The risk of venous thromboembolism (VTE) was examined in individuals with coeliac disease (CD). The Swedish national inpatient register was used to identify 14 207 individuals with a diagnosis of CD (1964–2003). These individuals were matched for age, sex, calendar year and county with 69 048 reference individuals. Cox regression was used to estimate hazard ratios (HRs) for subsequent thromboembolism in individuals with more than 1 year of follow-up and no prior VTE. CD was associated with an increased risk of subsequent VTE (HR = 1·86; 95% confidence interval (CI) 1·54–2·24). The risk increase was restricted to individuals with CD diagnosed in adulthood. Risk estimates were not affected by the presence of diabetes mellitus or concomitant surgery. Compared with inpatients as reference individuals, CD individuals remained at increased risk of subsequent VTE (adjusted HR = 1·27; 95% CI = 1·06–1·52). In conclusion, this study found a statistically significantly positive association between CD and VTE. This modest association might be explained by a combination of surveillance bias and chronic inflammation.
Individuals with CD are at a threefold increased risk of hyperhomocysteinaemia (Saibeni et al, 2005); and homocysteine levels correlate with the degree of intestinal injury (Saibeni et al, 2005). Although a recent trial found no protective effect of homocysteine supplementation on the risk of VTE (den Heijer et al, 2007), earlier studies have identified high levels of homocysteine and low levels of folic acid as risk factors for VTE (Cattaneo et al, 2001; Eichinger, 2006). It is therefore noteworthy that there is evidence of poor vitamin status and increased levels of homocysteine among patients with CD who are following a gluten-free diet (Hallert et al, 2002). Consuming a gluten-free diet will not always result in small-intestinal recovery [only 8/39 adults on a long-term gluten-free diet had a fully recovered duodenal mucosa (Lee et al, 2003)]. Therefore, inflammation and malnutrition often persist in CD. Other than hyperhomocysteinaemia (Cattaneo et al, 2001; Saibeni et al, 2005), high levels of thrombin-activatable fibrinolysis inhibitor (TAFI) (van Tilburg et al, 2000; Saibeni et al, 2004) and a positive association between CD and antiphospholipid syndrome (Shamir et al, 2003) may also contribute to an increased risk of thromboembolism in CD.
Several case-reports (Zenjari et al, 1995; Grigg, 1999; Kremer Hovinga et al, 1999; Andres et al, 2000; Manzano et al, 2002; Gabrielli et al, 2003) suggest an increased risk of VTE in CD. An Austrian study failed to find an increased risk of VTE in CD [odds ratio (OR) = 0·4; 95% confidence interval (CI) = 0·1–2·5] (Miehsler et al, 2004); this may be because of low power as this study was primarily designed to investigate VTE in inflammatory bowel disease and not in CD. Hence, we believe there is a need to conduct an appropriate study designed to assess the risk of VTE among patients with CD. We did so using the data obtained through linkage of Swedish national registers, identifying more than 14 000 individuals with CD and 69 000 age- and sex-matched reference individuals.
Design and methods
A detailed description of the methods used in the present study has been published elsewhere (Ludvigsson et al, 2007). Briefly, the Swedish National Board of Health accessed the Swedish National Inpatient Register (IPR) to identify all individuals with a hospital discharge diagnosis of CD between 1964 and 2003. CD, DM, and our outcome measure VTE, were defined according to the presence of the relevant international classification of disease code (see Appendix). We used the National Total Population Register to select up to five reference individuals for each CD individual, matched for age, sex, calendar year and county at the time of the first CD diagnosis.
The analyses of subsequent VTE in the current study were based on 14 207 individuals with CD and 69 048 matched reference individuals who did not have a diagnosis of CD (Fig 1). We excluded the first year of follow-up to minimize the risk of detection bias because of hospital admission in individuals with CD.
Data from a socio-economic index (SEI) (Table I) was available in a subset of individuals (8737 with CD and 36 318 without CD) . This index is based on a three-category occupational classification (Guteland, 1982). Some 6500 of the participants were children who had been assigned a socio-economic code on the basis of the occupation of the mother.
Table I. Characteristics of participants with at least 1 year of follow-up.
Reference (no coeliac disease) (%)
Coeliac disease (%)
*Age and duration in years. Median duration indicates the number of years from study entry (and diagnosis of coeliac disease) until first diagnosis of venous thromboembolism (VTE). Please note that values are based on individuals with VTE after study entry; individuals with VTE recorded before or within the first year of study entry were excluded from the main analyses (see text).
Socio-economic index, ‘I’ is the highest category (see also text). For reference individuals, the numbers of individuals who constituted the basis for the Cox regression are given. Data from the socio-economic index was actually available in another 6361 reference individuals; but these individuals were not part of the internally stratified calculations because of missing values on socio-economic index in the matched individual with coeliac disease. Adding the 6361 reference individuals to those presented above, the proportion of missing values was similar among individuals with coeliac disease and reference individuals.
Age at first recorded diagnosis of coeliac disease (years)
28 419 (41·2)
40 629 (58·8)
19 041 (27·6)
30 428 (44·1)
17 165 (24·9)
19 955 (28·9)
32 730 (47·4)
Median age (years) at first diagnosis of VTE (range)*
Median duration (years) between study entry and first diagnosis of VTE (range)*
Statistical methods and analyses
An internally stratified Cox regression was used to estimate the hazard ratio (HR) for subsequent VTE. Risk estimates are given as HRs. The internal stratification meant that each individual with CD was only compared with his/her matched reference individuals. The follow-up time started 1 year after study entry (date of first recorded diagnosis of CD and corresponding date in reference individuals) and ended on the date of first discharge after diagnosis of VTE, date of emigration, death or the end of the study period (31st December 2003), whichever happened first. The median follow-up time was 14 years (range: 1–40 years) in both individuals with CD and those without this diagnosis.
In separate analyses, we stratified for age at first recorded CD diagnosis (≤15 years vs. ≥16 years) and sex. The influence of sex and age on the risk of VTE in CD was also examined through the heterogeneity test described by Altman and Bland (2003). In separate analyses, we adjusted for DM as this is a potential confounder (Petrauskiene et al, 2005; Ludvigsson et al, 2006) and also may increase the risk of surveillance bias. We did not adjust for coronary heart disease because it is not associated with CD (West et al, 2004) and therefore is not a confounder. Individuals with CD are at an increased risk of certain forms of cancer (Askling et al, 2002). Finally, a heterogeneity test was used to examine if the risk estimates for VTE differed by presence of DM, cancer and surgery.
Conditional logistic regression was used to estimate the risk of CD (the dependent variable) associated with prior thromboembolism. Those with 1 year or less between the dates of VTE and study entry (equivalent to date of diagnosis of CD) were excluded. Risk estimates are given as OR.
To evaluate if a potential risk increase for thromboembolism is specific to CD or might have been influenced by surveillance bias because of hospital admission at study entry, post hoc analyses were performed in which reference individuals were restricted to those who had been inpatients within 1 year prior to or after the CD diagnosis in the matched index individual. These analyses did not use an internally stratified statistical approach, but instead the Cox regression was adjusted for sex, age and calendar period. This was because the loss of many reference individuals (who had not been inpatients) often resulted in strata only containing the index subject with CD, thus rendering the model inefficient. As all the subjects in the post hoc analyses had been inpatients, we included the first year follow-up to maximize the study power. Statistical significance was defined as 95% CIs for risk estimates not including 1·0. Statistics were calculated using spss 11.0 (Chicago, IL, USA, 2002).
We had 80% power to detect an increased risk of subsequent VTE at a significance level of 5% in individuals with CD if the HR exceeded 1·25. This study was approved by the Research Ethics Committee of the Karolinska Institutet. None of the participants was contacted. Subject information was anonymized prior to the analyses.
Some 60% of the subjects were female (Table I). The median age at study entry was 2 years (range 0–94 years) in both those with and those without CD. Although, the majority entered the study in childhood, this study also included 4841 individuals with CD diagnosed in adulthood. In all, 10 545 individuals with CD were adults at the end of the follow-up period. Table I shows the median age at first VTE as well as the median duration from study entry to first diagnosis of VTE.
CD and VTE
Of 15 439 individuals with CD, 406 (2·6%) had a diagnosis of VTE prior to or after diagnosis of CD compared with 1105/76 910 (1·4%) reference individuals. Restricting data to those with follow-up into adulthood, the incidence of subsequent VTE in individuals with CD was 102/100 000 person-years and 65/100 000 person-years in reference individuals.
There was a positive association between CD and subsequent VTE (HR = 1·86; 95% CI 1·54–2·24; Table II). This risk increase was restricted to individuals with a diagnosis of CD in adulthood, although heterogeneity testing found that the risk estimates for VTE did not differ statistically according to age at CD onset (P = 0·072). We found no difference in risk estimates of VTE between men and women (Table II; heterogeneity test: P = 0·739). HRs for VTE varied over time [1964–1973: 2·27; 1974–1983: 1·49; 1984–1993: 1·85; 1994–2003: 3·29 (all P-values <0·01)], and a formal interaction test showed that these risk estimates were statistically significantly different (P = 0·004).
Table II. Risk of subsequent venous thromboembolism in patients with coeliac disease.
Type of coeliac disease
HR* (95% CI)
*HR, Hazard ratios estimated through Cox regression.
CI, confidence interval.
First year excluded
No coeliac disease
Any coeliac disease
Age at first recorded coeliac disease diagnosis
First year included
No coeliac disease
Any coeliac disease
Adjustment for SEI or DM did not affect risk estimates (data not shown). The risk estimate for VTE increased slightly when the first year of follow-up was included (Table II). Restricting our outcome to those with VTE listed as the main diagnosis did not eliminate the statistically significantly positive association between CD and VTE (HR = 1·71; 95% CI = 1·35–2·16; P < 0·001; based on 100 and 364 positive events among those with and without CD respectively).
Six individuals were diagnosed with VTE by age of 10 years; all six were reference individuals who suffered from co-morbidity (four had heart malformations: tricuspid valve dysfunction, foramen ovale, double outlet syndrome and transposition of the great arteries; one had Budd-Chiari syndrome; and one had lymphosarcoma). When we excluded these six individuals from the analysis, the HR for subsequent VTE associated with CD was 1·86 (95% CI = 1·53–2·27, P < 0·001).
Heterogeneity testing showed that the risk of VTE in CD was not influenced by the presence of DM (heterogeneity test: P = 0·916), prior surgery (P = 0·277) or a cancer diagnosis (P = 0·384). A post hoc analysis, where all reference individuals were inpatients, found an adjusted HR for subsequent VTE of 1·27 (95% CI = 1·06–1·52; P = 0·011; based on 259 positive events in 15 292 individuals with CD compared with 210 events among 14 370 inpatients without CD). Individuals with prior VTE were at a 1·67-fold increased risk of having a later diagnosis of CD (95% CI = 1·34–2·06; P < 0·001; logistic regression).
This study found a positive association between CD and subsequent VTE. This risk increase was restricted to individuals with a diagnosis of CD in adulthood and the magnitude of the risk estimates was modest. When the reference individuals were restricted to inpatients, the association with VTE remained but just attained statistical significance. This study therefore rules out a substantially increased risk for thromboembolism in individuals with CD. Prior VTE was associated with subsequent CD.
To our knowledge, this is the largest cohort study of CD and thromboembolism conducted to date. During follow-up, there were 163 cases of VTE among 14 000 individuals with CD compared with 535 subjects with VTE among 69 000 reference individuals. This provided the study with considerable statistical power and enabled us to stratify the analyses. In contrast with the study by Miehsler et al (2004) we were able to study thromboembolism both prior to and after diagnosis of CD. We used an internally stratified Cox regression model to eliminate potential confounding by age, sex and calendar period. This is important because these three factors are potential confounders (Ivarsson et al, 1999, 2003; Silverstein et al, 1998; Tsai et al, 2002). We did not measure VTE because of pregnancy, childbirth or abortion (see Appendix). Women are at a highly increased risk of VTE during pregnancy and childbirth (Heit et al, 2005), and although recent data indicate otherwise (Tata et al, 2005), it has been suggested that women with CD are at an increased risk of infertility (Meloni et al, 1999). The inclusion of pregnancy-/childbirth-related thromboembolism could therefore potentially have biased our risk estimates. Heterogeneity testing found that the increased risk of thromboembolism in CD could not be explained by the presence of surgery or DM.
The lack of positive findings among children with CD may be due to their low median age at study entry [consistent with other Swedish data (Ivarsson et al, 2000)] and, more importantly, at the end of follow-up. Thromboembolism most often occurs in adulthood, particularly in old age [the risk of thromboembolism being fivefold higher in individuals aged >80 years than in those 40–49 years old (Glynn & Rosner, 2005)]. Heterogeneity testing found that the risk estimates for VTE in childhood CD did not differ statistically significantly from that in CD in adulthood.
Restricting our reference individuals to inpatients, we found a positive association between CD and VTE. The effect on the estimates of using an inpatient reference group suggests at least some of the association is because of surveillance bias: admission for one diagnosis increases the possibility that another, possibly unrelated diagnosis, will be made. Many of the inpatient reference individuals however suffered from diseases such as DM and cancer, both known to be associated with thromboembolism (Baron et al, 1998; Petrauskiene et al, 2005). In fact, our risk estimate for VTE (HR = 1·86) is very similar to that found in individuals with DM [standardized morbidity ratio = 2·3 (Petrauskiene et al, 2005)]. Despite the role of surveillance bias, chronic inflammation with villous atrophy in some individuals with CD may also constitute a modest increase in risk.
We had hypothesized that hyperhomocysteinaemia in some patients with CD could result in a risk increase for thromboembolism. Newly diagnosed CD is a risk factor for hyperhomocysteinaemia (relative risk = 3·4), and low folate levels (Saibeni et al, 2005), and hyperhomocysteinaemia may per se be a risk factor for thromboembolism (Wald et al, 2002). Furthermore, homocysteine levels correlate with mucosal appearance: as levels increase, so does the risk of severe duodenal involvement (Saibeni et al, 2005). High homocysteine levels may however not be confined to those with untreated CD (Hallert et al, 2002). In a recent study by Lee et al (2003), only 9/38 patients had a fully recovered mucosa several years (mean: 8·5 years) after diagnosis and institution of a gluten-free diet. Total villous atrophy was uncommon, but 69% of the patients had partial villous atrophy at follow-up (Lee et al, 2003). Even though vitamin status and homocysteine levels in the Italian patients normalized when they were on a gluten-free diet (Saibeni et al, 2005), this was not entirely confirmed by Swedish data (Hallert et al, 2002). In the Swedish study, individuals who had been on a gluten-free diet for at least 8 years had higher mean homocysteine levels than individuals without CD (Hallert et al, 2002). We therefore suggest that chronic inflammation and higher homocysteine levels contribute to the excess risk of thromboembolism seen in individuals with CD in this study.
This study lacked data on a number of potential confounders. Although smoking has sometimes (Petitti et al, 1978), but not always (Poulter & Meirik, 1995; Tsai et al, 2002; Glynn & Rosner, 2005) been implicated in the aetiology of thromboembolism, it has no effect (Miehsler et al, 2004; Ludvigsson et al, 2005a), or indeed a negative effect (Snook et al, 1996), on the risk of CD. Therefore, it cannot account for the associations observed here. Besides, we adjusted for SEI [strongly linked to smoking habits (Lindstrom et al, 2003)] with little effect on the risk estimate for VTE. Higher body weight is a risk factor for VTE (Tsai et al, 2002) but negatively associated with CD (Bardella et al, 2000). We cannot exclude the possibility that the inclusion of smoking and body weight as co-variates in our statistical model would have affected our risk estimates, but these factors are unlikely to explain the positive association between CD and thromboembolism. Use of oral contraceptives or hormone-replacement therapy is a risk factor for thromboembolism (Poulter & Meirik, 1995) but is not associated with CD (West et al, 2003; Miehsler et al, 2004). The IPR contains no data on ethnicity. In a recent study of Swedish pregnant women with CD, origin outside the Nordic countries was more common among the 2 million reference individuals than among the 2000 women with CD (Ludvigsson et al, 2005b). Considering that white race is the least associated with an increased risk of VTE (Tsai et al, 2002), ethnicity is unlikely to explain our findings. Finally, heterogeneity testing found that presence of cancer did not influence the risk of VTE. Cancer is often discovered after a diagnosis of VTE (Baron et al, 1998), and is slightly more common in CD (Askling et al, 2002); but cannot explain the positive association between CD and VTE.
The IPR contains no data on investigative procedures, such as small-bowel biopsies, in CD or, for example, Doppler ultrasonography, phlebography, angiography, ventilation-perfusion scintigraphy or spiral computer tomography in VTE. A recent study in Northern Sweden found that VTE is however seldom diagnosed without these investigations (Petrauskiene et al, 2005). It is also important to note that the risk estimates for VTE was highest in the last calendar period (1994–2003) when diagnostic specificity can be assumed to be higher than in the early part of the study period.
In Sweden, it is considered mandatory to perform a small-bowel biopsy prior to diagnosis of CD (in the study by Ivarsson et al (2002), 93% of individuals with ‘suspected CD’ had a positive biopsy); and a recent validation of the diagnosis of CD among patients with lymphoma found that 85% of individuals with CD diagnosis in the IPR actually had CD (Smedby et al, 2005). A large proportion of individuals with VTE will be cared for in hospital. The incidence of VTE was 65/100 000 person-years in our reference population compared with some 78/100 000 person-years in another Swedish study (Petrauskiene et al, 2005) and some 60/100 000 person-years among Americans without Protein S, Protein C and Antithrombin deficiency (Brouwer et al, 2006).
In contrast, CD does not necessitate hospital admission and this study may (i) have a low sensitivity for CD and (ii) include CD individuals with more severe disease than the average patient. However, in the first half of the study period, patients with CD were commonly admitted to hospital as part of the gastrointestinal investigation and only then transferred to an outpatient setting for further management. Even today, some individuals who undergo endoscopy with general anaesthesia are cared for as inpatients. Despite this, we cannot rule out the possibility that individuals with CD in this study had more severe CD than the average patient with CD. Considering that a study from northern Sweden found that only 1/1000 individuals had a diagnosis of CD (prior to screening), we believe that a sizeable proportion of individuals with diagnosed CD were identified in this study (Sweden has some 9 million inhabitants). In this study, any false-negative individuals with CD (classified as reference individuals) are unlikely to affect the risk estimates, as CD does not occur in more than 1% of the population (Dube et al, 2005).
In conclusion, we found a modest association between CD and VTE. This may be explained by a combination of surveillance bias and chronic inflammation in some individuals with CD.
JFL was supported by a grant from the Örebro University Hospital while writing this article. This project was supported by a grant from The Swedish Society of Medicine, the Swedish Research Council, the Sven Jerring Foundation, the Örebro Society of Medicine, the Karolinska Institutet, the Clas Groschinsky Foundation, the Juhlin Foundation, the Stiftelsen Samariten, the Majblomman Foundation and the Swedish Coeliac Society.
This project (04-030/1) was approved by the Research Ethics Committee of the Karolinska Institute, Stockholm, Sweden on the 18th March 2004.
JFL, SMM and AE designed the study. JFL wrote the manuscript and performed the statistical analyses. AW, SMM, AE and RL reviewed the manuscript for important intellectual content. All the authors read the final version of the manuscript and approved it.
Table I. International classification codes used in the current paper.
*The Swedish inpatient register does not distinguish between type 1 and type 2 diabetes mellitus.
†Venous thromboembolism (VTE) because of pregnancy, abortion, or childbirth has not been included. We did not include VTE in corpus cavernosum or in the fetal cord. ICD, International Classification of Diseases.