for (Tables 2, 3, 4 and 5). While it could be that if account was taken of yet further, as yet unknown, confounding factor(s) the gradients in the data could be reduced, but it seems highly unlikely that they could be reversed.

The evidence relating to men who had already had a vascular event is of interest, and it is unfortunate that no other study seems to have reported disease incidence in such men. Presumably, many of these men would have been advised to reduce their milk intake and/or to change to a low-fat milk. The evidence we have given above suggests, however, that they complied poorly with any such advice and the proportion of men with a vascular event who had changed to fat-reduced milk was no different to that in men who had had no vascular event.

This study is observational, and suffers from the limitations of all such studies. On the other hand, it is unlikely that an intervention study of drinking either no milk, or a fat-reduced milk, of adequate size and duration and with a sufficiently high level of compliance by subjects could ever be performed. Conclusions will, therefore, have to be based upon observational data. Results from nine other large cohort studies have been published and an overview of these, together with the present study, is presented in an accompanying paper (Elwood et al, 2003).


  1. Abbott RD, Curb JD, Rodriguez BL, Sharp DS, Birchfield CM & Yano K (1996): Effect of dietary calcium and milk consumption on risk of thromboembolic stroke in older middle aged men: The Honolulu Heart Program. Stroke 27, 813–818.

  2. Ackley S, Barrett-Connor E & Suarez L (1983): Dairy products, calcium and blood pressure. Am. J. Nutr. 38, 457–461.

  3. Bainton D, Baker IA, Sweetnam PM, Yarnell JWG & Elwood PC (1988): Prevalence of ischaemic heart disease: the Caerphilly and Speedwell Surveys. Br. Heart J. 59, 201–206.

  4. Baker IA, Pickering J, Elwood PC, Bayer A & Ebrahim S (2002): Fibrinogen, viscosity and white blood cell count predict myocardial infarction, but not cerebral infarction: evidence from the Caerphilly and Speedwell cohort. Thromb Haemost 87, 421–425.

  5. Barr SI, McCarron DA, Heaney RP, Dawson-Hughes B, Berga SL, Stern JS & Oparil S (2000): Effects if increased consumption of fluid milk on energy and nutrient intake, body weight, and cardiovascular risk factors in healthy older adults. J. Am. Diet. Assoc. 100, 810–817.

  6. Elwood PC (2001): Milk, coronary disease and mortality. J. Epidemiol. Commun. Health 55, 375.

  7. Elwood PC, Beswick A, Pickering J, McCarron P, O’Brien JR, Renaud SR & Flower RJ (2001a): Platelet tests in the prediction of myocardial infarction and stroke: evidence from the Caerphilly study. Br. J. Haematol. 113, 514–520.

  8. Elwood PC, Pickering J & Gallacher JEJ (2001b): Cognitive function and blood rheology: results from the Caerphilly cohort of older men. Age Ageing 30, 135–139.

  9. Elwood PC, Pickering J, Fehily AM, Hughes J & Ness A (2004): Milk drinking, ischaemic heart disease and ischaemic stroke. II. evidence from cohort studies. Eur. J. Clin. Nutr. 58, 718–724.

  10. Fehily AM & Hopkinson T (1993): Do we need information about portion sizes to rank individuals by their nutrient intakes? J. Hum. Nutr. Diet. 6, 101–111.

  11. Fehily AM, Yarnell JWG & Butland BK (1987): Diet and ischaemic heart disease in the Caerphilly study. Hum. Nutr. Appl. Nutr. 41A, 319–326.

  12. Fehily AM, Butland BK, Holiday RM & Yarnell JWG (1988): Dietary studies in the Caerphilly Heart Disease Survey. Food Sci. Nutr. 42F, 77–78.

  13. Gallacher JEJ, Yarnell JWG, Sweetnam PM, Elwood PC & Stansfeld SA (1999) Anger and incident heart disease in the Caerphilly study. Psychosomatic Med. 61, 446–453.

  14. Greenwood R, McCarron P, Elwood PC, Ben Shlomo Y, Bayer A, Baker IA, Frankell S, Ebrahim S, Murray L & Davey Smith G (2001): The incidence and aetiology of stroke in the Caerphilly and Speedwell Collaborative studies. I Methods and incidence of events. Public Health 115, 4–11.

  15. Hajjar IM, Grim CE & Kotchen TA (2003): Dietary calcium lowers the age-related rise in blood pressure in the United States: the NHANES III survey. J. Clin. Hypertens. (Greenwich) 5, 122–126.

  16. Iso H, Stampfer MJ, Manson JE, Rexrode K, Hennekens CH, Colditz GA, Speizer FE & Willett WC (1999): Prospective study of calcium, potassium and magnesium intake and risk of stroke in women. Stroke 30, 1772–1779.

  17. Jorde R & Bonna KH (2001): Calcium from dairy products, vitamin D intake and blood pressure: the Tromso Study. Am. J. Clin. Nutr. 73, 659–660.

  18. Naidoo J & Willis J (1994): Health Promotion: Foundation for Practice. London: Bailliere Tindall.

  19. Ness AR, Davey Smith G & Hart C (2001): Milk, coronary heart disease and mortality. J. Epidemiol. Commun. Health 55, 379–382.

  20. Nutritional Aspects of Cardiovascular Disease (1994): Report of the Cardiovascular Review Group: the Committee on Medical Aspects of Food Policy. London: HMSO.

  21. Pryer J, Cappuccio FP & Elliott P (1995): Dietary calcium and blood pressure: a review of the observational studies. J. Hum. Hypertens. 9, 597–604.

  22. Roberts DCK, Truswell AS, Sullivan DR, Gorrie J, Darnton-Hill I, Norton H, Thomas MA & Allen JK (1982): Milk, plasma cholesterol and controls in nutritional experiments. Atherosclerosis 42, 323–325.

  23. Segall JJ (1977): Is milk a coronary health hazard? Br. J. Prev. Soc. Med. 31, 81–85.

  24. St Onge MP, Farnworth ER & Jones PJH (2000): Consumption of fermented and nonfermented dairy products: effects on cholesterol concentrations and metabolism. Am. J. Clin. Nutr. 71, 674–681.

  25. Steinmetz KA, Childs MT, Stimson C, Kushi LH, McGovern PG, Potter JD & Yamanaka WK (1994): Effect of consumption of whole milk and skim milk on blood lipids profiles in healthy men. Am. J. Clin. Nutr. 59, 612–618.

  26. The Caerphilly & Speedwell Collaborative Group (1984): Caerphilly and Speedwell collaborative heart disease studies. J. Epidemiol. Commun. Health 38, 259–262.

  27. Vijver L Van der, Wall M Van der, Wetterings KGC, Dekker JM, Schouten EG & Kok FJ (1992): Calcium intake and 28-year cardiovascular and coronary heart disease mortality in Dutch civil servants. Int. J. Epidemiol. 21, 36–39.

  28. Yarnell JWG, Patterson CC, Sweetnam PM, Thomas HF, Bainton D, Elwood PC, Bolton CH & Miller N (2001): Do total and high density lipoprotein cholesterol and triglycerides act independently in the prediction of ischaemic heart disease? Arterioscler. Thromb. Vasc. Biol. 21, 1340–1345.

Download references


The Caerphilly study was conducted by the former MRC Epidemiology Unit (South Wales) and was funded by the Medical Research Council of the United Kingdom. The archive is now maintained by the Department of Social Medicine in the University of Bristol. We are grateful to Mr Stephen Swinglehurst, Operations Manager of Farmers and Dairymen, who supplied data on the distribution of milk in the Caerphilly area. We thank John Fennick for statistical help and advice. Janet Pickering and Janie Hughes are supported by the Food Standards Agency.

Author information

Correspondence to P C Elwood.

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Elwood, P., Pickering, J., Fehily, A. et al. Milk drinking, ischaemic heart disease and ischaemic stroke I. Evidence from the Caerphilly cohort. Eur J Clin Nutr 58, 711–717 (2004) doi:10.1038/sj.ejcn.1601868

Download citation


  • Caerphilly cohort
  • milk
  • ischaemic heart disease
  • stroke
  • cholesterol

Further reading