Sepsis is an infectious disease characterised by a systemic inflammatory response with tachycardia, fever, hyperventilation and affected white blood cell count.

Coeliac disease is an immune-mediated disorder found in 1% of the Western population.1 It is triggered by gluten exposure and treatment therefore consists of a life-long gluten-free diet. Although inflammation and villous atrophy in the small intestine is a prerequisite for diagnosis of the disease, coeliac disease also has extra-intestinal complications,2^–4 including lymphomas5 6 and infectious disease.7 8

Several factors may predispose to an increased tendency of severe infection in coeliac disease. Individuals with coeliac disease, especially those with other autoimmune disease, are at increased risk of hyposplenism.9^–12 This could potentially increase the risk of pneumococcal disease in coeliac disease.11 12 Individuals with coeliac disease also show an increased mucosal permeability,13^–15 a trait that has been linked to multiple organ dysfunction syndrome in critically ill patients16 and thus an increased risk of sepsis.

With the exception of one mortality study (showing a 7-fold increased risk of death from sepsis4 among individuals with coeliac disease), we know of only case reports of severe infection in coeliac disease.17^–19 In the mortality study, data on sepsis were limited to information from death certificates.4 That investigation also failed to consider the underlying aetiology of sepsis. For these reasons, we assessed the risk of sepsis in individuals with coeliac disease by using Swedish national inpatient register data.

METHODS

In this cohort study, we used the Swedish National Inpatient Register (IPR) to identify all individuals with an inpatient diagnosis of coeliac disease during the period 1964–2003. Coeliac disease, and the potential confounder diabetes mellitus (DM), as well as sepsis (our outcome measure) were defined according to the relevant international classification of disease (ICD) codes (see appendix 2). Subtypes of sepsis (according to the causative micro-organism) were identified through the IPR (see appendix 1). ICD diagnoses were not verified using patient charts.

Of 15 533 individuals with coeliac disease identified through the IPR, we then excluded 94 because of data irregularities (such as date of death preceding the date of diagnosis of coeliac disease. Statistics Sweden (the government agency for population statistics) then used the National Total Population Register20 to select up to five reference individuals for each patient with coeliac disease matched for age, sex, calendar year and county (table 1). These reference individuals constituted the comparison group in several analyses (table 1), but in the main analyses we restricted the reference population to inpatients in order to minimise the risk of ascertainment bias. Throughout the paper, we refer to these as “inpatient reference individuals”, while the more general term “reference individuals” is used to identify the comparison group drawn from the general population (which includes the inpatients). In all analyses, individuals with sepsis prior to study entry (and diagnosis of coeliac disease) were excluded (a general overview of the exclusion procedure has been published elsewhere21). The numbers of participants in each analysis are listed in table 1.

Statistical methods and analyses

Hazard ratios (HRs) for subsequent sepsis were estimated using an internally stratified Cox regression. This analysis resembles a conditional logistic regression as individuals with coeliac disease are compared with their matched reference subjects (the internal strata or risk-sets).

In the main analyses, the follow-up time started at study entry and ended on the date of first discharge diagnosis of sepsis, date of emigration, death or the end of the study period (31 December 2003), whichever occurred first. In separate analyses we stratified by sex and age at diagnosis. The statistical significance of difference in risk estimates between strata was tested with a formal interaction test. We also adjusted for DM and socioeconomic index (SEI,22 see also table 1) as DM is a potential confounder3 24 and SEI may affect health-seeking behaviour.25

To investigate the potential contribution of hyposplenism to the increased risk of sepsis in patients with coeliac disease, we specifically examined the risk of pneumococcal sepsis in coeliac disease (the IPR contains no data on spleen size so we were unable to use spleen size as a measure of hyposplenism). We also examined the risk of sepsis due to Staphylococci, Gram-negative bacteria, any Streptococci (not specified as Pneumococci), and Meningococci.

To increase the specificity of our definition of sepsis, we studied: (1) sepsis diagnosed in a department of infectious diseases; and (2) sepsis listed as the primary diagnosis (not secondary diagnosis). Both these restrictions are likely to increase the specificity of sepsis.23 We also examined the risk of having at least two diagnoses of sepsis. In these analyses, and the analyses described below, we used reference individuals from the general population.

In a recent study, Gedeborg et al23 validated the sensitivity and specificity of sepsis diagnoses among Swedish intensive care patients. In a post hoc analysis we estimated the risk of having sepsis according to the wide ICD-9-10 criteria defined by that study23 (see appendix 2) in a department of medicine (the primary diagnosis of sepsis is medical and not surgical). In this analysis we restricted our data to adults entering the study from 1986 and onwards since children with sepsis are cared for in departments of paediatrics. The specificity of sepsis in the IPR (wide criteria in an intensive care unit, see appendix 2) was according to Gedeborg et al23 97.5% for ICD-9 diagnoses and 92.6% for ICD-10 diagnoses (with sensitivity at a lower level: 45.7 and 52.5%).

In a final analysis we used a case–control design, where cases consisted of all individuals with a diagnosis of coeliac disease in the IPR. Sepsis was our exposure variable. Through conditional logistic regression we then estimated the risk of coeliac disease in individuals with prior sepsis. Those with 1 year or less between the dates of sepsis and study entry (equivalent to date of diagnosis of coeliac disease) were excluded. Risk estimates are given as odds ratios (ORs).

Statistical significance was defined as 95% confidence intervals for risk estimates not including 1.0. Statistics were calculated using SPSS 11.0.

None of the participants was contacted. Patient information was made anonymous prior to the analyses.

RESULTS

About 60% of the participants were female (table 2). The median age at study entry was 2 years (range, 0–94; mean, 19.7) in individuals with coeliac disease and 3 years in reference individuals (range, 0–94; mean, 20.2). Although the majority of study participants entered the study in childhood, most were adults by the end of follow-up (table 2). The median age at diagnosis of first sepsis was 62 years (range, 2–93; mean, 55) in individuals with coeliac disease, and 74 years (range, 1–97; mean, 62) in reference individuals. The median duration between study entry and first diagnosis of sepsis was 6 years (range, 1–31) in individuals with coeliac disease and 10 years (range, 1–39) in reference individuals. A total of 207 individuals with coeliac disease and 493 reference subjects had a subsequent diagnosis of sepsis. Forty-seven individuals with coeliac disease (22.7%) and 94 reference individuals (19.1%) died within 28 days of their first diagnosis of sepsis. Of the 14 250 individuals with coeliac disease and follow-up for >1 year, 10 291 (72.3%) had coeliac disease listed as the main diagnosis at study entry (childhood coeliac disease, 7893/9334  =  84.6%; adulthood coeliac disease, 2398/4916  =  48.8%).

Comparisons with inpatient reference individuals

We found a positive association between coeliac disease and subsequent sepsis (HR  = 1.6, 95% CI  = 1.2 to 1.9, p<0.001, based on subsequent sepsis in 221 reference individuals and 285 individuals with coeliac disease), but this risk increase was restricted to coeliac disease diagnosed in adulthood (HR  = 1.5, 95% CI  = 1.1 to 2.1, p = 0.006) (childhood coeliac disease: HR  = 1.0, 95% CI  = 0.6 to 1.9, p = 0.908) (table 3). There was no difference in risk estimates for subsequent sepsis between males and females with coeliac disease or between individuals with coeliac disease diagnosed in childhood as opposed to in adulthood. Adjustment for SEI in a subset of patients with available SEI data did not influence the risk estimates indicating that SEI is not a confounder (data not shown). Although the risk estimates for sepsis decreased with adjustment for DM, the statistically significant association between coeliac disease and subsequent sepsis remained (HR  = 1.3, 95% CI  = 1.1 to 1.7, p = 0.001).

In separate analyses we evaluated the association between coeliac disease and subsequent sepsis due to specific bacteria (table 4). coeliac disease was associated with sepsis due to Pneumococci (HR  = 2.5, 1.2 to 5.1, p = 0.014) and Staphylococci (1.9, 1.2 to 3.3, p = 0.013), but not due to Gram-negative bacteria (1.4, 0.9 to 2.2), all Streptococci (not specified as Pneumococci) (1.1, 0.6 to 2.1), and Meningococci (0.4, 0.1 to 3.6) (table 4).

Comparisons with reference individuals from the general population

Coeliac disease was positively associated with subsequent sepsis (HR  = 2.6, 95% CI  = 2.1 to 3.0; table 3). This risk increase was seen in both males and females, as well as in children and adults (table 3). A formal interaction test did not indicate that risk estimates differed by age at coeliac disease diagnosis (p = 0.200). When we stratified individuals by age (five groups), we found an increased risk of sepsis in all age bands but in the oldest participant quintile (aged ⩾74 years at coeliac disease diagnosis) where the HR for sepsis did not attain statistical significance (HR  = 1.4, 95% CI  = 0.8 to 2.2). The HR for sepsis in individuals diagnosed with coeliac disease before the age of 3 years was 1.9 (95% CI  = 1.2 to 3.0).

Adjustment for socioeconomic status in a subset of individuals with data on SEI showed that the increased risk of sepsis in coeliac disease could not be explained by SEI (crude HR  = 3.1; HR adjusted for SEI  = 3.1). HRs for sepsis did not vary notably over time (calendar periods): 1964–1973, HR  = 2.2; 1974–1983, 2.0; 1984–1993, 2.9; and 1994–2003, 2.9. Exclusion of individuals with a diagnosis of DM did not influence the risk estimate (HR  = 2.6, 95% CI 2.1 to 3.2, based on 172 diagnoses of sepsis in subjects with coeliac disease and 379 among the reference group). When we excluded the first 5 years of follow-up there was a statistically significantly increased risk of sepsis in subjects with coeliac disease (HR  = 2.2, 95% CI  = 1.8 to 2.8). When we included the first year of follow-up, the HR for sepsis in subjects with coeliac disease was 3.2 (95% CI  = 2.7 to 3.7, based on 285 individuals with coeliac disease and sepsis versus 613 reference individuals with subsequent sepsis). Restricting our outcome measure to sepsis diagnosed in adults at the department of infectious diseases (HR  = 3.0, 95% CI  = 2.1 to 4.1), or to sepsis as defined by Gedeborg et al23 (HR  = 2.5, 95% CI  = 1.7 to 3.7) we found similar risk estimates. Coeliac disease was also associated with having a subsequent diagnosis of sepsis listed as the primary diagnosis (HR  = 2.6, 95% CI  = 2.1 to 3.2). The HR for having at least two diagnoses of sepsis was 3.0 (95% CI  = 2.1 to 4.3).

Table 5 presents data on the incidence of sepsis. This table shows consistently increased incidence rates for sepsis in subjects with coeliac disease in all strata.

A more pronounced association of coeliac disease was identified with subsequent sepsis due to Pneumococci (HR  = 3.9, 95% CI  = 2.2 to 7.0; table 4). The risk increase for pneumococcal sepsis was seen in both children and adults with coeliac disease (table 4).

Conditional logistic regression found an increased risk of subsequent coeliac disease in individuals with prior sepsis (OR  = 2.2, 95% CI  = 1.7 to 3.0, p<0.001).

DISCUSSION

This study found a positive association between coeliac disease and subsequent sepsis. This risk increase cannot be explained by potential confounding factors such as concomitant DM or socioeconomic factors. Our findings are therefore consistent with earlier case reports of coeliac disease and severe infection17^–19 as well as with the increased risk of death from sepsis reported by Peters et al4 (standardised mortality ratio  = 7.1, 95% CI  = 1.9 to 18.2). Using inpatient reference individuals, we found a 1.6-fold increased risk of subsequent sepsis in coeliac disease. When compared with reference individuals from the general population, the risk increase was even higher (HR  = 2.6).

Suggested mechanisms of action

It has been advocated that individuals with coeliac disease should be given prophylaxis or vaccinations against Pneumococci.11 26 The underlying idea is that hyposplenism is common in coeliac disease,10^–12 27 and coeliac disease may even be the most common cause of non-surgical asplenia in the Western world.28 In a recent Italian study,11 the authors counted pitted red cells (erythrocytes with membrane abnormalities) as a measure of hyposplenism and found higher average pitted red cell counts in coeliac disease patients than in patients with other autoimmune disease or in healthy volunteers.11 What is also interesting about the study by Di Sabatino et al11 is that the prevalence of hyposplenism in those groups of individuals with coeliac disease at greatest risk of hyposplenism (coeliac disease and complications or coeliac disease and other autoimmune disorder) was not related to the duration of gluten-free diet. Thus it may be assumed that hyposplenism in many individuals with coeliac disease persists after diagnosis of coeliac disease. Hyposplenism is associated with fewer IgM memory B cells; these cells are important in the defence against encapsulated bacteria.29 30 We suggest that hyposplenism in coeliac disease may explain the increased risk of pneumococcal sepsis in coeliac disease. A limitation in this context is, however, that the current study contained no data on spleen size. Such data might otherwise have helped us explain the association between coeliac disease and sepsis.

The lack of positive relationship between coeliac disease and meningococcal sepsis may be due to low power in the analysis since very few patients received a diagnosis of meningococcal sepsis in our study or because the association between coeliac disease and sepsis operates specifically though impaired immunity against Pneumococci. Another explanation might be that hyposplenism due to coeliac disease takes years to develop and when it has, in adult age, meningococcal infections are uncommon. Duration of untreated coeliac disease with concomitant exposure to gluten, may also explain that the highest HRs for sepsis were seen in coeliac disease diagnosed in adulthood. Only in coeliac disease diagnosed in adulthood did we see an increased risk of sepsis when we compared with inpatient reference individuals. This is consistent with earlier findings31 that long-term exposure to gluten in patients with coeliac disease is linked to increased co-morbidity.

There was evidence that those with coeliac disease were at increased risk also of other bacterial forms of sepsis; so it is possible that mechanisms other than hyposplenism may contribute to the increased risk of sepsis in coeliac disease. Individuals with coeliac disease have increased mucosal permeability.14^–16 In 2004, Forsberg et al32 reported that coeliac disease is linked to an increased prevalence of bacteria attached to the intestinal epithelial cells, and that this might be due an altered composition of the intestinal glycocalyx. Different patterns of short-chain fatty acids in active and treated coeliac disease as compared with that in controls, also suggest a difference in the metabolic activity of the intestinal flora in those with coeliac disease.33

Finally, some patients may have suffered from malnutrition that, per se, may increase the risk of sepsis. All these mechanisms may contribute to the increased risk of sepsis in people with coeliac disease. However, as most of the risk estimates were reduced when the comparison group was limited to inpatients this could account for some of the associations, particularly where sepsis was not due to Pneumococcal infection.

Strengths and weaknesses

This is one of the largest studies of coeliac disease and sepsis. The use of an internally stratified Cox regression means that potential confounding due to sex, age and calendar year was eliminated. This is of major importance since these factors influence the incidence of coeliac disease34 and sepsis.35 However, we cannot rule out the possibility that unidentified confounding factors have influenced our data. Individuals with coeliac disease are at increased risk of cancer.36 Cancer treatment often consists of chemotherapy administered through central venous catheters that, per se, increase the risk of sepsis. Some individuals with coeliac disease have co-existent autoimmune liver disease,37 while others have refractory sprue.5 Such patients are often given immunosuppressive treatment that could increase the risk of sepsis. Unfortunately, the IPR does not contain data on drug treatment or the use of central venous catheters. This study benefited from the use of two reference groups: one from the general population and one from among inpatients. Using the general population as a comparison, it is possible that studies such as this can produce somewhat inflated estimates of association due to selection and surveillance bias making it more likely that diagnoses are recorded among inpatients. In contrast, the use of an inpatient comparison group reduces the influence of surveillance bias, but may produce a somewhat conservative estimate of association between some diagnoses.

The characteristics of sepsis among our reference individuals from the general population are similar to those found in Americans with sepsis.35 In our study, the sepsis incidence was 148/100 000 (years 1964–2003), compared with 83/100 000 (in 1979–1984) and 240/100 000 (in 1995–2000) among Americans35). Likewise, the mean age at first diagnosis of sepsis among the reference population was 62 years in this study, compared with 57–61 years (at any sepsis diagnosis) in the US.35 The 28-day death rate was 19% in our study as compared with 28%–19% (during hospital admission) in the American study.35 Also, the relationship between Gram-positive and Gram-negative bacteria was similar in the two studies (our study, ratio  = 1.2; Martin et al,35 ratio  = 1.4).

A potential weakness of this study is that of misclassification. The misclassification rate of sepsis, however, is low,23 and misclassification will only affect risk estimates if it is differential by coeliac disease status. We find such bias unlikely. We calculated the risk of sepsis according to the criteria of Gedeborg et al23 (with sepsis specificity levels above 90%), and in the analysis using these criteria there was a more than 2-fold increased risk of sepsis associated with coeliac disease. The requirement of a small intestinal biopsy prior to the diagnosis of coeliac disease increases the specificity and the likelihood that a diagnosis of coeliac disease represents true coeliac disease. Smedby et al38 recently found that the inpatient diagnosis of coeliac disease was correct in 85% of patients with concomitant lymphoma. Unfortunately, we have no data on endoscopy in individuals with coeliac disease. Such data could otherwise have been used to increase the specificity of coeliac disease diagnoses. Neither do we have any data on the proportion of adults or children who underwent endoscopic examination as inpatients, now or during the early part of the study period. There is evidence to indicate that our study has identified a large proportion of those with diagnosis of coeliac disease in Sweden. An earlier study suggests that 1/1000 individuals in Sweden had a diagnosis of coeliac disease in the late 1990s,39 and this figure is consistent with the number identified by our study.

Although diagnostic criteria for both coeliac disease and sepsis have changed since 1964, we found similar risk estimates over time with the highest risk estimates in the second half of our study period when the specificity can be assumed to be higher than in the early half of the study period. However, we lack data on the presence of tachycardia, fever, hyperventilation and leukocyte counts; and these might have been used to further verify inpatient diagnoses of sepsis.

The sensitivity of both coeliac disease and sepsis diagnoses may be low in this study, and only some 50% of those with signs of sepsis in Swedish intensive care units in the study by Gedeborg et al23 received an inpatient diagnosis of sepsis. Instead these patients may have received diagnoses such as meningitis or pneumonia. There is no reason to suspect that differential bias has been introduced because sensitivity for diagnosis of one disease varies by the presence of the second. Low sensitivity is likely to have produced more conservative estimates of risk rather than creating spurious associations.

In conclusion, our study found a modestly increased risk of sepsis in coeliac disease. The highest risk estimates were seen for pneumococcal sepsis. This may be due to hyposplenism in coeliac disease.