The estimated true prevalence of celiac disease is 1–3% [1
], emphasizing the importance of efficient and practical diagnostic strategy for this common condition. At the same time, diagnosis of a life-long disease must be based on solid evidence. This has long been achieved by demonstrating characteristic mucosal damage in duodenal biopsy, but such a histology-based approach is invasive and technically challenging [3
]. Together with the high specificity of modern serological tests, the aforesaid challenges have led to the introduction of new pediatric guidelines for diagnosing celiac disease, enabling a non-biopsy diagnosis in selected children with anti-transglutaminase 2 antibodies (TGA) > 10× the upper limit of normal (ULN) and positive anti-endomysial antibodies (EmA) [5
It has been suggested that the biopsy-omitting approach could also be extended to adult patients [6
] and, supporting this, we recently reported a positive predictive value (PPV) of 100% for the serological criteria in adults with variable pre-test probabilities for celiac disease [8
]. The results, however, were obtained utilizing only one TGA assay and thus cannot be directly generalized due to a lack of standardization between the commercial tests. In fact, the artificial 10× ULN cutoff and requirement of EmA were introduced mainly to overcome the variation in the diagnostic performance of the TGA-based tests [5
]. So far only a few-and exclusively pediatric-studies have directly compared the accuracy of non-biopsy approach for celiac disease with different TGA assays [9
We investigated this issue by applying four widely used commercial TGA tests in two large and well-defined cohorts of adults with either clinical suspicion (high pre-test probability) or family risk (moderate pre-test probability) of celiac disease.
Altogether 125 subjects in the clinical cohort had a diagnostic duodenal lesion in either the primary EGD or after prolonged gluten consumption, and a further 12 received the diagnosis based on special investigations (Figure 1
A, Table S1
). Correspondingly, 85 subjects in the family cohort received the diagnosis either directly or after additional investigations (Figure 1
B, Table S1
The two study cohorts had comparable median ages, while there was a female predominance in the clinical cohort compared with the almost even gender distribution in the family cohort (Table 1
). By definition, 100% of the subjects in the family cohort had relative(s) with celiac disease, while the corresponding proportion in the clinical cohort was approximately one-fifth. All participants receiving the diagnosis had HLA DQ2/8 genotype consistent with celiac disease (Table 1
The overall frequency of seropositivity using manufacturer’s cut-offs for the TGA assays tested ranged from 48.5% to 62.3% in the clinical cohort and from 15.4% to 43.0% in the family cohort. The corresponding numbers in those receiving a celiac disease diagnosis were 84.7–96.6% and 76.5–98.8% (Table 1
When applying a cut-off 10× ULN, all four TGA assays showed a PPV of 100% in both clinical (95% CIs from 88.0–100% to 92.0–100%) and family (95% CIs from 78.1–100% to 87.0–100%) cohorts (Table 2
). With the pre-defined 1× ULN cut-offs the corresponding PPVs ranged in clinical cohort from 83.6% to 100% (95% CIs from 76.0–89.2% to 96.0–100%) and in family cohort from 90.3% to 100% (95% CIs from 82.0–95.2% to 90.7–99.9%), respectively (Table 2
). The ULNs calculated by exploiting the highest positive TGA value without celiac disease diagnosis for each assay ranged from 1.0× to 5.1× in the clinical cohort and from 1.3× to 4.9× in the family cohort (Table 3
Assuming that only cases with Marsh 3 at any time in the duodenal biopsy or confirmed DH were correctly diagnosed, the PPV for 10× ULN remained 100% in all tests in the family cohort but dropped to 98.1% with QUANTA Lite and to 98.0% with Eurospital in the clinical cohort (Table S2
). The corresponding figures for 1× ULN were 76.9–94.8% in the clinical cohort and 88.2–97.0% in the family cohort (Table S2
). For Celikey and Orgentec, in which the PPV for 10x ULN remained 100% even with “worst case scenario”, the highest values for negative biopsy were 9.6× and 5.3× ULN respectively.
EmA were positive in 89.8% and 98.8% of those with celiac disease in the clinical and family cohorts, respectively (Table 1
). Altogether, EmA was positive in 95.7% of the Celikey, 90.1% of the Orgentec, 78.5% of the Eurospital, and 54.7% of the Inova positive patients; for those who were eventually diagnosed with celiac disease the corresponding figures were 95.6%, 95.0%, 93.6%, and 93.4%, respectively. One subject with TGA >10× ULN in all four tests was EmA negative, as were four subjects with only Eurospital or Inova 10× ULN. All EmA positive participants had HLA DQ2/8.
All four commercial TGA assays tested here demonstrated a PPV of 100% for celiac disease when applying the 10× ULN cutoff as specified by the European Society for Paediatric Gastroenterology Hepatology and Nutrition (ESPGHAN) [5
]. The excellent accuracy of the serology-based criteria observed here, if used as recommended, is in line with the majority of recent retrospective and prospective pediatric studies [9
]. Together with the previous single-assay study by us and some similar reports by other research groups [6
], these findings provide further evidence that the biopsy-sparing guidelines could also be applied to adult celiac disease patients.
Only a limited number of studies have compared the performance of two or more TGA assays in the serological diagnosis of celiac disease in the same patients; in fact, to the best of our knowledge, such comparisons have been reported only in children [7
]. A few previous adult studies have nevertheless utilized two TGA assays in separate cohorts and the result can thus be indirectly compared with our findings [7
]. Zanini et al. [22
] found 5×ULN to be 100% specific for celiac disease with Eurospital and Celikey. However, the ULNs used were 7 U/mL and 16 U/mL for Eurospital (two groups) and 8 U/mL for Celikey, and thus differed from those used here. Efthymakis et al. [7
] observed PPVs of 96% and 100% for two tests with 10×ULN, but the assays used were not reported. In children, Werkstetter et al. [9
] tested the ESPGHAN criteria meticulously with eight TGA-IgA assays, including Celikey, Inova and Eurospital, and found PPVs of 99.8–100% for >10× ULN. In addition, Rozenberg et al. [24
] reported PPVs of >99% for four automated TGA analyzers and 98% for one ELISA kit.
Nevertheless, lower PPVs for non-biopsy approach have also been reported [25
]. These discrepancies may, at least partially, be attributable to methodological differences, including e.g., variable use of ULN cutoff for a given assay, lack of confirmatory EmA testing, and challenges in applying histology as the reference standard [3
]. A further explanation for inconclusive histology could be early developing celiac disease presenting with only mild/patchy mucosal changes or extraintestinal form with no apparent intestinal lesion [5
], as also demonstrated here in those who developed diagnostic lesion only after gluten challenge or who had DH. A subgroup of our patients was also diagnosed with a so-called mild enteropathy celiac disease [12
]. This could be criticized, but they were rigorously investigated with sophisticated diagnostic methods and demonstrated objectively measured treatment response, strongly supporting the presence of celiac disease. Moreover, even if they had all been considered to be non-celiacs, the PPVs for 10× ULN would still have remained excellent.
Of note, performance of the TGA assays was not affected by the assumed pre-test probability for celiac disease, as they worked equally well in both study cohorts. This is in line with our previous single-assay study [8
] and dispels the fears that lower pre-test probability for the disease, particularly in screen-detected individuals, would lead to poorer diagnostic accuracy [6
]. In fact, results of the present and previous studies indicate that serology actually correlates better with the degree of histological lesion than the severity or nature of the symptoms [29
]. Likewise, in recent studies the non-biopsy criteria have been equally reliable in symptomatic and asymptomatic children [19
] and ESPGHAN now allows a non-biopsy approach regardless of the clinical presentation [5
There was also excellent compatibility between TGA values >10× ULN and EmA positivity. This is important, as EmA is considered as the serological reference test that can be utilized to control for the performance variation of the TGA assays [5
]. This finding also indicate that this laborious confirmatory step which is not feasible in all centers could be omitted in case of well-validated TGA tests with values >10× ULN. EmA could nevertheless remain useful in borderline cases. In contrast, as also reported in other recent studies [8
], measurement of celiac disease-associated HLA did not provide additional benefits. Accordingly, HLA is no longer required by ESPGHAN [5
] although it can be valuable in the differential diagnostics of seronegative duodenal lesion [34
Depending on the TGA assay, lower than 10× ULN (1.0×–5.1×) also showed 100% PPV for celiac disease, and similar findings have been reported in a few earlier studies [22
]. This suggests that the proportion of subjects eligible for non-biopsy approach could be increased. The lack of standardization between the TGA tests and their different diagnostic performances nevertheless complicates this issue. As also seen here, some of the assays are clearly intended more as sensitive screening tests whereas others have higher specificity even with low positive values. One possible option might be serial testing with two TGA assays, but at this point the scarcity of evidence makes it challenging to decide optimal combination. EmA confirmation could still be useful in these circumstances, particularly if aiming to lower the required ULN factor or when testing a new TGA combination. On the other hand, the chosen strategy should not be too complicated for routine clinical use. More studies designed to investigate specifically this issue are urgently needed.
4.1. Strengths and Weaknesses
The main strengths of our study were the use of two large and well-defined cohorts of patients with different pre-test probabilities for celiac disease [3
]. We were also able to exploit sophisticated diagnostic methods in inconclusive cases, although special investigations were not conducted systemically on all participants. There were also limitations. First, EGDs were not centralized to a single hospital, although this should not be a major source of bias as celiac disease diagnostics are harmonized throughout Finland [36
]. Second, some of the patients in the clinical cohort had already been tested for TGA before their referral, but significant bias is again unlikely since the results were comparable with those of the previously untested family cohort. Finally, one more potential shortcoming was that we did not control for a possible reduction of the dietary gluten while awaiting the biopsy.
We found the serology-based diagnosis of celiac disease in adults to be accurate with different commercial TGA assays in individuals with variable pre-test probabilities, further supporting the transition towards a less invasive diagnostic approach also after childhood. The results also suggest that, at least in well-validated TGA assays, the specified cut-off for non-biopsy approach could be lowered, but more research on this issue is needed.