Are we not over-estimating the prevalence of coeliac disease in the general population?

Abstract

Until the 1980s, coeliac disease was considered to be a rare disease, but in the 1990s it became clear that it was a frequent condition. Recently, it was suggested to affect 1 out of 100 subjects in the Western world. To understand what the true prevalence of coeliac disease is in the general population, we conducted a systematic review of published papers. The overall prevalence of coeliac disease in the general population appears to be around 1/160 (6.2‰), but this figure varies widely according to the diagnostic criteria used in the original papers. Prevalence obtained with tissue transglutaminase antibodies only was markedly higher than that obtained through a histological diagnosis. We conclude that the prevalence of coeliac disease in the general population has been over-estimated. This is mainly due to tissue transglutaminase antibodies being used as the only diagnostic tool.

Key words::

• Coeliac disease
• endomysial antibodies
• epidemiology
• tissue transglutaminase antibodies
• villous atrophy

Key messages

• We conducted a review of published papers to understand the true prevalence of coeliac disease in the general population.

• The overall prevalence of coeliac disease in the general population appears to be high (1/160–6.2‰) but not as high as suggested in the most recent papers.

• The prevalence varies widely according to the diagnostic criteria used in the papers.

Introduction

In the last 20 years our knowledge on the prevalence of coeliac disease (CD) has dramatically changed (1). Until the 1980s, CD was considered to be a rare disease, but in the 1990s, thanks to ‘coeliac antibodies’, it became clear that CD was a frequent condition, and in 1995 Fasano suggested that the prevalence of CD in the general population could be around 1/250 (2). This prevalence has progressively increased and, recently, it was proposed that it might be higher than 1/100 (3,4).

Although it is obvious that this rise in the prevalence of CD is mainly due to the use of coeliac antibodies, which have allowed only those subjects found to be ‘serologically’ positive to be referred for duodenal biopsies (3), the astonishing rise in the latest published papers deserves some considerations. A possible explanation is that the prevalence of CD has actually increased (4). On the other hand, we feel that other factors could have a role.

Several studies addressing the prevalence of CD in the general population have been published over the last 20 years. To understand what the true prevalence of CD is in the general population, we decided to conduct a systematic review. We also investigated whether there are any differences in prevalence when different types of populations, sample sizes, years of publications, geographic regions, and diagnostic criteria are taken into account.

Methods

A literature search was started by entering the following query on http://www.ncbi.nlm.nih.gov/ pubmed/advanced: (“1990”[Publication Date]: “3000” [Publication Date]) AND “Celiac Disease/epidemiology” [MAJR] AND English [Language]. This search found 519 papers. Studies performed with small sample sizes (< 400) and studies set up in primary care or endoscopy units were not taken into account. Similarly to Dubé et al., duplicate and triple publications were eliminated (1). Finally, we found 40 papers focusing on the prevalence of CD in the general population (4–43), plus the paper published in the present issue (44).

We rapidly realized that in all these studies types of populations, diagnostic criteria, and sample sizes differed so widely that a proper meta-analysis could not be performed. Nevertheless, the papers were divided and regrouped according to their different characteristics (Table I). It should be noted that we also classified all the studies according to the diagnostic criteria used in the original papers. Although the diagnosis of CD should be based on coeliac antibodies followed by duodenal biopsy, we found that this strategy had been used in only 32 papers. In nine papers endomysial antibodies (EMA) were used to make the diagnosis, and in the last four papers positive tissue transglutaminase antibodies (TTA) were the only criterion to support the diagnosis of CD. It should be noted that in five histology-based studies, only some of the subjects found to be positive to either EMA or TTA had actually undergone a duodenal biopsy to confirm the diagnosis (14,29,34,41,42).

Table I. Different categories in which the forty papers (4–43) were classified.

	Subgroups	References
Type of population	Adults	4,5,7–10,16,19,22,23,27,28,30,39,41
	Children	6,14,17,21,24,25,28,29,32–34,40,42
	Blood donors	11–13,15,18,20,26,31,35–38,43
Sample size	≤1999	5,8,10,13,16–19,24–26,33,35,38,43
	2000–3999	4,9,11,15,20–23,28,29,31,32,36,37,39,41,42
	≥4000	6,7,12,14,27,30,34,40
Year of publication	≤1999	5–18
	2000–2004	19–32
	≥2005	4,33–43
Geographic region	Northern Europe	4,5,10,15,16,18,24,29,32,43
	Central Europe	13,14,17,25,26,42
	Southern Europe	6,7,9,12,19,21,23,39
	Anglo-Saxons country	8,11,27,30,41
	Latin America	20,22,28,35,36
	Middle East and India	31,33,34,37,38,40
Diagnostic criteria used in the original papers	Duodenal biopsy	4,6–10,12–25,28,29,32–34,36,38,43
	EMA	5,11,14,26,27,29,30,37,42
	TTA	31,34,35,41

EMA = endomysial antibodies; TTA = tissue transglutaminase antibodies.

The prevalence of the disease was computed as the number of affected subjects over the number of subjects studied in each article, together with its Poisson exact 95% confidence intervals. The numbers of affected subjects were compared between different categories of patients by means of the negative binomial regression, using the number of studied subjects as exposure variable. Stata 11 (StataCorp, College Station, TX, USA) was used for computation.

In order to provide the most objective evaluation, we performed the analyses with three different sets of data. The first one included the prevalence of CD shown in the 40 papers (4–43); as far as references 14, 29, 34, 41, and 42 are concerned, the serological results were included. On the other hand, the histological results of references 14, 29, 34, 41, and 42 were used for the second analysis. Finally, in the third one, these five papers were excluded, and the analysis was performed without them. Theoretically, the first analysis should provide an over-estimation of the prevalence of CD in the general population, while the second one should give an under- estimation. The third one should be the most objective, but it suffers from a reduction of the sample size.

Results

We were surprised to see that a prevalence of CD higher than 1/100 was suggested in only four studies performed with a histological diagnosis (4,10,17,24). On the other hand, seven biopsy-based papers suggested a prevalence lower than 1/400 (7,9,12,20,28,38,39). Although claiming a prevalence of 1/100 is certainly more appealing than one of 1/400, we cannot see any scientific reason for preferring either.

Table II shows that the overall prevalence of CD in the Caucasian general population is around 1/160 (6.2‰) with a very narrow 95% confidence interval. Although there were no differences as far as types of population, sample size, and year of publication were concerned (data not shown), Table III shows that the prevalence of CD varied widely according to the diagnostic criteria used in the original papers. Post-hoc comparisons showed that the prevalence obtained with TTA was significantly higher than that obtained using histology or EMA.

Table II. Overall prevalence of coeliac disease. Serological results of references 14, 29, 34, 41, and 42 were included in analysis 1. Histological results of references 14, 29, 34, 41, and 42 were used for analysis 2. Analysis 3 was performed without any of the results of references 14, 29, 34, 41, and 42.

	Analysis 1	Analysis 2	Analysis 3
Prevalence ‰ (1/n)	6.9 (1/144)	5.8 (1/172)	6.0 (1/166)
95% CI ‰ (1/n)	6.4–7.4 (1/135–1/156)	5.3–6.9 (1/161–1/188)	5.5–6.6 (1/151–1/181)

CI = confidence interval.

Table III. Prevalence of coeliac disease according to the diagnostic criteria used in the original papers. Analysis 1 was performed using the serological results of references 14, 29, 34, 41, and 42. Analysis 2 was performed using the histological results of references 14, 29, 34, 41, and 42. Analysis 3 was performed without any of the results of references 14, 29, 34, 41, and 42.

	Analysis 1	Analysis 2	Analysis 3
Histology (‰) (95% CI ‰)	5.2 (4.6–5.8)	5.1 (4.6–5.6)	5.2 (4.6–5.8)
EMA (‰) (95% CI ‰)	9.0 (8.0–10.1)	7.3 (6.1–8.6)	7.3 (6.1–8.6)
TTA (‰) (95% CI ‰)	11.3 (9.4–13.6)	16.6 (12.3–21.9)	16.6 (12.3–21.9)
Overall model P value	0.018	0.010	0.018
Post-hoc comparisons P value:
Histology versus TTA	0.007^a	0.006^a	0.010^a
EMA versus TTA	0.088	0.014^a	0.018
Histology versus EMA	0.265	0.979	0.892

CI = confidence interval; EMA = endomysial antibodies; TTA = tissue transglutaminase antibodies.

^aSignificant at the 5% level after Bonferroni correction.

Although the prevalence of CD was different in different geographic regions (for example, 1/110 in Northern Europe and Anglo-Saxon countries, 1/285 in Southern Europe), the comparison was far from statistical significance (data not shown).

Discussion

On the basis of our review, it emerges that in the last few years the prevalence of CD in the general population has been over-estimated and that this was mainly due to the use (or should we write ‘abuse’?) of TTA as the only diagnostic tool.

TTA are a fantastic serologic help in the diagnosis of CD. However, unlike EMA, their specificity is not 100%. This is a crucial point when we use them to study populations at low risk of CD. In a population with a disease prevalence of 1/100, a test with a sensitivity of 100% but a specificity of ‘only’ 99% will give 50% of false positive results. We must not forget that (45).

Another explanation for the over-estimation of the prevalence of CD in the general population is related to the fact that the most recent papers did not only include new patients with positive coeliac antibodies and frank villous atrophy. We found that patients already known to be affected by CD had been considered in the final count. Again, patients with minimal intestinal lesions and/or negative coeliac antibodies were always considered to be affected by CD. Although these patients certainly exist, in our clinical experience they are very rare and most of the time they are not affected by CD but are the result of diagnostic mistakes (46–48).

We could not find any significant difference between adults, children, and blood donors. This finding is very difficult to explain. First of all we cannot forget that CD was considered an almost exclusively paediatric disease until the mid 1980s and that we ourselves showed a lower CD prevalence as age increases (9). Then, blood donors are accurately selected. Anaemia or insufficient body weight are universally recognized as conditions precluding enrolment as blood donors. So they are certainly ‘super-healthy’, and the prevalence of CD among blood donors must be lower than in the adult general population. We feel that we did not find this difference because of the wide heterogeneity of the published papers which hampered a specific analysis focusing on this point.

Very interestingly, we could not find any difference between different geographic regions. It is possible that this is a true finding due to the fact that the amount of gluten needed to trigger CD is very small (49,50). So it will inevitably be ingested whatever the ‘cultural’ gluten consumption is. On the other hand, we cannot exclude the possibility that the wide heterogeneity of the published papers did not make it possible to highlight any relevant difference.

We conclude that the prevalence of CD in the general population is not as high as has recently been claimed. At least in some cases, the over-estimation was due not only to the use of TTA as the sole diagnostic tool but also to a debatable interpretation of both experimental results and already published data. According to our analysis, the prevalence of CD is ∼1/160, it seems to have been stable over the last 20 years, it is the same in both children and adults, and it does not vary in different geographical regions.

Acknowledgements

We are grateful to Susan West for reading and correcting the manuscript.

Declaration of interest: The authors state no conflict of interest and have received no payment in preparation of this manuscript.