Multicenter study to compare the diagnostic performance of CLIA vs. FEIA transglutaminase IgA assays for the diagnosis of celiac disease

Introduction

Celiac disease (CD) is an immune-mediated enteropathy caused by gluten ingestion in genetically susceptible patients [1]. Dietary intake of gluten peptides induces an intestinal inflammatory response characterized by activation of gliadin specific CD4⁺ T-cells, and subsequent activation of CD8⁺ intra-epithelial lymphocytes [2]. These cytotoxic cells can cause apoptosis of enterocytes, which leads to villous atrophy. Accordingly, clinical symptoms of CD patients include malabsorption, wasting and diarrhea [3]. Improved symptom awareness has led to better disease recognition, but apart from this effect, the true incidence rate of CD is also increasing among all age groups. Both developments have contributed to the high global disease prevalence rate of approximately 1% [1, 4]. Currently, the only widely accepted treatment for CD is a gluten-free diet (GFD), which usually resolves symptoms and leads to mucosal recovery [5]. A lifelong GFD can, however, negatively impact quality-of-life and strict adherence is challenging. Therefore, correct diagnosis of patients with suspected CD is highly important.

In adult CD patients, the diagnosis is currently based on duodenal biopsies showing villous atrophy, crypt hyperplasia and intra-epithelial lymphocytosis in combination with positive CD-specific serology [6]. The presence of IgA-antibodies directed against transglutaminase-2, or tissue transglutaminase (tTG), is an important cornerstone in the diagnosis of both adult and pediatric patients. tTG deamidates glutamine in gluten peptides. In genetically predisposed CD patients, this can lead to the formation of epitopes with an increased affinity to the disease-associated human leukocyte antigen (HLA) molecules, i.e. DQ2 and DQ8. Through subsequent activation of B-cells, IgA autoantibodies against tTG (tTG-IgA) are produced [7]. Other serological findings in CD patients include antibodies against endomysium (EMA) that are also directed against tTG and are highly specific for the disease. EMA is often used to confirm tTG-IgA antibodies, especially in patients with low tTG-IgA levels. Currently, the first step in the diagnostic process is testing for tTG-IgA antibodies [5, 6].

In children with suspected CD, duodenal biopsy can be omitted if the tTG-IgA level is ≥10× upper limit of normal (ULN) and the presence of EMA is confirmed in a second diagnostic sample [8, 9]. In the previous ESPGHAN guideline positivity for HLA-DQ2 or DQ8 was necessary for diagnosis based on EMA confirmed high tTG-IgA levels without biopsy, however the 2020 guideline states that HLA-typing does not add to the certainty of diagnosis if the other two criteria for CD diagnosis are fulfilled [9, 10]. Therefore, HLA-typing is no longer required. The current guideline states that also asymptomatic children can be diagnosed accordingly, provided that the parents concur. Recent studies showed approximately 50% of diagnostic gastroduodenal biopsies can be avoided with the current diagnostic strategy [11, 12]. However, in adults this approach is not widely adopted in clinical practice, probably due to the lack of sufficient high-quality data across different populations [13].

Currently, in the Netherlands the most frequently used assay techniques to detect tTG-IgA antibodies are the fluorescence enzyme immunoassay (FEIA) and the chemiluminescent immunoassay (CLIA). With chemiluminescence, energy in the form of light is emitted as a result of a chemical reaction, whereas with fluorescence light is indirectly emitted as a result of previously absorbed energy [14]. In a recent study, a higher dynamic range has been reported for CLIA compared to FEIA in the measurement of tTG-antibodies [14]. Manufacturers of tTG assays utilize different strategies to define cut-off values, and as such, low tTG IgA levels may lead to discrepant outcomes in different assays.

Therefore, in this retrospective, multicenter study, we investigated the diagnostic performance of the FEIA and two CLIA tTG-IgA assays for the diagnosis of CD in both adult and pediatric patients. Furthermore, we assessed if the proportion of patients at diagnosis with a tTG-IgA level ≥10× ULN was different between the three assays as this may impact the rate of duodenal biopsies in pediatric patients.

Materials and methods

Results

Diagnostic performance

In this study, we included diagnostic samples of 95 adult and 65 pediatric CD patients. Patient characteristics are shown in Table 1. Predominantly females were diagnosed with CD, especially in the pediatric cohort. According to the ESPGHAN guidelines first published in 2012, a no-biopsy diagnostic strategy was chosen for pediatric patients with an original tTG-IgA level ≥10× ULN [10]. However, in adults the majority of cases was diagnosed with duodenal biopsies (Table 1). Consecutive control samples in this study comprised 479 adult and 253 pediatric patients. This resulted in a case-control ratio of 5.04 and 3.89 in adult and pediatric samples, respectively.

Table 1:

CD patient characteristics.

	Adults (n=95)	Children (n=65)
Mean age, years (range)	44 (18–85)	7 (0–16)
Sex, n (%)	Male 41 (43)	Male 19 (29)
	Female 54 (57)	Female 46 (71)
Marsh classification, n
No biopsy	7	49
Marsh 0a	3	0
Marsh 1a	1	1
Marsh 2	4	1
Marsh 3A-C	80	14
DQ2/DQ8 status, n
Unknown	65	10
DQ2 positive	27	47
DQ2 DQ8 positive	2	4
DQ8 positive	1	4
EMA titer at diagnosis, n
Unknown	23	4
Negative	2	0
Weakly positive	5	1
Positive	32	31
Strongly positive	33	29
Original tTG level, n (%)
> 10× ULN	61 (64)	51 (78)
> 3 and ≤ 10× ULN	20 (21)	9 (14)
> 1 and ≤ 3× ULN	14 (15)	5 (8)

1. ^aCD diagnosis based on a clinical diagnosis and/or a clinical response to GFD.

Each sample was measured with all three tTG-IgA assays. The data showed a high quantitative correlation between each of the three platforms (p<0.0001). The Pearson correlation coefficient between the FEIA and QUANTA Flash was 0.717, between FEIA and ChLIA 0.555, and between the two CLIA assays 0.838. The clinical performance characteristics for adult and pediatric patients are shown in Table 2. Sensitivity of the CLIA assays was higher compared to the FEIA assay, especially in adult patients. In adult patients, sensitivity of the FEIA assay was 88.4% compared to 100% in both CLIA assays. Specificity was high (≥97.6%) for all assays. The FEIA had the highest specificity (adults 99.8%; children 100%) when compared to the QUANTA Flash (adults 99.4%; children 98.0%) and ChLIA assay (adults 98.7%; children 97.6%). This was also reflected in a higher positive likelihood ratio (Table 2).

Table 2:

Diagnostic performance characteristics of three tTG-IgA assays.

	EliA Celikey IgA		QUANTA Flash® h-tTG IgA		Euroimmun ChLIA
	Adults	Children	Adults	Children	Adults	Children
Sensitivity, %	88.4	96.6	100	100	100	100
Specificity, %	99.8	100	99.4	98.0	98.7	97.6
Positive likelihood ratio	442	∞	167	50	77	42
Negative likelihood ratio	0.116	0.031	0	0	0	0

Subsequently, the discordant cases and controls were studied. Overlapping positivity for the different tTG-IgA assays is shown as Venn diagrams (Figure 1). A high rate of overlap for the two CLIA assays in the group of false-positive controls was observed, especially in pediatric patients. In children, 5/6 false positive controls were positive with both CLIA assays. In false positive adult control samples, concordance between the two CLIA assays was 3/6. In all discordant samples, EMA-IgA was determined to study the additional value of this assay in these discrepant cases and controls. The results varied highly among the different cases and controls, although EMA-IgA were exclusively detected in CD cases (data not shown).

Figure 1:

Venn diagrams of overlapping positivity for three tTG-IgA assays tested in a cohort of CD patients and control patients.

Next, a receiver operating characteristic (ROC)-curve analysis was performed to determine the optimal cut-off value for each assay (Figure 2). All assays demonstrated a high area under the curve (AUC). In adult patients, the FEIA showed an AUC of 0.998, slightly lower compared to the CLIA assays. ROC-analysis revealed an optimal cut-off level for the FEIA of 6.5 U/mL, close to the manufacturer’s lower value of the equivocal range of 7.0 U/mL. The use of this cut-off increases sensitivity, especially in adult patients, without a significant effect on specificity (Table 3). ROC-analysis for both CLIA did not reveal improved cut-off values.

Figure 2:

Receiver operator characteristic (ROC)-curve for three tTG-IgA assays. (A) ROC-curve for adult patients. (B) ROC-curve for pediatric patients; the EliA Celikey IgA and QUANTA Flash h-tTG IgA ROC curves are overlapping.

Table 3:

Diagnostic performance characteristics of the tTG-IgA assay EliA Celikey IgA (Phadia) for different cut-off values.

	6.5 U/mL		7.0 U/mL		10 U/mL
	Adults	Children	Adults	Children	Adults	Children
Sensitivity, %	95.8	100	93.7	100	88.4	96.6
Specificity, %	99.6	100	99.6	100	99.8	100
Positive likelihood ratio	240	∞	234	∞	442	∞
Negative likelihood ratio	0.042	0	0.063	0	0.116	0.031

tTG-IgA level ≥10× ULN

If tested with the FEIA, 41/95 (43%) adult cases had a diagnostic tTG-IgA level ≥10× ULN (Table 4). In pediatric patients, this was 46/65 (71%). No adult or pediatric control patients demonstrated a tTG-IgA level ≥10× ULN. Consequently, if a no-biopsy diagnostic strategy based on tTG-IgA ≥10× ULN was chosen, 43% of adult and 71% of pediatric patients tested in this cohort based on FEIA results would not have needed to undergo a duodenal biopsy. If a cut-off of 6.5 U/mL was chosen, these percentages increased to 59 and 74% for adult and pediatric patients, respectively. No control samples tested positive for a tTG-IgA value ≥10× ULN if 6.5 U/mL was selected as the cut-off value.

Table 4:

The percentage of diagnostic cases with a tTG-IgA level ≥10× ULN.

Threshold	EliA Celikey IgA 10 U/mL		EliA Celikey IgA 7.0 U/mL		EliA Celikey IgA 6.5 U/mL		QUANTA Flash® h-tTG IgA 20 CU		Euroimmun ChLIA 10 CU/mL
	Adults	Children	Adults	Children	Adults	Children	Adults	Children	Adults	Children
tTG-IgA level ≥10× threshold, %	43	71	59	72	59	74	82	91	81	92

For the CLIA assays the percentage of adult and pediatric diagnostic samples with a diagnostic tTG-IgA level ≥10× ULN was significantly higher (Table 4). In adult patients this was 82 and 81% for the QUANTA Flash and ChLIA platform, respectively. In pediatric patients the percentages were 91 and 92% for the CLIA assays. No control samples tested positive with a level ≥10× ULN. Taken together, a higher proportion of patients in this cohort had ≥10× ULN when tested with a CLIA compared to the FEIA, and pediatric patients more often present with levels ≥10× ULN as compared to adult patients.

In accordance with a higher proportion of levels ≥10× ULN in pediatric patients, the median tTG-IgA value in pediatric patients was higher compared to adult patients for all tTG-IgA assays. For the FEIA, the median tTG-IgA level was 2.85× higher in children with CD compared to adults. This was 2.62× and 1.01× for the QUANTA Flash and ChLIA, respectively. Thus, the mean level difference between adult and pediatric patients is smaller for the CLIA assays as compared to the FEIA assay. Indeed, relatively more low-positive tTG-IgA levels were observed in adult patients tested with FEIA (Figure 3). For example, 12/95 adult cases had a tTG-IgA level 1–3× ULN with the FEIA assay compared to 5/95 measured by both CLIA assays. A scatter plot of tTG IgA levels in individual samples illustrates the difference in tTG-IgA level distribution of the FEIA assay as opposed to the CLIA assays as well (Supplemental Figure 1). The tTG-IgA levels were lower in relation to the cut-off in samples analyzed with FEIA as compared with CLIA, especially in adults.

Figure 3:

The distribution of diagnostic tTG levels of CD patients amongst different ULN categories measured by three tTG-IgA assays.

Discussion

tTG-IgA serology is highly important in the diagnosis of CD, and in pediatric cases a diagnostic duodenal biopsy can be omitted if the level is ≥10× ULN [9]. However, tTG-IgA assays are based on different techniques, standardization is lacking and test performance may vary [15, 16]. Therefore, the aim of this retrospective multicenter study was to compare the diagnostic performance of two commonly used assays and one recently launched tTG-IgA assay for the diagnosis of CD. All cases and controls, especially the false positives and false negatives, were carefully reviewed to establish a correct classification of disease vs. control patient.

Previously, a retrospective study comparing the diagnostic performance of the FEIA and the QUANTA Flash in adult patients concluded that the sensitivity of the FEIA was lower than the CLIA if the upper value of the equivocal range (10 U/mL) was used as cut-off. If the lower value of the equivocal range (7.0 U/mL) was used, the sensitivity was similar for both assays [17]. Vermeersch et al. demonstrated that the FEIA had a slightly lower sensitivity and a higher specificity compared to the QUANTA Flash if a cut-off of 7.0 U/mL was used in an adult population [18]. In pediatric patients, a recent retrospective study showed no difference in sensitivity between the FEIA and the QUANTA Flash assays if the lower value of the equivocal range (7.0 U/mL) was selected as cut-off for the FEIA [19]. Specificity was ≥98% for both assays, similar to our pediatric cohort. Others showed high specificity (100%) and sensitivity (97.7%) of the FEIA in a retrospective pediatric cohort study, at the 10 U/mL cut-off [20]. Recent data showed a lower sensitivity of the FEIA in adults compared to children [21]. All these data support our findings that the FEIA has less diagnostic accuracy in adults as opposed to pediatric patients, but that assays are better aligned if a lower cut-off is used. The difference in performance characteristics between the three assays especially become manifest if the 10 U/mL cut-off is used for the FEIA. On the other hand, at both the 10 U/mL and 7.0 U/mL cut-offs the positive likelihood ratio of the FEIA is higher compared to the CLIA assays, exemplifying the higher specificity. These assay characteristics should be taken into account by the requesting physician. Limitations of previous studies include that these were performed in specific populations (e.g. children or adults) and samples were included from one center, possibly increasing selection bias [17, 18]. Furthermore, some studies predominantly included healthy controls instead of disease controls. Novel aspects of our study include the multicenter study design, inclusion of CD patients across all age groups, using relevant disease controls and the inclusion of two distinct CLIA assays.

Taken together, our data show in accordance with previous studies that lowering the cut-off of the FEIA improves diagnostic characteristics especially in adult patients who more often present with lower levels of tTG-IgA in comparison to pediatric cases [17]. Furthermore, concordance between both CLIA assays was higher than concordance between FEIA and either of the CLIA assays, suggesting that assay techniques affect tTG-IgA test results.

In our study, a tTG-IgA level ≥10× ULN was only observed in CD patients and not in the disease control group. A recent study showed a positive predictive value (PPV) >99.5% for a tTG-IgA level ≥10× ULN for 10 different tTG-IgA assays in pediatric patients; in this study an ULN of 7.0 U/mL was used for the FEIA [11]. In pediatric patients tested with the QUANTA Flash similar diagnostic performance results were obtained [22]. In this study, a higher cut-off value than 10× ULN was advised to increase the PPV regarding the presence of CD-associated mucosal damage. For diagnostic purposes this is less relevant as CD diagnosis was established based on other parameters, but for follow-up this is relevant considering the tTG-IgA level was positively correlated to the degree of mucosal damage [22]. Our data demonstrated that depending on the tTG-IgA platform used between 71 and 92% of children did not need to undergo a duodenal biopsy for CD diagnosis, which is higher than the approximately 60% reported by Riznik et al. [23] This lower percentage may be caused by a small subset of patients in this study that had IgA-deficiency, which was an exclusion criterion for our study or the assay used to test for tTG-IgA. Other studies, which also included asymptomatic patients, showed percentages of 77 and 69% of newly diagnosed pediatric CD patients with a tTG-IgA ≥10× ULN [12, 24]. Using two tTG-IgA platforms simultaneously to determine if at least one test shows a level ≥10× ULN may even decrease the need for duodenal biopsy to <11%, as was previously shown [25].

In adult CD patients, a recent study involving three cohorts with different pretest probabilities and using two tTG-IgA platforms including FEIA supported a no-biopsy strategy for all cohorts if the tTG-IgA level was ≥10× ULN [26]. A different study showed a PPV of 100% in different at-risk populations when the tTG-IgA level was ≥10× ULN in combination with the presence of EMA [27]. In conclusion, our data show that, similar to other studies, a tTG-IgA level ≥10× ULN may lead to an accurate CD diagnosis in adults without the need for duodenal biopsies. The proportion of adult patients with a diagnostic tTG-IgA level ≥10× ULN was higher for both CLIA assays compared to the FEIA assay, which is in agreement with previous work [17]. Based on these findings, using a CLIA will result in less need for upper gastrointestinal endoscopy in adult patients, hereby decreasing patient discomfort, potential procedural complications and healthcare costs.

In our study, the ULN selected to determine the need for an intestinal biopsy was defined as the recommended manufacturer’s cut-off, which is similar to previous studies investigating diagnostic characteristics of tTG-IgA assays [17, 19, 23]. However, manufacturers utilize various strategies to define the cut-off based on a cohort of patients and healthy and/or disease controls [28]. Moreover, in case of an equivocal range, both clinical laboratories as well as studies make different choices by selecting either the lower value or the higher value as the ULN. By definition, the ULN should be determined as a percentile of healthy, i.e. normal, controls. However, the original ESPGHAN 2012 guideline defined ULN as a ROC optimized cut-off obtained with CD patients and disease controls [10]. One should be aware of the discrepancy in the definition and use of ULN as this may impact clinical practice.

The main limitations of our study include the retrospective study design, which introduces a possible selection bias, and that no patient samples were selected by a laboratory using the tTG-IgA ChLIA (Euroimmun) assay. However, with our strategy a similar number of study patients were selected by a CLIA assay and FEIA assay.

A previous study showed predictive value of tTG-IgA levels for mucosal damage, although no comparison was made between the different tTG-IgA assays [22, 29]. Future studies comparing different tTG-IgA assays and their predictive value for villous atrophy, both at diagnosis and after introduction of GFD, may provide data to better predict response to GFD and reduce follow-up biopsies.

In conclusion, our study shows high diagnostic accuracy for three different tTG-IgA assays. CLIA assays have a higher sensitivity compared to the FEIA assay, particularly for adult patients. Specificity is high for all assays with the FEIA having the highest specificity. Both CLIA assays show a high level of concordance. Furthermore, a higher proportion of CD patients test ≥10× ULN with the CLIA assays compared to the FEIA assay and would thus not need to undergo duodenal biopsies.