Approximately 1% of the Western population suffers from celiac disease (CD). On exposure to gluten, these individuals develop an immune-mediated response in the small intestine characterized by inflammation, crypt hyperplasia, and villous atrophy (VA).

Clinical Perspective on p490

One in 5 deaths in the United States is caused by ischemic heart disease (IHD),¹ and ≈500 000 people die of IHD each year in the United States. This makes IHD the leading cause of death for American men and women.¹ Individuals with CD are at increased risk of death,² infectious disease,³ and fractures.⁴ However, research findings on IHD are contradictory. Most^2,5–7 but not all^8–10 studies have shown an increased risk of incident IHD, death resulting from IHD, or cardiovascular disease in CD. These contradictory results may be explained in part by differences in study design, method of data collection, and small sample sizes. The main objective of this study was therefore to examine the risk of IHD in a population-based large cohort of patients with CD.

In recent years, there has been a growing interest in CD with only minor mucosal abnormalities (no VA),¹¹ not fulfilling traditional criteria for CD.¹² Despite this interest, we know of no study on IHD in patients with inflammation without VA or in patients with latent CD (defined by the National Institutes of Health as normal mucosa but positive CD serology).¹³ A second objective was therefore to estimate the risk of IHD in patients with inflammation without VA or latent CD.

Methods

In this cohort study, we estimated the risk of incident IHD (defined as first myocardial infarction [MI] and angina pectoris [AP]) in patients with CD according to small intestinal histopathology and no prior history of IHD. We did so through linkage of nationwide histopathology data with inpatient and mortality data on IHD from Swedish national registers.

Collection of Biopsy Data

From October 2006 through February 2008, we collected data from biopsy reports at Sweden's 28 regional pathology departments.¹⁴ The biopsies had been performed since 1969 (Table 1). Data were restricted to computerized biopsy reports; for this reason, most patients in this study had been biopsied after 1990 (Table 1). Data searches were carried out by local information technology technicians who obtained data on arrival date of the biopsies, personal identity number,¹⁵ morphology based on the Swedish SnoMed classification codes,¹⁴ and topography (duodenum or jejunum).

We graded morphology as VA (Marsh stage 3 and equal to CD), inflammation (Marsh 1 to 2), and normal mucosa (Marsh 0) (for a detailed list of relevant morphology codes, see the online-only Data Supplement). In this study, the CD diagnosis was based on small intestinal morphology and was not conditional on positive CD serology. Small intestinal biopsy is routine clinical practice in Sweden before CD diagnosis,¹⁴ and 95% of individuals with VA have CD.¹⁴ For a diagnosis of inflammation, we required a SnoMed code corresponding to duodenal/jejunal inflammation¹⁴ but never a biopsy with VA. In the most commonly used Swedish histopathology classification (Kvalitets och standardiseringskommittén; see the online-only Data Supplement), inflammation is equivalent to intraepithelial lymphocytosis.

After the exclusion of duplicates and biopsies with data irregularities (eg, biopsy performed before birth or after death of the individual), we had data from 351 403 biopsy reports in 287 586 unique individuals (CD, 29 148; inflammation, 13 446; normal biopsy, 244 992).

To identify individuals with latent CD, we linked data from biopsy reports in a subset of individuals with normal mucosa (n=121 952) with data on CD serology from 8 university hospitals.¹⁶ These hospitals care for individuals from both urban and rural areas corresponding to 49% of the Swedish population. Through this linkage, we obtained data on date of test, type of test (anti-gliadin, anti-endomysial, and anti–tissue transglutaminase), antibody levels (immunoglobulin A/immunoglobulin G), and age-specific reference values in all individuals with normal mucosa.¹⁶ In this study, we defined latent CD as a normal mucosa with positive serology for CD ≤180 days before biopsy and ≤30 days after biopsy. We included only those individuals who never had a biopsy with VA or inflammation. When these criteria were used, the number of individuals with latent CD was 3736¹⁶; 351 of these individuals had positive immunoglobulin A endomysial or tissue transglutaminase antibodies.

Records from individuals with CD, inflammation, or latent CD (total n=46 330) were then sent for matching by the government agency Statistics Sweden. Personal identity numbers were replaced with serial numbers to guarantee the anonymity of the participants. After exclusion of 1884 individuals undergoing biopsy for various reasons (the Figure), 44 446 individuals (CD, n=28 190; inflammation, n=12 598; and latent CD, n=3658) remained.

• Download figure
• Download PowerPoint

Reference Individuals

For each individual with CD, inflammation, or latent CD, Statistics Sweden identified up to 5 unexposed reference individuals matched on age, sex, county, and calendar year from the Total Population Register. Reference individuals were sampled from all Swedish residents for whom the 28 regional pathology registers had not indicated prior duodenal/jejunal biopsy. In total, Statistics Sweden identified 229 800 reference individuals. The Figure delineates the reasons for exclusion of 10 408 reference individuals. The final analyses included 219 392 reference individuals. Some of the reference individuals may have undergone testing for CD and tested negative.

Outcome Measure

In the main analyses, we used a composite measure for IHD based on MI (incident disease and death caused by MI) and AP (incident AP). To maximize the validity of the diagnoses, only main causes of hospitalization or death were considered. The following International Classification of Diseases (ICD) codes defined MI: ICD-8, 410; ICD-9, 410; and ICD-10, I21 through I22. The following defined AP: ICD-8, 411; ICD-9, 411B, 413, and 414A; and ICD-10, I20. In a separate analysis, we evaluated unstable AP (ICD-8, 411; ICD-9, 411B; and ICD-10, I20.0). Data on incident disease were obtained from the Swedish national Hospital Discharge Register. Data on death from MI or AP were obtained from the national Cause of Death Register.¹⁷

Other Covariates

Through the Hospital Discharge Register, we identified individuals with type 1 diabetes mellitus. Earlier versions of the Swedish ICD (versions 7 through 9) did not distinguish between type 1 and type 2 diabetes mellitus. In this study, type 1 diabetes mellitus was defined as a relevant ICD-8 to ICD-10 code (ICD-8, 250; ICD-9, 250; and ICD-10, E10–14) recorded before 30 years of age. We also estimated the risk of IHD adjusted for anemia resulting from iron deficiency, B₁₂ deficiency, or folic acid deficiency according to relevant ICD codes (ICD-8, 280 and 281; ICD-9, 280 and 281; and ICD-10, D50, D51, and D52) and for rheumatoid arthritis in another analysis (ICD-8, 712.3 and 714.93; ICD-9, 714; and ICD-10, M05 and M06). When adjusting for education, we used 7 previously defined educational categories determined by Statistics Sweden.

Statistical Analyses

We used Cox regression, internally stratified for age at first biopsy (and corresponding age in reference individuals), calendar period, sex, and county, to estimate hazard ratios (HRs) for IHD. Follow-up started on the date of first biopsy (CD, inflammation, or normal mucosa in an individual in whom CD serology was positive at the time of biopsy) and the corresponding date in matched reference individuals. Only individuals without a history of IHD at the time of biopsy (or the corresponding date in control subjects) were included in the analyses.

Follow-up ended with a diagnosis of IHD, death caused by IHD, or death resulting from other cause; with emigration; or on December 31, 2007, whichever occurred first. In reference individuals, follow-up could end if the individual had a biopsy with CD, inflammation, or latent CD after the time of matching. The proportional hazards assumption was tested through log-minus-log curves.

In a priori subset analyses, we examined the risk of IHD according to follow-up (<1, 1 to 4.99, and ≥5 years), sex, age (0 to 19, 20 to 39, 40 to 59, and ≥60 years at first biopsy), and calendar period of first biopsy (1989 and before, 1990 to 1999, and 2000 and later). We calculated incidence rates based on the number of events divided by the number of person-years at risk (PYAR). Excess risks were estimated as rate differences between exposed and matched unexposed study participants. The attributable fraction was calculated as (1−1/HR). We calculated separately the HRs for MI and AP. We also estimated the risk of having at least 2 episodes of IHD according to the Hospital Discharge Register and having a diagnosis of unstable AP as opposed to any AP.

In a separate analysis, we adjusted our data for education. Furthermore, we adjusted for type 1 diabetes mellitus, anemia, or rheumatoid arthritis. Finally, we estimated HRs for death from IHD in separate analyses.

Statistical significance was defined as 95% confidence intervals (CIs) for risk estimates not including 1.0. We used SPSS version 16.0 (SPSS Inc, Chicago, IL) for all analyses.

Posthoc Analyses

Given that 2 earlier cohort studies have found a positive association between CD and stroke,⁵,⁸ patients with stroke (and at increased risk of later IHD) may undergo screening for CD. To rule out that the increased risk of IHD in individuals undergoing biopsy was due to prior cerebrovascular disease leading to testing for CD (and then IHD), we also examined the risk of IHD in individuals with no record of ischemic or hemorrhagic stroke before their first biopsy. We also estimated the risk of vascular disease as defined by the Antiplatelet Trialists' Collaboration (nonfatal MI, nonfatal stroke, or death resulting from vascular disease).¹⁸ Finally, we also estimated the risk of IHD in 2 smaller subsets of individuals with limited data on smoking, body mass index, nationality, and medication. A description of these posthoc analyses and their results can be found in the online-only Data Supplement.

This study was approved by the Research Ethics Committee of Karolinska Institutet. The funding sources did not influence the study's design, conduct, and reporting. Dr Ludvigsson 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.

Results

Background Data

The majority of study participants were female (Table 1). The median age at first recorded IHD was 73 years in CD, 74 years in inflammation, and 69 years in latent CD. Other patient characteristics are given inTable 1.

Morbidity From Any IHD by Small Intestinal Histopathology

During follow-up, there were 991 events of IHD in CD, 809 in inflammation, and 62 in latent CD. We found a positive association between CD and IHD that was independent of the histopathological appearance of the small intestine (Table 2). The overall HRs for IHD were 1.19 in CD (95% CI, 1.11 to 1.28), 1.28 in inflammation (95% CI, 1.19 to 1.39), and 1.14 in latent CD (95% CI, 0.87 to 1.50).

The absolute risk (incidence) of IHD was 375, 808, and 256 per 100 000 in individuals with CD, inflammation, or latent CD, respectively. The corresponding excess risks were 60, 179, and 31 per 100 000 person-years. The attributable fraction (ie, the percentage of IHD events in patients undergoing small intestinal biopsy that would not have happened without their small intestinal abnormality) was 16% in CD, 22% in inflammation, and 12% in latent CD. Risk estimates did not change after adjustment for type 1 diabetes mellitus, anemia, rheumatoid arthritis, or education (see the online-only Data Supplement).

The highest risk estimates were seen in the first year of follow-up, in which individuals with CD were at a 1.77-fold increased risk of IHD (95% CI, 1.46 to 2.14). Relative risks were similar in inflammation and latent CD (Table 2). We found no increased risk of IHD in years 1 through 4.99 after diagnosis, followed by a statistically significantly increased risk of IHD ≥5 years after first biopsy (HR, 1.23; 95% CI, 1.12 to 1.36). Excluding the first year of follow-up, CD (HR, 1.12; 95% CI, 1.04 to 1.21) and inflammation (HR, 1.24; 95% CI, 1.14 to 1.35) remained associated with IHD, whereas latent CD was not (HR, 1.04; 95% CI, 0.76 to 1.42).

None of the 13 962 individuals undergoing biopsy in childhood were registered with any IHD during follow-up compared with 4 of the 69 604 reference individuals (Table 3). Excluding individuals biopsied in childhood (thereby restricting our analyses to those diagnosed at 20 years of age and older) had little effect on risk estimates (CD: HR, 1.19; 95% CI, 1.11 to 1.28; inflammation: HR, 1.28; 95% CI, 1.19 to 1.39; and latent CD: HR, 1.14; 95% CI, 0.87 to 1.50). In patients with CD diagnosed in adulthood, the HRs varied only slightly with age (20 to 39 years: HR, 1.16; 95% CI, 0.79 to 1.70; 40 to 59 years: HR, 1.21; 95% CI, 1.08 to 1.37; and ≥60 years: HR, 1.19; 95% CI, 1.08 to 1.30;Table 3), although absolute risks and excess risks increased with age as expected. The positive association between CD and IHD was independent of calendar period (Table 3), although the association did not reach statistical significance in individuals biopsied before 1989.

When IHD was defined as at least 2 hospital discharges listing a relevant diagnosis, individuals with CD, inflammation, and VA were still at a 20% to 30% increased risk that was independent of small intestinal histopathology (see the online-only Data Supplement).

Morbidity From IHD by IHD Subtype

CD was associated with both MI (HR, 1.11; 95% CI, 1.01 to 1.21) and AP (HR, 1.27; 95% CI, 1.15 to 1.40) incidence (Table 4). Inflammation was also associated with an increased risk of MI and AP, whereas latent CD was associated only with AP incidence (Table 4). Restricting our outcome to unstable AP showed positive associations with only CD and inflammation (see the online-only Data Supplement).

Mortality From IHD

During follow-up, there were 271 deaths resulting from IHD in CD, 264 in inflammation, and 8 in latent CD. We found an increased risk of death resulting from IHD in CD (HR, 1.22; 95% CI, 1.06 to 1.40) and inflammation (HR, 1.32; 95% CI, 1.14 to 1.52) but not in latent CD (HR=0.71; 95% CI=0.34 to 1.50). The absolute mortality rate from IHD was 101 per 100 000 PYAR in CD, 255 per 100 000 PYAR in inflammation, and 33 per 100 000 PPYR in latent CD. The excess risk in CD and inflammation was 18 and 62 deaths, respectively, per 100 000, whereas no excess risk was observed in latent CD. Roughly 18% of IHD deaths in individuals with CD could be attributed to the underlying CD. HRs for death from acute MI were almost identical to those mentioned above (see the online-only Data Supplement) as indicated by the finding that all but 21 of the 543 individuals who died of IHD had MI (and not AP) as the underlying cause of death.

Posthoc Analyses

Adjustment for anemia or rheumatoid arthritis did not affect HRs more than marginally (see the online-only Data Supplement). Restricting the analyses to individuals without prior cerebrovascular stroke, we found an increased risk of IHD in all 3 cohorts, although the association was significant only in CD (HR, 1.18; 95% CI, 1.10 to 1.27) and inflammation (HR, 1.28; 95% CI, 1.18 to 1.39) and not in latent CD (HR, 1.13; 95% CI, 0.86 to 1.50).

When we restricted our outcome to nonfatal MI, nonfatal stroke, or death resulting from vascular disease,¹⁸ HRs decreased somewhat but remained statistically significantly increased in CD and inflammation (CD: HR, 1.10; 95% CI, 1.03 to 1.28; inflammation: HR, 1.20; 95% CI, 1.11 to 1.29) but not in latent CD (HR, 1.00; 95% CI, 0.78 to 1.29).

Discussion

This nationwide cohort study found an increased risk of IHD in patients with CD that was independent of small intestinal histopathology. This is the largest study to date on IHD risk in CD, with almost 1000 deaths from IHD during follow-up. It is also the first study to examine IHD in inflammation and latent CD. Although relative risks were modest, IHD may be of concern to the CD community given that cardiovascular disease was the most common cause of death in patients with CD.²

We found a 19% increased risk of IHD in CD, which translated into an absolute risk of 375 events per 100 000. In patients ≥60 years of age at diagnosis, 15% of all IHD events could be attributed to CD. Only in individuals diagnosed with CD in childhood was there no association with IHD; this lack of association is probably due to the short follow-up because IHD events are unlikely to occur in younger age.

Our risk estimates for IHD were lower than those of 2 previous Swedish studies^5,6 (compared with the general population^5,6). Whereas we used biopsy registers to ascertain CD, those authors used the Hospital Discharge Register. The use of hospital registers to identify a condition that does not generally require inpatient care as part of the investigation can lead to the selection of more serious cases with a decreased generalizability of the results. Besides, neither of the 2 Swedish studies examined both incident IHD and death resulting from IHD.^5,6 Two British studies reported an inverse relationship between CD and IHD,^8,9 but none of these associations was statistically significant, and after men and women were combined in the study by Whorwell et al,⁹ CD did not protect against IHD. Wei et al⁷ reported a 2.5-fold increased risk of cardiovascular disease in patients without previous cardiovascular disease, but they based their study on endomysial antibody–positive individuals without histopathology data and did not differentiate between stroke, heart failure, and IHD.⁷

Many patients with CD have been exposed to long-term inflammation in the bowel, both before and after CD diagnosis.¹⁹ In the general population, chronic inflammation is a major risk factor for IHD. Although we lack individual data on dietary adherence for the full study sample, an earlier patient chart review suggested that roughly 17% of our patients with CD have low dietary adherence.¹⁴ Low dietary adherence is associated with persistent inflammation and therefore might explain the increased risk of IHD observed in patients with CD.

We found a positive association between inflammation and IHD, confirming reports of a positive association between nonspecific inflammation/autoimmunity and IHD^20,21 or subclinical signs of IHD.²² The overall risk of IHD in latent CD was 1.14, which can be compared with 1.19 in CD and 1.28 in inflammation without VA. However, in latent CD, there was no association with MI or with death from IHD but only with incident AP (This latter association disappeared when we examined only unstable AP). Latent CD also was not associated with our secondary composite outcome measure (nonfatal MI, nonfatal stroke, or death from vascular disease). Regrettably, we did not have data on ECGs, angiographies, or clinical symptoms to verify the AP diagnosis. Given the absence of an association with MI and unstable AP, we cannot rule out that part of the risk increase for IHD in latent CD is due to surveillance bias or gastrointestinal symptoms misinterpreted as being of cardiac origin (and vice versa), nor can we rule out that the increased risk of IHD seen in this study is due to general factors in gastrointestinal disease rather than CD per se. Bernstein et al²³ reported a 26% increased risk of IHD in inflammatory bowel disease (ulcerative colitis or Crohn disease). The overall aim of this study, however, was to examine the risk of IHD in CD and not in other gastrointestinal diseases (for which we have no data).

This study has several strengths. Ascertaining CD through biopsy registers minimized the risk of selection bias because small intestinal biopsy is routine clinical practice in Sweden.¹⁴ Although we did not require a positive CD serology for diagnosis of CD, 88% of the individuals undergoing patient chart validation had positive CD serology before biopsy.¹⁴ However, we were unable to compare the risk of IHD in those with VA and positive CD serology against those with VA and negative CD serology because CD serology was obtained to identify latent CD, not to define CD. Another strength is the large number of patients, which allowed analyses of patient subsets. We used only primary diagnoses for IHD. In addition, we saw a positive association between CD and IHD when we restricted our outcome to having at least 2 diagnoses with IHD or death resulting from IHD.

One of the main limitations of the study is the lack of data on blood pressure, smoking, body mass index, lipids, exercise, and other established IHD risk factors. We cannot rule out the existence of residual confounding.

In this study, we adjusted for education and type 1 diabetes mellitus, neither of which influenced the risk estimate for IHD. We lacked data on hemoglobin levels. The presence of anemia is otherwise associated with both CD²⁴ and IHD.²⁵ Adjusting for diagnosed anemia, however, had very marginal effects on the HRs for IHD.

We found the highest HRs in the first year after biopsy. This may be due to intense inflammation in patients with CD or inflammation, resulting in medical investigation for CD (and biopsy) and in IHD. It is also possible that psychological stress²⁶ caused by the CD investigation may have contributed to the increased risk of IHD just after biopsy. Another potential explanation is the more frequent follow-up in patients with newly diagnosed CD. This is a study limitation. Surveillance, especially in the first year of follow-up, and consequent diagnosis of IHD may also explain why CD patients were at no increased risk of IHD in years 1 through 5, followed by a 20% increased risk of IHD >5 years after biopsy. Although we lack data on dietary adherence, we speculate that dietary adherence may decrease over time. This could lead to increased inflammation and a higher risk of IHD.

Earlier validation has shown that Swedish pathologists correctly classify 90% to 96% of samples with VA or normal mucosa but sometimes misclassify inflammation.¹⁴ On average, biopsy classification in this study was based on 3 tissue samples.¹⁶ Still, we cannot rule out that some biopsies with inflammation were classified as normal mucosa (latent CD), thereby potentially inflating the risk estimate in analyses of this group. It is also possible that inflammation without VA did not represent an early stage of CD; nevertheless, when 2 independent reviewers manually examined biopsy reports from 1534 individuals, diseases other than CD were rare (1.6% of biopsy reports with inflammation and 0.3% of reports with VA indicated presence of inflammatory bowel disease).¹⁴ However, it should be noted that suspected CD is the major indication of small intestinal biopsy in Sweden, and we expect most individuals with duodenal/jejunal inflammation to have early CD. This view is supported by our findings that patients with inflammation had symptoms similar to those with VA.¹⁴

Conclusions

We found a positive association between CD and IHD. Because cardiovascular disease is the most common cause of death in this group of patients, our findings may have important implications in patients with CD.² Increased awareness of IHD risk factors in this patient group is warranted, and if our findings are corroborated by others, clinical guidelines of treatment of CD may have to include a more detailed cardiovascular risk assessment than what is current practice.

Sources of Funding

Dr Ludvigsson was supported by a grant from Örebro University Hospital while writing this article. This project was supported by a grant from the Swedish Society of Medicine, Swedish Research Council, Sven Jerring Foundation, Örebro Society of Medicine, Karolinska Institutet, Clas Groschinsky Foundation, Juhlin Foundation, Majblomman Foundation, Uppsala-Örebro Regional Research Council, and Swedish Celiac Society. None of the funders had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.

Disclosures

None.

Footnotes