Diagnostic test accuracy of anti-glycopeptidolipid-core IgA antibodies for Mycobacterium avium complex pulmonary disease: systematic review and meta-analysis

Currently, an anti-glycopeptidolipid (GPL)-core IgA antibody assay kit for diagnosing Mycobacterium avium complex (MAC) is commercially available. We conducted this systematic review and meta-analysis to reveal the precise diagnostic accuracy of anti-GPL-core IgA antibodies for MAC pulmonary disease (MAC-PD). We systematically searched reports that could provide data for both sensitivity and specificity by anti-GPL-core IgA antibody for clinically diagnosed MAC-PD. Diagnostic test accuracy was estimated using the bivariate model. Of the 257 articles that we had found through primary search, we finally included 16 reports consisted of 1098 reference positive subjects and 2270 reference negative subjects. The diagnostic odds ratio was 24.8 (95% CI 11.6–52.8, I2 = 5.5%) and the area under the hierarchical summary receiver operating characteristic curves was 0.873 (95% CI 0.837–0.913). With a cutoff value of 0.7 U/mL, the summary estimates of sensitivity and specificity were 0.696 (95% CI 0.621–0.761) and 0.906 (95% CI 0.836–0.951), respectively. The positive and negative likelihood ratios were 7.4 (95% CI 4.1–13.8) and 0.34 (95% CI 0.26–0.43), respectively. The demanding clinical diagnostic criteria may be a cause of false positive of the index test. The index test had good overall diagnostic accuracy and was useful to ruling in MAC-PD with the cutoff value.

Index test. We evaluated anti-GPL-core IgA antibody for MAC-PD as an index test. Along with a commercialized kit, Capilia MAC Ab ELISA (TAUNS, Shizuoka, Japan), in-house assays were also allowed. Sensitivity and specificity were evaluated using a cutoff value of 0.7 U/mL 27 .
Reference test. Reference diagnoses based on ATS/IDSA 2007 statement and American Thoracic Society 1997 statement were preferred 4,28 . However, other clinical diagnoses were also accepted.
Outcomes. After extracting the number of subjects with true positives/false negatives/false positives/true negatives, we composed a two-by-two contingency table. Overall diagnostic test accuracy was assessed using a diagnostic odds ratio (DOR), and the area under the hierarchical summary receiver operating characteristic (HSROC) curves (AUC). The summary estimates of sensitivity and specificity, positive likelihood ratio (PLR), and negative likelihood ratio (NLR) were also assessed. Positive predictive value and negative predictive value were obtained across the pretest probability ranging from 0% to 100%.
Literature search strategy. In the electronic search, we systematically searched Pubmed, EMBASE, the Cochrane Library, and Web of Science on February 18 th , 2016. We used the following search formula for Pubmed without limitation: ((Mycobacterium avium complex) OR (Mycobacterium avium-complex) OR MAC OR MAC-PD OR (non-tuberculosis Mycobacterium)) AND (glycopeptidolipid OR anti-glycopeptidolipid OR anti-glycopeptidolipid-core OR GPL OR anti-GPL OR Capilia OR tauns OR (EIA kit) OR (ELISA kit) OR (enzyme immunoassay kit)) AND (sensitivity OR specificity OR "predictive value" OR "likelihood" OR "true positive" OR "true negative" OR "false positive" OR "false negative" OR diagnostic OR diagnosis). We used similar search formulas for EMBASE, the Cochrane Library, and Web of Science (Supplementary Text 1).
We hand-searched published reviews and included original studies.
Study selection. Two investigators (YS, NH) independently screened the candidate reports by reading only the title and abstract. Then, the two investigators independently scrutinized the full text of reports that had not been excluded by at least one investigator. Duplicate use of the same data was carefully assessed. The final inclusion was determined by discussion between the two investigators.
Data extraction. The two investigators (YS, NH) independently extracted the data from the original studies.
These data were cross-checked. If necessary, an investigator (NH) tried to contact the author of original reports by e-mail.
Quality assessment for bias and applicability. The two investigators (YS, NH) independently scored the seven domains of the Revised Tool for the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) 26 . When a study might include definite-MAC patients under treatment, the study was scored for high risk of bias concerning patient selection. The final score was determined after discussion between the two investigators. A study that had no domain with a high risk of bias and no domain with high applicability concerns was regarded as a high-quality study.
Statistical analysis and quantitative synthesis. Data synthesis and interpretation. We obtained the DOR using a DerSimonian-Laird random-model and the AUC using Holling's proportional hazard model. Based on Jones' criteria 29 , we interpreted AUC > 0.97, 0.93-0.96, 0.75-0.92, and 0.5-0.75 as "excellent, " "very good, " "good, " and "reasonable, " respectively. We obtained a paired forest plot, HSROC curve, and the summary estimate of the sensitivity and the specificity using the bivariate model 24 . PLR and NLR were calculated from the summary estimates of sensitivity and specificity. Following Grimes's criteria 30 , we interpreted PLR in the range of 2-5, 5-10, and > 10 as representing small, moderate, and large increases of probability when the index test was positive. We also interpreted the NLR in the range of 0. summary estimate of sensitivity and specificity were presented as variables depending on the pretest probability of MAC-PD ranging from 0% to 100%.
Sensitivity analysis. As part of the sensitivity analysis, subgroup analysis was conducted focusing on reports that evaluated Capilia MAC Ab ELISA (cutoff: 0.70 U/mL), high-quality reports, and reports evaluating RA cases 5,26 .
Heterogeneity. We evaluated the heterogeneity using I 2 statistics. I 2 < 40% is usually considered "not important" heterogeneity 31 .
Software. We used commands of the statistics software R as follows: "madauni" command for DOR, "phm" command for AUC, and "reitsma" command for the HSROC curve and the summary estimates of sensitivity and specificity 32,33 .

Study search and study characteristics.
Of the 257 articles that we had found through primary search, 86, 122, and 33 were excluded through removal of duplication, screening, and full-article reading, respectively ( Fig. 1). Notably, Kitada et al. reported a number of reports from a hospital, some of which were excluded due to a possible overlap of included subjects. Our hand search found no eligible article. We finally included 16 reports, comprising 10 full-length articles and six conference abstracts, all of which were written in English (Table 1) [7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22] . We obtained non-published data concerning four reports through e-mail communication with the authors [16][17][18][19] . Among the 16 reports, 13 were from Japan and one each was from of Korea, the USA, and Taiwan. The Capilia kit was used in 14 studies. The cutoff value of 0.70 U/mL was used in 15 studies. The diagnostic criteria of ATS/IDSA 2007 were used in 10 studies. Four studies used the one-gate cohort approach, all of which were regarded as high-quality studies (Supplementary Figure 1). Three studies included only subjects with rheumatoid arthritis (RA).
The number of participants in each study ranged from 18 to 906 with a median of 143. The total number of subjects was 3368. This total consisted of 1098 reference positive subjects and 2270 reference negative subjects. Across the 16 studies, the sensitivity ranged from 0.20 to 1 with a median of 0.75, and the specificity ranged from 0.33 to 1 with a median of 0.92 (Fig. 2).
Sensitivity and specificity. The summary estimates of sensitivity and specificity were calculated from 14 studies that used the Capilia kit with the cutoff value of 0.7 U/mL. The summary estimates of sensitivity and specificity were 0.696 (95% CI 0.621-0.761) and 0.906 (95% CI 0.836-0.951), respectively. (Table 2 and Fig. 3B) According to a sensitivity analysis using four high-quality studies, the summary estimates of sensitivity and specificity were 0.646 (95% CI 0.519-0.756) and 0.918 (95% CI 0.706-0.981), respectively ( Table 2 and Fig. 3C). The summary estimates sensitivity of 0.790 (95% CI 0.301-0.971) and specificity of 0.979 (95% CI 0.873-0.997) were obtained from three reports including only RA cases.
Positive and negative likelihood ratios. Based on the summary estimate of sensitivity and specificity calculated from 14 reports using the Capilia kit with the cutoff value of 0.7 U/mL, PLR and NLR were 7.4 (95% CI 4.1-13.8) and 0.34 (95% CI 0.26-0.43). According to Grimes' criteria, a positive index test moderately increases the probability of MAC-PC, while a negative index test suggests a small decrease of the probability ( Table 2 and Fig. 4).
Positive and negative predictive values. We calculated PPV and NPV from the summary estimate of sensitivity and specificity calculated from 14 reports using the Capilia kit with the cutoff value of 0.7 U/mL, i.e. a sensitivity of 0.696 and a specificity of 0.906 (Fig. 4). This figure indicates that the positive test has a stronger impact than the negative test.
Positive rate in non-MAC-PD subjects. Some included studies provided data concerning the positive rate in non-MAC-PD subjects. The anti-GPL-core IgA antibody was positive in 27-100% of RGM cases, while the test was negative in 93-100% of pulmonary-healthy controls, 90-100% of M. tuberculosis cases, and 97-100% of non-MAC non-RGM NTM-PD cases ( Table 3). The pooled positive rate for rapid growing Mycobacterium was 0.64 (95% CI 0.27-1.00) (Supplementary Figure 2).

Discussion
To the best of our knowledge, the current report is the first systematic review and meta-analysis evaluating the diagnostic test accuracy of anti-GPL-core IgA antibody for MAC-PD. We think the results from our study are Scientific RepoRts | 6:29325 | DOI: 10.1038/srep29325 robust for various reasons. First, we included a sufficient number of studies and subjects. Second, we used the recently recommended hierarchical meta-analysis approach. Third, across all analyses, the observed heterogeneities were not important (< 40%). Fourth, a sensitivity analysis revealed consistent results.    M. abcessus and other RGM commonly have GPL antigen in their cell walls 5 . Thus, the anti-GPL-core antibody assay intrinsically has cross-reactivity for MAC and RGM. This is a serious pitfall when interpreting the results of the anti-GPL-core antibody assay because different antibiotic regimens should be prescribed to treat MAC and RGM 3,4 . When the anti-GPL IgA antibody is positive, we should carefully consider the possibility of RGM infection, especially M. abcessus 9 . While pulmonary imaging studies for MAC-PD commonly present multiple discrete pulmonary nodules 34 , common CT findings of RGM-PD are bilateral bronchiectasis, bronchiolitis, and upper lobe cavities 35 . Second, the universally used diagnostic criteria by ATS/IDSA 2007 statement, which is simpler than the previous American Thoracic Society 1997 statement, are still demanding 4 . The ATS/IDSA 2007 diagnostic criteria overlook some of true MAC-PD population. Although the culture isolation of the microbe has been the gold standard diagnostic method of infectious disease, the sero-diagnosis by anti-GPL-core IgA antibody, for example combining one MAC culture positive plus anti-GPL-core IgA antibody positive, can be used for diagnostic criteria 5 . Kitada et al. proposed a cutoff value for the anti-GPL-core IgA antibody of 0.7 U/mL 27 . This is because this cutoff can provide high specificity, though it is not sensitive. The summary estimate of sensitivity was 69% in our analysis (Table 2), which means that almost a third of MAC-PD cases were overlooked by the anti-GPL-core IgA antibody assay. The serum enzyme immunoassay requires a sufficient quantity of targeted antibody to identify the MAC-infected individual. However, non-extensive pulmonary lesions, especially solitary nodules, and an immunocompromised state, namely an HIV-infected state and use of immunosuppressants, lead to a low serum antibody level 13 .

Figure 1. Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow chart for study search.
We need to discuss the limitations of our study. First, most of the included studies had a high risk of bias due to the two-gate study design. Because the prevalence of MAC is not high, it is difficult to recruit a large number of patients with MAC in a cohort study. Nonetheless, we believe the results of our analysis are reliable. This is because the high-quality studies based on sensitivity subgroup analysis that included only one-gate studies yielded consistent results (Table 2 and Fig. 3C). Second, most of the included studies were from Japan and only three were from other countries (Table 1). To reconfirm the diagnostic test accuracy of the sero-diagnosis for MAC, Kitada et al. evaluated 152 subjects in the USA and revealed that the sensitivity and specificity were 70.1% and 93.9%, respectively 12 . Third, subgroup analysis focusing on RA cases revealed higher accuracy than results from overall reports ( Table 2 and Fig. 3D). However, the analysis was not conclusive due to the lack of a   plausible biological explanation and the limited number of included reports and subjects. Fourth, it is not likely that through checkup including BALF was done for every non-MAC persons to completely deny the possibility of MAC due to ethical regulation.
In conclusion, we conducted the first systematic review and meta-analysis concerning the diagnostic test accuracy of anti-GPL-core IgA antibody for MAC-PD. According to our analysis using data of 16 reports and 3368 subjects. The DOR was 24.8 and the AUC was 0.873. The summary estimates of sensitivity and specificity were 0.691 and 0.919, respectively. Considering the demanding clinical diagnostic criteria of the ATS/IDSA 2007 statement, the true specificity of the anti-GPL-core IgA antibody may be higher. A positive index test moderately increases the probability of MAC-PC, while a negative index test suggests a small decrease of the probability. This test is useful to ruling in MAC-PD.