Dr J. West, Division of Epidemiology and Public Health, Queen's Medical Centre, University of Nottingham, Nottingham NG7 2UH, UK. E-mail: firstname.lastname@example.org
Background : It has been suggested that vascular disease mortality may be reduced in coeliac disease because of lower levels of blood pressure, cholesterol and body mass.
Aim : To examine whether people with coeliac disease are at reduced risk of various vascular diseases.
Methods : We identified 3790 adults with diagnosed coeliac disease and 17 925 age- and sex-matched controls in the General Practice Research Database. We estimated odds ratios for diagnosed hypertension, hypercholesterolaemia and atrial fibrillation and hazard ratios for myocardial infarction and stroke.
Results : Adults with coeliac disease, compared with controls, were less likely to have had a diagnosis of hypertension [11% vs. 15%, odds ratio 0.68 (95% confidence interval: 0.60–0.76)] or hypercholesterolaemia [3.0% vs. 4.8%, odds ration 0.58 (95% confidence interval: 0.47–0.72)] but slightly more likely to have had atrial fibrillation [2.1% vs. 1.7%, odds ratio 1.26 (95% confidence interval: 0.97–1.64)]. The hazard ratio for myocardial infarction was 0.85 (95% confidence interval: 0.63–1.13), while the hazard ratio for stroke was 1.29 (95% confidence interval: 0.98–1.70).
Conclusions : Although rates of myocardial infarction and stroke were not substantially different, adults with coeliac disease do have a lower prevalence of hypertension and hypercholesterolaemia compared with the general population. The effect of a gluten-free diet on cardiovascular risk factors should be determined before any screening programmes for coeliac disease are instituted.
The development of accurate serological tests has facilitated the diagnosis of coeliac disease and made screening of the asymptomatic population feasible. As a result coeliac disease has been shown to affect between one and two-hundred of the population in many European countries including the UK.1–7 People with coeliac disease are known to have an increased risk of developing several potentially serious conditions including osteoporosis and small intestinal malignancy.8–10 It has also been suggested that the risks of developing other autoimmune diseases increases with duration of untreated coeliac disease.11 Because of these risks some people have advocated population screening for coeliac disease.12, 13
Whether coeliac disease might also afford protection from certain disease needs consideration particularly as any protection might be reduced by treatment with a gluten-free diet. This possibility was first raised by Whorwell et al. who found a 40% reduction in ischaemic heart disease mortality in people with diagnosed coeliac disease.14 In two recent studies, we have found some evidence of decreased cardiovascular morbidity. In a survey of people with diagnosed coeliac disease there was lower reported antihypertensive medication use among people with coeliac disease compared with the control group15 and in a recent cross-sectional population screening study we found both blood pressure and serum cholesterol levels were lower in subjects antiendomysial seropositive for coeliac disease compared with those seronegative.1 In addition coeliac disease like ulcerative colitis appears to be associated with non-smoking although it is unclear whether this is a causal association.16–18
As vascular disease is the most important single cause of mortality in coeliac disease, accounting for 40% of all deaths, any reduction in its mortality would be of great importance and might even outweigh larger increases in mortality from less common conditions.19 We have therefore carried out a general population-based study to explore the relationship between coeliac disease and risk of vascular disease within the General Practice Research Database (GPRD), a UK-based, longitudinal primary care database.
In the UK, over 95% of people are registered with a general practitioner for their primary health care. The GPRD, established in 1987, is a longitudinal primary care database and contains the computerized medical records from general practice of more than 8 million of these registered people. When people are seen in primary care in the UK the majority of significant medical diagnoses, information from hospital letters and discharge summaries, and prescriptions are entered onto a desktop computer. These data are then aggregated and anonymized to maintain patient confidentiality. The data from GPRD practices are audited regularly and at least 95% of all morbidity and prescribing information must be included for the participating practice to be contributing ‘up-to-standard’ data.20 We used two approaches in this study: first, we compared the risk of ‘persistent’ conditions (hypertension, high cholesterol and atrial fibrillation) in coeliac disease compared with the general population using all available GPRD data, and secondly we used a historical-matched cohort study design for acute myocardial infarction and stroke where the observation time (subject years at risk) for individuals included in our study started at the beginning of their up-to-standard GPRD record.
We extracted the records of all subjects within the GPRD between June 1987 and April 2002 with a recorded diagnosis of coeliac disease aged 25 or older at the start of their GPRD record. Where possible we selected five control subjects matched to each individual with coeliac disease by age, sex, general practice and follow-up time. When selecting control subjects we excluded individuals who had any record of a gluten-free prescription or a non-specific reference to coeliac disease (e.g. ‘gluten-free diet’, ‘gluten sensitivity’). Each subject with coeliac disease was assigned a date of diagnosis defined as the date of the first record of coeliac disease. As general practitioners enter some data for important historical events retrospectively this date preceded the start of the GPRD record on some occasions. We defined ‘incident’ subjects with coeliac disease as those individuals whose diagnosis date of coeliac disease or first prescription for a gluten-free product occurred at least 1 year after the beginning of their GPRD record. Further details about our study population have been described elsewhere.21
Diseases and confounders
We investigated the risk of a diagnosis of hypertension, high cholesterol or atrial fibrillation at any time in the available data and rate of first myocardial infarction or stroke during up to standard GPRD data for the cohort analysis. In addition, we calculated a composite measure of treated hypertension that was positive only if both the subjects had a diagnosis of hypertension and had ever had a prescription for an antihypertensive medication. Similarly we looked at the proportion of those with treated hypercholesterolaemia as those with both a recorded high cholesterol diagnosis and a prescription for either a statin or bezofibrate. We extracted information on the following potential confounders: height, weight and smoking habit from the whole of each individual's data period. Body mass index (BMI, per kgm2) was calculated for all individuals where data was available. Other potential confounders including the recorded presence or absence of diabetes and other comorbid diagnoses (e.g. thyroid disease) were extracted from all the available data.
We compared cases and controls using chi-squared test, or t-test, where appropriate. We calculated odds ratios (OR) for the diagnosis of hypertension, high cholesterol and atrial fibrillation using conditional logistic regression. For the cohort analysis we calculated crude age- and gender-specific myocardial infarction and stroke rates for the two cohorts and then used Cox regression modelling to estimate the hazard ratio (HR) of myocardial infarction or stroke in the coeliac disease cohort compared with the matched control cohort. We plotted Kaplan–Meier graphs and checked the proportional hazards assumption of the models. The impact of potential confounders was assessed using a series of multivariable models, retaining variables that led to a change in the hazard or odds ratios for coeliac disease of 10% or more. For confounder variables, missing data were fitted as a separate category to ensure that nested models contained the same number of individuals. We checked for any evidence of interaction between disease status and both BMI and prevalent/incident status by fitting multiplicative interaction terms. We used stata 7 (Stata Corportation, College Station, TX, USA) for all analyses.
Our study included 3590 subjects with coeliac disease and 17 925 matched controls contributing 21 248 and 117 210 observed years at risk respectively. The groups were closely matched on age at the start of the GPRD record and gender (Table 1). There were more current smokers in the control cohort (16.9% vs. 13.7%, P < 0.01) and more individuals who were underweight (BMI ≤ 18.5) in the coeliac disease group (4.3% vs. 1.1%, P < 0.01). There was no excess of recorded diabetes in the coeliac disease group (3.5% vs. 3.7%, P > 0.05) and 7.0% had a diagnosis of thyroid disease compared with 3.2% of controls (P < 0.01). Mean systolic blood pressure was 5 mmHg lower in the coeliac disease group (P < 0.01).
Table 1. Characteristics of the adults with coeliac disease and their matched controls
n = 3590
n = 17 925
Median observed time (years)
Total observed time (years)
Age groups at start of GPRD record (years)
Body mass index (per kgm2)
Diabetes ever recorded
Thyroid disease ever recorded
Systolic blood pressure, mmHg (mean, s.d.)
Diastolic blood pressure, mmHg (mean, s.d.)
Blood pressure not recorded
Gluten-free prescriptions per year of follow-up
10 or more
Overall 408 (11%) of the adults with coeliac disease had ever had a diagnosis of hypertension compared with 2765 (15%) in the control group giving an unadjusted OR of 0.68 [95% confidence interval (CI): 0.60–0.76] as shown in Table 2. Part of this relationship was explained by BMI, as after adjusting for this variable the OR was 0.78 (95% CI: 0.69–0.87). When we repeated our analyses using the composite measure of treated hypertension (including prescriptions), the results were similar. The unadjusted ORs for atrial fibrillation and high cholesterol diagnoses were 1.26 (95% CI: 0.97–1.64) and 0.58 (95% CI: 0.47–0.72) respectively. Adjusting for BMI had no appreciable effect on these latter two estimates. When we repeated our analyses using the composite measure of treated hypercholesterolaemia our results were similar.
Table 2. Analysis of risk of hypertension, atrial fibrillation and high cholesterol diagnoses in 3590 people with coeliac disease compared with 17 925 controls using conditional logistic regression
Overall number with disease (%)
* Baseline category.
† Adjusted for body mass index.
There were 52 recorded first myocardial infarctions in the coeliac disease cohort, and 62 recorded first strokes. The overall rate of myocardial infarction for the coeliac disease cohort was 24.7/10 000 subject years compared with 29.2/ 10 000 subject years in the control cohort (Table 3). There was approximately a 15% decrease in the risk of myocardial infarction for the coeliac disease cohort compared with the control cohort HR 0.85 (95% CI: 0.63–1.13). The overall rate of stroke was 29.4/10 000 subject years in the coeliac disease group and the HR was 1.29 (95% CI: 0.98–1.70). In the multivariate analyses, only BMI of the potential confounders we assessed made any substantial impact on the coefficient for myocardial infarction in coeliac disease altering the overall HR to 0.95 (95% CI: 0.71–1.27). For stroke, only the presence or absence of diagnosed hypertension altered the HR appreciably to 1.40 (95% CI: 1.06–1.84). In contrast, when adjusted for smoking status, the HRs were 0.87 (95% CI: 0.65–1.16) and 1.33 (95% CI: 1.00–1.75) respectively. When we analysed only the subjects with an ‘incident’ diagnosis of coeliac disease we found a slightly reduced HR for myocardial infarction [HR 0.75 (95% CI: 0.46–1.23)] and a slightly increased HR for stroke [HR 1.60 (95% CI: 0.99–2.59)] in comparison with both the overall HRs for all subjects and compared with those in the ‘prevalent’ group. However, we found no statistically significant evidence of interaction. Figure 1 shows the Kaplan–Meier plots for myocardial infarction and stroke. There was no statistically significant evidence against the proportional hazards assumption in any of the presented models. When we repeated our analyses restricted to only those adults with coeliac disease who had at least one gluten-free prescription all our results were similar.
Table 3. Number, rate and crude hazard ratios for the coeliac disease cohort compared with the control cohort, overall and limited to prevalent and incident subjects with coeliac disease
Prevalent subjects with coeliac disease†
Incident subjects with coeliac subjects
Number of events
Rate (10 000/year)
Rate (10 000/year)
Rate (10 000/year)
* Baseline category.
† For these analyses only the matched controls of those subjects with coeliac disease included were used.
Coeliac disease cohort
Coeliac disease cohort
The results of our study show a marked decrease in the risk of diagnosed hypertension and high cholesterol, yet a slight increase in the risk of atrial fibrillation in adults with coeliac disease compared with the general population. Although modest, and despite our lack of adequate power to precisely assess it, these factors appear to have some effect on vascular disease. Our point estimates suggest there might be approximately a 15% reduction in the risk of myocardial infarction but approximately a 30% increase in the risk of stroke.
Our study has the advantage of being large, population-based and reliant on recording of diseases and prescriptions in general practitioners’ records. However, there are some possible limitations. Adults with coeliac disease are more frequent attenders at their general practitioner than members of the general population.21 As a consequence, if ascertainment bias is present, it is possible that we have underestimated the risk of stroke and atrial fibrillation in the control cohort because of less complete recording of medical events in a ‘healthy’ group. Similarly, we may have underestimated the decrease in relative risk of hypertension, high cholesterol and myocardial infarction. It is also possible that we have underestimated the effect sizes through misclassification of unrecorded coeliac disease, i.e. that some of the control group may have coeliac disease. However, there have been several validation studies of GPRD recording, including one that has looked in detail at the accuracy of the diagnosis of inflammatory bowel disease.22–24 In this study, Lewis et al. found that the inflammatory bowel disease diagnosis within the GPRD was highly probable or probable in 92% (95% CI: 86–96) of their surveyed cases. Inflammatory bowel disease is analogous to coeliac disease in that it is a diagnosis made in secondary care and their findings are likely to be generalizable to our study. In addition, we believe that there will be very few people with a recorded diagnosis of coeliac disease and coexisting prescriptions for one or more gluten-free products who do not have coeliac disease. When we restricted our analyses to those adults with coeliac disease who also had at least one gluten-free prescription, to increase the specificity of the coeliac disease diagnosis, there were no substantial changes in the effect estimates.
As the GPRD does not contain information on socio-economic status we have been unable to control for this important variable in our analysis and as adults with coeliac disease have tended to be in higher socio-economic groups, this could explain part of their lower incidence of cardiovascular disease.1 Our control cohort was closely matched in terms of age, gender and community to minimize the potential for confounding by these factors and we had the ability to assess the impact of potential confounders such as BMI, smoking status and comorbidity on vascular disease risk. Although adjusting for BMI and the diagnosis of hypertension altered the effect estimates for myocardial infarction and stroke respectively, both can be considered as intermediate steps between coeliac disease and the outcome. Rather than being alternative explanations for the observed association, they are more likely to be on the causal pathway between coeliac disease and vascular disease. We found no other evidence of substantial confounding.
Although not statistically significant, the point estimates for the risks of stroke and myocardial infarction are intriguing. It is possible, for example, that the observed relationship with myocardial infarction may have been attenuated by the effect of treatment with a gluten-free diet. Although we have previously shown that people with undetected coeliac disease have lower serum cholesterol than the general population1 it is possible that following treatment, and the subsequent improvement in intestinal absorption, serum cholesterol may increase therefore attenuating any protective effect. The relationship is clearly complex in view of the finding that plasma homocysteine levels remain high in people with treated coeliac disease even after many years of gluten exclusion.25 It is also possible that there is actually no difference between the two groups as both the estimates for stroke and myocardial infarction include unity. Another reason why any differences are not so apparent in our study could be due to the presence of people with undetected coeliac disease in the control cohort, although we think the effect of this will be small.
The analysis of incident and prevalent cases adds some support to the idea that treatment with a gluten-free diet may alter the vascular disease risk profile of these individuals as those in the incident group include people with some untreated subject time, and appear to have a slightly lower risk of myocardial infarction. Our finding of a slightly increased risk of stroke was not explained by the greater prevalence of thyroid disorders or atrial fibrillation in the coeliac disease group, although it is possible that the latter condition is under recorded in primary care as only 13% of the subjects who had had a stroke in our study had atrial fibrillation. With less hypertension, lower BMI and presumably lower cholesterol, it was surprising to find a different relationship to that with myocardial infarction. One explanation may be the previously documented finding of a high prevalence of coeliac disease in idiopathic cardiomyopathy,26 suggesting that the mechanism of increased risk in coeliac disease might be arrythmogenic or thromboembolic. Alternatively the recently postulated neurotoxic effects of gluten may play an important role.27
Our findings confirm the hypothesis that adults with diagnosed coeliac disease have a decreased risk of hypertension and hypercholesterolaemia. The finding of a slightly decreased risk of myocardial infarction and a small increased risk of stroke in coeliac disease are intriguing and lead to speculation about the mechanisms of vascular disease, particularly in relation to nutritional status. Our data suggest that the effects of treatment with a gluten-free diet on these risk factors should be determined before any screening programmes for coeliac disease are instituted.
The authors would like to thank Hassy Dattani and all the staff at EPIC for their help and advice with GPRD data. Also would like to thank The Wellcome Trust for their grant support and Laila Jal Tata and Andrew Fogarty for their comments on the manuscript. Joe West had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Grant support: The Wellcome Trust. Joe West is a Wellcome Research Training Fellow in Clinical Epidemiology, grant number 063800.