Systematic Review and Meta-Analysis: Accuracy of Both Gamma Delta+ Intraepithelial Lymphocytes and Coeliac Lymphogram Evaluated by Flow Cytometry for Coeliac Disease Diagnosis

It has been suggested that in doubtful cases of coeliac disease, a high CD3+ T-cell receptor gamma delta+ (TCRγδ+) intraepithelial lymphocyte count increases the likelihood of coeliac disease. Aim: To evaluate the diagnostic accuracy of both an isolated increase of TCRγδ+ cells and a coeliac lymphogram (increase of TCRγδ+ plus decrease of CD3− intraepithelial lymphocytes) evaluated by flow cytometry in the diagnosis of coeliac disease. Methods: The literature search was conducted in MEDLINE and EMBASE. The inclusion criteria were: an article that allows for the construction of a 2 × 2 table of true and false positive and true and false negative values. A diagnostic accuracy test meta-analysis was performed. Results: The search provided 49 relevant citations, of which 6 were selected for the analysis, which represented 519 patients and 440 controls. Coeliac lymphogram: The pooled S and Sp were 93% and 98%, without heterogeneity. The area under the SROC curve (AUC) was 0.98 (95% CI, 0.97–0.99). TCRγδ+: Pooled S and Sp were both 95%, with significant heterogeneity. The AUC was 0.97 (95% CI, 0.95–0.98). Conclusions: Both TCRγδ+ count and coeliac lymphogram assessed by flow cytometry in duodenal mucosal samples are associated with a high level of diagnostic accuracy for and against coeliac disease.


Introduction
Coeliac disease (CD) is an immune-mediated systemic disorder elicited by the ingestion of gluten in genetically susceptible individuals. The pooled global prevalence of CD based on serological test results is 1.4% and based on biopsy results is 0.7% [1]. CD is characterized by the presence of a varied array of gluten-dependent clinical manifestations, CD-specific antibodies, HLA-DQ2 or HLA-DQ8 haplotypes, and enteropathy [2][3][4][5]. Generally speaking, CD diagnosis presents no difficulties when the biopsy shows severe villous atrophy and crypt hyperplasia. In clinical practice, however, diagnosis is often less straightforward. Diagnostic difficulties arise especially when biopsy findings are borderline. In such cases, there exists the risk of both under-and over-diagnosis.
Tissue transglutaminase IgA class autoantibodies (anti-tTG2) are the serological markers of choice for the detection of CD [3][4][5]. Serological tests have a high specificity and sensitivity, but can fluctuate in cases with mild intestinal damage and a low gluten intake [6,7]. Moreover, serology seems to have a lower sensitivity and specificity in adults [8]. Table 2 describes the flow cytometry technique used in the selected studies, showing some differences, mainly in the definition of the CD3 − IEL subsets.   Table 1 summarizes the six selected cross-sectional studies, all of which were focused on European populations, all except one having been carried out in Spain. One of them studied children and adult populations separately [25], and these were included separately in the analysis. For two studies, only the coeliac lymphogram, and for one study, only TCRγδ + IEL were described; thus, there were five eligible studies for the analysis of coeliac lymphogram (in one of them, separately for children and adults), and four eligible studies for the analysis of TCRγδ + IEL diagnostic accuracy. Table 2 describes the flow cytometry technique used in the selected studies, showing some differences, mainly in the definition of the CD3 − IEL subsets.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled non-consecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1. Table 3. Results of Quality Assessment of Studies of Diagnostic Accuracy included in Systematic Reviews (QUADAS-2) checklist.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1. Table 3. Results of Quality Assessment of Studies of Diagnostic Accuracy included in S Reviews (QUADAS-2) checklist.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1. Table 3.

Results of Quality Assessment of Studies of Diagnostic Accuracy included in Systematic
Reviews (QUADAS-2) checklist.

Reference Standard
Camarero

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 to (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled no consecutive patients with known disease and a control group without the condition, which m exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index te had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multip logistic regression was developed to calculate the probability of having CD using both TCRγδ + a CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studi the test threshold was not described in the paper but was provided by the authors after contacti them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementa Table S1.
Nutrients 2019, 11, x FOR PEER REVIEW

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the Q (Table 3). In two studies, the risk of bias about patient selection was high, since they consecutive patients with known disease and a control group without the conditio exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct o had introduced a bias in one study [15], since the test threshold was not pre-specified, logistic regression was developed to calculate the probability of having CD using bo CD3 − counts, which could lead to overoptimistic estimates of test performance. In thre the test threshold was not described in the paper but was provided by the authors a them and was the threshold pre-specified for clinical use in their labs [14,25,26]. concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in S Table S1.
Nutrients 2019, 11, x FOR PEER REVIEW

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated (Table 3). In two studies, the risk of bias about patient selection was hig consecutive patients with known disease and a control group without exaggerate diagnostic accuracy [14,15]. There was also a high risk that th had introduced a bias in one study [15], since the test threshold was not p logistic regression was developed to calculate the probability of having C CD3 − counts, which could lead to overoptimistic estimates of test perform the test threshold was not described in the paper but was provided by t them and was the threshold pre-specified for clinical use in their labs concerns regarding test applicability.
The methodological quality of the flow cytometry methods is sum Table S1.

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUADAS-2 tool (Table 3). In two studies, the risk of bias about patient selection was high, since they enrolled nonconsecutive patients with known disease and a control group without the condition, which may exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the index test had introduced a bias in one study [15], since the test threshold was not pre-specified, and a multiple logistic regression was developed to calculate the probability of having CD using both TCRγδ + and CD3 − counts, which could lead to overoptimistic estimates of test performance. In three other studies, the test threshold was not described in the paper but was provided by the authors after contacting them and was the threshold pre-specified for clinical use in their labs [14,25,26]. There were no concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supplementary Table S1.

Low risk;
Nutrients 2019, 11, x FOR PEER REVIEW

Qualitative Analysis of Included Studies
The methodological quality of the included studies was evaluated using the QUAD (Table 3). In two studies, the risk of bias about patient selection was high, since they enr consecutive patients with known disease and a control group without the condition, w exaggerate diagnostic accuracy [14,15]. There was also a high risk that the conduct of the had introduced a bias in one study [15], since the test threshold was not pre-specified, and logistic regression was developed to calculate the probability of having CD using both TC CD3 − counts, which could lead to overoptimistic estimates of test performance. In three oth the test threshold was not described in the paper but was provided by the authors after them and was the threshold pre-specified for clinical use in their labs [14,25,26]. Ther concerns regarding test applicability.
The methodological quality of the flow cytometry methods is summarized in Supp Table S1.

Quantitative Analysis of Included Studies
Coeliac lymphogram: As mentioned, a coeliac lymphogram consists of both an i TCRγδ + IEL and a decrease in CD3 − IEL. Pooled sensitivity and specificity ( Figure 2) were CI, 89-96%) and 98% (95% CI, 93-99%), respectively. The SROC curve is described in Figure  the  The methodological quality of the flow cytometry methods is summarized in Supplementary  Table S1.

Quantitative Analysis of Included Studies
Coeliac lymphogram: As mentioned, a coeliac lymphogram consists of both an increase in TCRγδ + IEL and a decrease in CD3 − IEL. Pooled sensitivity and specificity ( Figure 2) were 93% (95% CI, 89-96%) and 98% (95% CI, 93-99%), respectively. The SROC curve is described in Figure 3A, being the area under the curve (AUROC) of 0.98 (95%, 0.97-0.99), which implies a high diagnostic accuracy for diagnosing CD. There was no significant heterogeneity.  A summary of LRP and LRN with 95% CIs is described in Supplementary Figure S1, suggesting that the test is useful for confirmation (when positive) and exclusion (when negative) of CD. Fagan's nomogram is presented in Figure 4A. Among patients with a pre-test coeliac disease probability of 55%, post-test probabilities were 98% and 8% for positive and negative coeliac lymphogram, respectively.  A summary of LRP and LRN with 95% CIs is described in Supplementary Figure S1, suggesting that the test is useful for confirmation (when positive) and exclusion (when negative) of CD. Fagan's nomogram is presented in Figure 4A. Among patients with a pre-test coeliac disease probability of 55%, post-test probabilities were 98% and 8% for positive and negative coeliac lymphogram, respectively. A summary of LRP and LRN with 95% CIs is described in Supplementary Figure S1, suggesting that the test is useful for confirmation (when positive) and exclusion (when negative) of CD. Fagan's nomogram is presented in Figure 4A. Among patients with a pre-test coeliac disease probability of 55%, post-test probabilities were 98% and 8% for positive and negative coeliac lymphogram, respectively.  TCRγδ + IEL: Pooled sensitivity ( Figure 5) was 95% (95% CI, 82-99%) with significant heterogeneity (I 2 = 95.6%; p < 0.001). Pooled specificity ( Figure 5) was 95% (95% CI, 91-97%), with heterogeneity (I 2 = 62%; p = 0.05). The SROC curve is described in Figure 3B, being the AUROC of 0.97 (95%, 0.95-0.98), which implies a high diagnostic accuracy for diagnosing CD.
Supplementary Figure S1 and Figure 4B describe the likelihood ratio matrix and the Fagan's nomogram, respectively. It is suggested that the test is useful for confirmation (when positive) and exclusion (when negative) of CD. Among patients with a pre-test coeliac disease probability of 55%, post-test probabilities were 96% and 6% for positive and negative TCRγδ + , respectively.  Supplementary Figure S1 and Figure 4B describe the likelihood ratio matrix and the Fagan's nomogram, respectively. It is suggested that the test is useful for confirmation (when positive) and exclusion (when negative) of CD. Among patients with a pre-test coeliac disease probability of 55%, post-test probabilities were 96% and 6% for positive and negative TCRγδ + , respectively.

TCRγδ + IEL Evolution after a Gluten-Free Diet
Four of the selected studies described the percentage of TCRγδ + IEL in patients on a GFD [14,15,26,27]. One of them prospectively assessed the evolution of this percentage after 1 year on a GFD [14]. In addition, we added data from our series, published only as an abstract [28], also prospectively evaluating the changes in TCRγδ + IEL after a GFD. Ultimately, 201 patients were evaluated (Table 4). Results demonstrated the persistence of the increased values of this intraepithelial T lymphocyte subset after long-term follow-up on a GFD in both children and adults. Regrettably, it was not possible to perform a meta-analysis of these data.

Discussion
Recent ESsCD guidelines for CD diagnosis states in the "areas of future research" section that studies are needed to evaluate T-cell flow cytometry and make it widely available for clinical use [5]. The results reported herein suggest that both coeliac lymphogram (an increase in TCRγδ + IEL plus a decrease in CD3 − IEL) and the isolated increase in TCRγδ + IEL, assessed by flow cytometry in duodenal specimens, would be appropriate tests for CD diagnosis, since they were associated with an AUROC of 0.97-0.98 and LR values representing strong evidence both in favor of and against CD. Coeliac lymphogram was associated with a higher specificity than the assessment of isolated TCRγδ + cells, whereas an isolated TCRγδ + cell count was associated with a higher sensitivity.
The advantages of the use of flow cytometry to assess TCRγδ + cells are considerable compared to other user-dependent techniques [18]. Results are obtained in an objective, quantitative, reproducible way, allowing for the analysis of a greater number of cells than with immunohistochemistry. The concomitant measurement of CD3 − IEL adds specificity to the assay, as has been demonstrated in the present and previous studies [18]. Although the analysis of IEL subpopulations is, in general, not needed for CD diagnosis in seropositive patients, it may have high diagnostic value in doubtful cases that involve differentiating between CD and non-CD atrophy, those in which the mucosal lesion is equivocal [29], when HLA-DQ2.5 and HLA-DQ8 are negative [30], or when serum tTG levels are negative or with low titers (mainly those associated with negative EmA) [31]. The World Gastroenterology Organisation Global Guidelines for CD diagnosis recommend second biopsies to be performed in patients in whom the first biopsies and serological tests have been inconclusive (e.g., seronegative enteropathy) [32], but this strategy implies a second invasive procedure and a considerable delay in the final diagnosis. Finally, it may be useful in HLA-DQ2/8 + patients presenting with a grade 1 mucosal lesion since it is a non-specific lesion [10,24,33], which is caused by CD in around 5-15% of cases [34], and serology is positive in only 20-30% of them [35]. In this sense, subjecting a patient to an invasive procedure like gastroscopy with biopsies should involve trying to get as much information as possible from the procedure in order to achieve an accurate diagnosis. Taking an additional biopsy for flow cytometry is an easy procedure and may yield substantial information. Most labs in tertiary and even secondary hospitals have a flow cytometer for diagnostic purposes and analyzing the lymphocyte subpopulations in the duodenal mucosa is an affordable technique.
From the 958 articles selected in this systematic review, only 6 of the 49 references that were classified as potentially relevant were finally included in the meta-analysis. Some of the excluded studies had a high methodological quality, but did not meet the inclusion criteria and, consequently, were not useful for the aim of our meta-analysis, since most of them used immunohistochemistry instead of flow cytometry to assess TCRγδ + cells. Other reasons for excluding studies were reasonable doubts on how CD diagnosis was achieved or on how CD was ruled out in the control group, and the absence of the data required to construct the 2 × 2 tables. The six studies included in the present meta-analysis comprised 959 individuals. Most of them were cross-sectional studies carried out in Spain, where the intraepithelial lymphogram technique was first described and is widely utilized, and has been recently included in the Ministry of Health guidelines for the early diagnosis of CD [36]. Two studies included only pediatric patients, three evaluated a mix of pediatric and adult populations, one of them studied children and adults separately, and one included only adults. Since there were only two studies independently assessing an adult population, a meta-regression analysis could not be performed. In addition, the median age of included patients in the study performed only in adults was 55 years, much higher than in other ones; a TCRγδ+ IEL count resulted in 66% sensitivity for CD diagnosis in that study [27], likely being a reason for the heterogeneity observed in meta-analysis. It is noteworthy that the CD patients with normal TCRγδ + IEL in that study were significantly older than those with abnormal values [27]. The authors argued that advanced age may be a factor preventing the characteristic increase of TCRγδ + IEL in CD for unknown reasons, and thus might explain the lower sensitivity found in that study. In the validation cohort of the cut-off for TCRγδ + established in our laboratory, we did not find differences in TCRγδ + sensitivity (≈94%) for CD diagnosis between different age groups, but the patients included were under 50 years of age [37]. It is possible that some subjects with suspected CD who are over 50 years of age could have a negative coeliac lymphogram; thus, further studies will be required to assess the TCRγδ + sensitivity in these patients. Specificity, in contrast, was very high regardless of age.
As mentioned, it has been suggested that TCRγδ + values remain elevated in CD patients despite a GFD. The present systematic analysis included 200 patients supporting this observation. The persistence of the increased TCRγδ + cell values after a long-term GFD (1 year to a mean of 5 years) opens up the possibility of using this biomarker to confirm CD in patients who were started on a GFD, and for whom serology and histology may have yielded misleading results. In addition, it may prove useful for distinguishing between CD and non-coeliac gluten sensitivity in patients who are symptom-free after a GFD and reluctant to undergo a gluten challenge. Further studies are required to evaluate the persistence of the increased TCRγδ + cell populations after longer follow-up.
There were small differences in the flow cytometry technique used between the six studies. Aspects of the sampling, the number of biopsies used, and the sampling locations differed between studies or were not appropriately described. However, unlike histopathological analyses, the IEL pattern is uniformly distributed in the distal duodenum and the duodenal bulb in healthy donors and coeliac patients [38], and the sampling location is therefore not a critical factor. All of the six studies used calcium chelation and shaking for the isolation of IEL, with minor differences between them (time, temperature, and buffer). The number of isolated IELs, the viability of the obtained cell suspension and the amount of starting material were not always reported, which is not critical due to the high efficiency of the isolation protocols [18]. The major differences between the six papers were in the staining panel, gating strategy, and IEL subset definition. Only one of the six papers excluded dead cells in its gating strategy, which is an essential step for reducing nonspecific stained cells that would interfere in the results. Of particular importance is the definition of the CD3 − subset, given the heterogeneous nature of this subset and its importance in defining the "coeliac lymphogram": a consensus regarding the most adequate definition should be reached to reduce confusion in the field. In fact, the CD3 − subset evaluated by Nijeboer et al. was completely different from that in the rest of studies (Table 2); it corresponded to a small proportion of IELs lacking surface CD3 but expressing intracellular CD3, which was massively expanded in refractory CD type II [39]. The aforementioned differences in analysis strategy might explain, at least in part, the differences in the cut-off, which might not be transferable from lab to lab. In fact, four different cut-off values for TCRγδ + were used in the six studies. All these differences could account for the heterogeneity observed in the present meta-analysis. Heterogeneity is to be expected in meta-analyses of diagnostic test accuracy, and for this reason, a random effects model was used for the present analysis.
None of the papers fulfilled the criteria to be considered reproducible according to the MIFlowCyt guidelines [22]. Considering the proven clinical usefulness of the IEL pattern, an effort should be made by all researchers to report methodologies at a sufficient level of detail to allow other groups to implement them in a standardized manner.
A limitation of the present study was the heterogeneity of the control group (a small number of patients with non-coeliac atrophy or other enteropathies, and a majority with functional bowel disease patients, gastro-esophageal reflux disease, Helicobacter pylori gastritis, parasitic infections, etc., all with normal small bowel histology). In this sense, the absence of a subgroup of seronegative CD among cases and a subgroup with enough non-CD atrophy patients among controls was a drawback, since these are the patient subgroups in which coeliac lymphogram has the highest diagnostic interest. However, two of the included studies provided data on a small sample of non-coeliac villous atrophy patients, suggesting that TCRγδ + count may be useful to rule out CD in this setting. Finally, since all included articles except one were carried out in Spain, it could be argued whether the results might be extrapolated to other populations. However, the increase of TCRγδ + in patients with CD has been described in studies from other countries around the world [13,[40][41][42][43][44].

Conclusions
In conclusion, both TCRγδ + count and coeliac lymphogram assessed by flow cytometry in duodenal mucosal samples have been associated with a high level of diagnostic accuracy for and against coeliac disease. Further studies are warranted to confirm the diagnostic value of the technique in cases in which diagnosis is not straightforward.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6643/11/9/1992/s1, Figure S1: Likelihood matrix for the overall distribution of the studies: (A) Coeliac lymphogram; (B) TCRγδ + . Each point corresponds to a study. The summary positive likelihood ratio (LRP) and negative likelihood ratio (LRN) for index test are found on the left upper quadrant of the matrix, and showed a high accuracy for both exclusion and confirmation of CD., Table S1: Methodological quality of the flow cytometry methods