The diagnosis of CD has traditionally depended upon intestinal biopsies and has been extended to include an array of serological markers. The guidelines of the European and North American societies for gastroenterology require a biopsy for diagnosis [13, 14]. Recently, the European Society for Pediatric Gastroenterology, Hepatology, and Nutrition published guidelines allowing the diagnosis of CD without a biopsy in some situations . CD is usually diagnosed when the duodenal and jejunal mucosa display villous atrophy, crypt hyperplasia, and an increase in intraepithelial lymphocytes [16–19]. However, different diseases not related to gluten- sensitive enteropathy can induce a flat mucosa, thus mimicking CD. Moreover, patients with gluten-sensitive enteropathy and normal small bowel mucosal architecture have also been described [20–24]. Most likely because of a lack of technical proficiency with grasping biopsy forceps or endoscopic procedure, biopsy specimens have been shown to be sufficient for diagnosis of CD in only 90% of cases . Furthermore, CD may be missed during histological examinations owing to variations in different pathologists’ assessments . Because of this, and because of the inconvenience and high cost associated with jejunal biopsy and the high prevalence of CD in the general population, less-invasive tests are required. In the last 20 years, serological tests for the diagnosis of CD have improved substantially [27–33]. For practical and ethical reasons, patients with negative serology sometimes did not undergo a biopsy unless clinical indications of CD were evident. This procedure causes a verification bias because the gold standard (histology of the mucosa) is not always available for negative tests . On the other hand, a positive test result demanded a biopsy even when there was only a slight clinical suspicion of CD. Today, it is nearly impossible to overcome this bias for ethical reasons; therefore, the bias may be present in many studies.
The data contained in Table 3, however, indicate that the criteria for choosing the best tests must be defined. For clinicians who want to reduce the number of jejunal interventions in a population with a high frequency of CD, the best test is the one with the lowest sum of false-positive and false-negative diagnoses: the test with the highest ppv, the highest npv, a high likelihood ratio positive, and a low likelihood ratio negative. In our study, a combination of four antibody tests yielded a ppv of 99%, an npv of 100%, an lr+ of 86, and an lr- of 0.00. For practical reasons, we may omit EMA from our combination of antibody tests, and instead chose the test combination of IgG anti-dpgli + IgA anti-dpgli + IgA anti-tTG (Tables 2 and 3), with a ppv of 99%, an npv of 98%, an lr+ of 87, and an lr- of 0.01, as the first step in our diagnostic procedure. Out of 268 patients, 208 (78%) were correctly classified with these serological tests: they had either three tests above or three tests below the cut- off. Intestinal biopsy was necessary as a second diagnostic step in the remaining 60 patients (22%), who had discordant antibody results. This two-step diagnostic procedure reduces the number of intestinal biopsies and increases the sensitivity of the entire diagnostic procedure; only CD patients without any CD-specific antibodies would be missed.
In 1998 , we suggested the above combination of serological tests as a low-risk and cost-effective algorithm for diagnosing various forms of CD. This combination — still using anti ngli — was confirmed in a total of 1,873 patients with jejunal biopsy [30–32]. Because of patient preselection according to their symptoms, the prevalence of CD was 59%. The ppv of three tests with congruent positive antibody results was >99%. The npv of all three antibody tests was 98%. However, 37% (599/1,873) of the patients with discordant antibody results could not be classified by antibody tests alone. In the present study, the number of nonclassified patients was reduced from 37% to 22% (P < 0.001) because anti-dpgli performed better than anti-ngli. Thus, in a population with a high pretest probability of CD, using a combination of three or four antibody tests should obviate the need for as many as 78% of jejunal biopsies.
Several recent studies have questioned the necessity of performing a jejunal biopsy on all individuals with suspected CD [34–39]. One approach defined five criteria, including clinical signs, four of which had to be fulfilled for a diagnosis of CD . Other studies describe the association of very high IgA anti-tTG antibody titers with Marsh 3 histopathology [35–37]. Therefore, recent guidelines, released by the European Society for Pediatric Gastroenterology, Hepathology, and Nutrition stated that intestinal biopsy is redundant in patients with high anti-tTG antibody titers (>10 times the upper limit of normal) . These proposals attain a high specificity and will result in few patients receiving a false-positive diagnosis. However, many CD patients do not have such high anti-tTG titer, and will therefore require intestinal biopsy. Sugai et al.  investigated single antibody tests and various combinations of two-antibody tests in populations with different pretest probabilities for CD. As in our study, sensitivity was lower in combined tests than in single tests; however, ppv increased significantly not only in the population with high pretest probability for CD, but also in the group with low pretest probability for CD. They concluded that: “Appropriate use of CD serology might accurately identify the vast majority of CD patients in populations with different pretest probabilities”.
Recently, Vermeersch et al. illustrated the utility of likelihood ratios for the interpretation of CD serology.  The likelihood ratio for CD was much higher for double positive test than for single positive test results.  Our results showed comparable test results for single and double positive analyses. (Table 3) Similarly, triple positive tests had a high likelihood ratio. However, the best test for CD exclusion was the triple negative test which had a significantly lower likelihood ratio than the double negative test reults reported by Vermeersch et al. (p=0.000037).
Therefore, we speculate that the combined tests with the very high likelihood ratio positive and the very low likelihood ratio negative achieved in the present study group will also identify patients in populations with a low CD frequency.