Performance of Purified Antigens for Serodiagnosis of Pulmonary Tuberculosis: a Meta-Analysis

Study selection.Initially, two reviewers independently screened citations retrieved from all sources for relevance. Screening of full-text articles by using prespecified inclusion criteria was carried out by two reviewers, and the articles included were checked by a third reviewer. Disagreements were resolved by consensus.

Data extraction.A data extraction form was created and pilot tested with a subset of eligible studies and then finalized. Two reviewers (each of whom was responsible for approximately 50% of the studies) extracted data from all included studies with the standardized form. To verify reproducibility, a third reviewer independently performed data extraction on all studies. Differences among reviewers were resolved by consensus. When necessary, authors were contacted for additional information.

Assessment of study quality.The quality of studies was appraised by using a subset of criteria from QUADAS, a validated tool for diagnostic studies (see Table S1 in the supplemental material) (110).

Antigen classification.Antigens were classified into five categories according to the type of compound: (i) recombinant proteins, (ii) native proteins, (iii) lipids, (iv) multiple antigens (protein-protein or lipid-lipid additive reactivity), and (v) protein-lipid antigens. Several investigators evaluated antibody responses to multiple antigens in the same patient population to enhance sensitivity. These studies have taken two approaches. In some cases, different antigens (or portions thereof) have been cloned as single protein entities (polyproteins) and tested for their reactivities with sera. In other cases, multiple antigens have been tested as single entities and cumulative results (additive reactivity) were calculated. In the former case, we considered polyproteins to be single antigens; in the latter case, we classified the entities as multiple antigens.

Data analysis.Estimates of sensitivity and specificity from individual studies and their exact 95% confidence intervals (CIs) were obtained by using Meta-DiSc (version 1.4) software (117). Sensitivity refers to the proportion of TB patients with positive test results; specificity refers to the proportion of participants without TB with negative test results. For sensitivity, we included studies that used sputum smear as the reference standard along with studies that used culture. For specificity, we noted the type of comparison group, e.g., healthy participants or patients with nontuberculous respiratory disease. Likelihood ratio positive was calculated as sensitivity/(1 − specificity); likelihood ratio negative was calculated as (1 − sensitivity)/specificity.

Selection of subgroups for meta-analysis.We recognized that studies were heterogeneous in many respects, particularly concerning antigen characteristics and antibody class. Therefore, in order to address heterogeneity and combine study results, subgroups of comparable antigens were determined (Fig. 1). Initially, studies were grouped by the class of antibody detected by the test: (i) IgG and/or IgA, (ii) IgM, and (iii) other IgM-containing combinations (IgM-IgG, IgM-IgA, and IgM-IgG-IgA). This division was based on the understanding that IgM antibodies are expressed transiently and earlier in infection than other antibodies. Next, studies were stratified by antigen number (single or multiple antigens); the type of compound (protein or lipid); and, for proteins, the source of the compound (recombinant or native). Finally, for each distinct single antigen or multiple-antigen combination, studies were stratified by patient sputum smear status and HIV status. At least four studies were required to be available for inclusion in a subgroup in order to strengthen the results and reduce the possibility of finding a significant result by chance. In this way, we identified 16 subgroups.

• Open in new tab
• Download powerpoint

FIG. 1.

Flow diagram for selection of subgroups, IgG and/or IgA antibody detection: an example with MPT51. The same sequence of steps was repeated for each antigen. Having at least four studies available was a condition for inclusion in the meta-analysis. *, other antibody combinations included IgG and IgM (n = 3 studies); IgM and IgA (n = 0); IgG, IgA, and IgM (n = 7); and not reported (n = 10); **, other recombinant antigens included 38 kDa (n = 13 studies), CFP-10 (n = 9), malate synthase (n = 8), TbF6 (n = 4), α-cystallin (n = 4), Mtb48 (n = 3), Ag85C (n = 2), DPEP (n = 2), ESAT6 (n = 2), and other antigens (n = 16) that appeared in only one study each.

To summarize sensitivity and specificity within each subgroup, separate meta-analyses were performed by using the hierarchical summary receiver operating characteristic curve model (72). The advantages of the hierarchical summary receiver operating characteristic are that it jointly models both sensitivity and specificity, weights studies according to the number of participants, and takes into account unmeasured heterogeneity between studies by using random effects (31). The model was estimated by using a Bayesian approach with noninformative prior distributions and was implemented with WinBUGS (version 1.4.1) software program (91).

The average sensitivity, specificity, and likelihood ratios from each meta-analysis were estimated. From the posterior distribution of each parameter of interest, we extracted the mean and the 95% credible interval (the Bayesian equivalent of the classical confidence interval) on the basis of the 2.5% and 97.5% quartiles. When feasible, specificity estimates were stratified by type of comparison group. Finally, a summary receiver operating characteristic (SROC) curve from each meta-analysis was obtained. The SROC curve plots sensitivity versus 1 − specificity for the range of specificity values observed for each study, as extrapolation beyond this range is not advisable (42). The SROC curve gives an idea of the overall performance of a test across different thresholds (54, 61). The closer that the curve is to the upper-left-hand corner of the plot (sensitivity and specificity are both 100%), the better the performance of the test is (42). The plots were made by using the R (version 2.6.1) software program (70).

Descriptive analysis.Descriptive analyses were performed by using SPSS (version 14.0.1.366) software (92). Forest plots were made by using Meta-DiSc (version 1.4) software (117).

Description of studies included.The literature searches identified over 5,000 citations, of which 49 publications (254 studies) were included (Fig. 2) (2-4, 6, 8, 10, 12, 15, 18-20, 23, 24, 26, 27, 29, 33-37, 39, 43, 44, 55, 58, 59, 66-69, 73, 76, 77, 81-87, 99, 101-103, 105, 108, 109, 118). Mycobacterial culture was used as the reference standard in 199 (78%) studies, sputum smear was used as the reference standard in 29 (11%) studies, and sputum smear and/or culture was used as the reference standard in 26 (10%) studies. Two hundred thirteen (84%) studies involved smear-positive patients, and 41 (16%) involved culture-confirmed smear-negative patients. Four studies involved children younger than 15 years of age, and 30 (12%) studies involved HIV-infected persons. The vast majority (96%) of studies performed antibody tests by ELISA. The median number of participants with TB was 51 (interquartile range, 39 to 105); the median number of participants without TB was 57 (interquartile range, 35 to 83).

• Open in new tab
• Download powerpoint

FIG. 2.

Flow of studies through the review process. PTB, pulmonary tuberculosis.

Two hundred fifty-four studies evaluating 51 distinct single antigens (9 native proteins, 27 recombinant proteins, and 15 lipids) and 30 distinct multiple-antigen combinations were identified. Many of these antigens were evaluated in only one study. In order to accommodate the large number of antigens identified in the review, only those antigens appearing in two or more studies are included in Tables S2 though S6 in the supplemental material. A guide to the tables and figures is presented in Table 1. The antigens and their alternative names are listed in Table 2. The most frequently evaluated antigens are described below and in additional tables and figures, as noted.

Assessment of study quality.The majority of studies used a case-control study design. Only 65 (26%) studies reported blinded interpretation of index test results. Almost all studies provided sufficient detail describing the execution of the index test (Table 3).

• Open in new tab
• Download powerpoint

FIG. 3.

SROC curves of antigen performance for serodiagnosis of pulmonary TB. (A) Recombinant proteins; (B) native proteins; (C) lipids.

38 kDa, a major protein present in culture filtrates of M. tuberculosis, has been studied extensively (1, 17, 22). Several studies have shown an association between the presence of anti-38 kDa antibodies and advanced cavitary TB (14, 22, 75). In smear-positive patients, recombinant 38 kDa yielded a sensitivity of 47% (95% CI, 39 to 55) and a specificity of 94% (95% CI, 86 to 98) (8, 27, 33, 35, 37, 55, 77, 81).

Malate synthase (81 kDa), present in M. tuberculosis culture filtrates, the cell wall, and cytoplasmic subcellular fractions, is an enzyme of the glyoxylate pathway used by M. tuberculosis during intracellular replication in macrophages (90) and has adapted to function as an adhesin that enhances bacterial adherence to host cells (46). In sputum smear-positive patients, malate synthase achieved a sensitivity of 73% (95% CI, 58 to 85) and a specificity of 98% (95% CI, 95 to 100) (35, 37, 85, 86, 108). The likelihood ratio positive (40.78) was considerably higher for malate synthase than for other antigens. Earlier studies have demonstrated that whereas antibodies to the 38-kDa antigen are present in patients with extensive cavitary lesions, anti-malate synthase antibodies are elicited earlier during the progression of TB, being present in patients who have not yet developed cavities (74). This is reflected in the higher sensitivity of TB diagnosis provided by malate synthase.

(c) Recombinant MPT51 (Rv3803c) (Table A2).The 27-kDa protein MPT51, a culture filtrate protein, is closely related to the antigen 85 (Ag85) complex, which comprises Ag85A, Ag85B, and Ag85C. MPT51 is an adhesin (108) reported to be a fibronectin-binding protein of M. tuberculosis (113). A sufficient number of studies evaluating the performance of MPT51 were available to stratify the results by HIV infection status. In sputum smear-positive patients, MPT51 provided equivalent sensitivities in both HIV-negative TB patients (59%; 95% CI, 38 to 76) and HIV-positive TB patients (58%; 95% CI, 30 to 82); the specificities were 94% and 97%, respectively (10, 69, 85, 108).

Culture filtrate protein 10 (CFP-10), a culture filtrate and cell wall protein, has been identified as one of the earliest proteins expressed by M. tuberculosis during culture in bacteriological media (9). In sputum smear-positive patients, CFP-10 provided a sensitivity of 48% (95% CI, 29 to 68) and a specificity of 96% (95% CI, 83 to 99) (27, 33, 59, 118).

(e) Recombinant TbF6 (see Table S2 in the supplemental material).TbF6 is a single antigen combining four distinct antigens (CFP-10, MTB8, MTB48, and 38 kDa) as a genetically fused polyprotein (37). In sputum-smear positive patients, TbF6 achieved a sensitivity of 70% (95% CI, 37 to 90) and a specificity of 93% (95% CI, 69 to 99) (6, 37). The high sensitivity obtained with TbF6 is likely due to the fact that it comprises immunogenic domains from multiple antigens.

(ii) Native proteins. (a) Native 38 kDa (Rv0934) (Table A3).In sputum smear-positive patients, native 38 kDa provided a sensitivity of 49% (95% CI, 37 to 61). In sputum smear-negative patients, the sensitivity reported was lower (31%; 95% CI, 15 to 52). Specificities were 97% in both subgroups (15, 18, 59, 69, 76, 77, 102).

(b) Native Ag85B (Rv1886c) (Table A4).Ag85B, present in M. tuberculosis culture filtrates and cell walls, is a major component of the Ag85 complex (112). Like MPT51, Ag85B is a fibronectin-binding protein (113). In HIV-negative TB patients, native Ag85B yielded a sensitivity of 53% (95% CI, 20 to 83), and in HIV-positive TB patients, a it yielded a sensitivity of 62% (95% CI, 19 to 92). The specificities were ≥95% (23, 67, 77, 103).

(c) Native α-crystallin (2031c) (see Table S3 in the supplemental material).α-Crystallin is a 14/16-kDa cell wall protein (106) shown to be induced in bacteria under hypoxia (78). In sputum smear-positive patients, α-crystallin provided a sensitivity of 48% (95% CI, 29 to 68) and a specificity of 96% (95% CI, 83 to 99) (66, 103).

(iii) Lipids.A variety of lipid-containing antigens are common to mycobacterial species (71). Several lipid moieties have been purified and intensely studied for their serological potential for TB diagnosis (44). Four lipid antigens, all acylated trehaloses (107), were evaluated in smear-positive patients and included in the meta-analysis.

2,3-Diacyltrehalose (DAT), a component of the M. tuberculosis cell wall, has been postulated to play a role in modulating host immune responses (50). DAT yielded a sensitivity of 63% (95% CI, 45 to 78) and a specificity of 81% (95% CI, 50 to 96) (43, 44, 83, 105).

(b) TAT (see Table S4 in the supplemental material).2,3,6-Triacyltrehalose (TAT) is an antigenic glycolipid compound found in the M. tuberculosis cell wall (41, 43). TAT provided a sensitivity of 81% (95% CI, 21 to 99) and a specificity of 44% (95 CI, 24 to 67) (43, 44).

(c) SL-I (see Table S4 in the supplemental material).2,3,6,6′-Tetraacyltrehalose 2′-sulfate (sulfolipid I [SL-I]), a compound found abundantly in the M. tuberculosis cell wall, may affect the human immune system and play a role in the pathogenesis of TB (51). SL-I yielded a sensitivity of 80% (95% CI, 56 to 93) and a specificity of 59% (95% CI, 8 to 96) (43, 44).

(d) Cord factor (see Table S4 in the supplemental material).Cord factor (trehalose 6,6′-dimycolate), a major component of M. tuberculosis cell walls, is named for its central role in aggregating mycobacteria into cord structures (7, 30). Cord factor may contribute to the virulence of M. tuberculosis by facilitating cavity formation (38). Cord factor achieved a sensitivity of 69% (95% CI, 28 to 94) and a specificity of 91% (95% CI, 78 to 97) (43, 44, 101).

(e) TbF6 plus DPEP (see Table S5 in the supplemental material).TbF6 polyprotein plus DPEP was the multiple-antigen combination most frequently evaluated; four studies involved HIV-uninfected individuals, and one study involved HIV-infected individuals. As described above, TbF6 is a polyprotein. DPEP, also known as MPT32, is a proline-rich 45/47-kDa antigen suggested to have a role in the cross-linking of molecules produced by or bordering M. tuberculosis (49). Earlier studies with native MPT32 have demonstrated that it is a highly immunogenic protein that provided higher sensitivity than the 38-kDa protein when it was tested in the same patient cohort (74). In HIV-negative TB patients, TbF6 plus DPEP achieved a sensitivity of 75% (95% CI, 50 to 91) and a specificity of 94% (95% CI, 86 to 99) (6, 37). The single study evaluating the serodiagnostic potential of TbF6 plus DPEP in HIV-infected individuals is described below.

Sufficient numbers of studies evaluated five antigens, four proteins (recombinant 38 kDa, native 38 kDa, malate synthase, and CFP-10) and one lipid (DAT) for comparison of the specificities for healthy and diseased controls. For the four proteins, both subsets showed similar specificity values. For DAT, studies involving patients with nontuberculous respiratory disease yielded a significantly higher specificity, 57% (95% CI, 30 to 76), than studies with healthy volunteers, 97% (95% CI, 88 to 100).

Overall performance of antigens (Fig. 3).Among recombinant proteins, malate synthase and TbF6 plus DPEP (multiple antigen) provided the highest sensitivities for the specificities reported (Fig. 3A). For native proteins, Ag85B in HIV-infected TB patients achieved better performance than other native antigens (Fig. 3B). Among lipid antigens, cord factor had the best performance (Fig. 3C).

• Open in new tab
• Download powerpoint

FIG. 4.

Performance of antigens for the serodiagnosis of pulmonary TB in HIV-infected patients. (A) Sensitivity; (B) specificity. The circles and the lines represent the point estimates and the 95% CIs, respectively. The size of the circle indicates the study size. MS, malate synthase; n, native; r, recombinant; 1, IgG; 2, IgM; 3, IgA; 4, IgG/A; 5, IgG/IgA/IgM.

Thirty studies evaluating antigens (all proteins) in HIV-infected TB patients were identified; all studies involved sputum smear-positive patients. Of the total, 23 (77%) studies evaluated assays for the detection of IgG and/or IgA antibodies (23, 35, 37, 69, 103, 108), four studies evaluated assays for the detection of IgM antibodies (69, 103), and three studies evaluated assays for the detection of IgG plus IgA plus IgM antibodies (69, 103). Four studies based on multiple-antigen combinations are described in more detail below (35, 37, 108).

Antibodies to all five single antigens (38 kDa, malate synthase, Ag85B, α-crystallin, and recombinant MPT51) evaluated in these studies were detected. As discussed, only MPT51 and Ag85B were investigated in a sufficient number of studies for inclusion in the meta-analysis (Table 4). With assays for the detection of IgG and/or IgA antibodies, the sensitivities reported for 38 kDa (range, 35% to 68%) and α-crystallin (range, 15% to 58%) were similar to those provided for MPT51 and Ag85B. Malate synthase achieved higher sensitivities (range, 73% to 92%). Compared with tests for the detection of only IgG and/or IgA antibodies, tests for the detection of IgM antibodies provided considerably lower sensitivities (range, 4% to 5%). The inclusion of tests for the detection of IgM (IgG plus IgA plus IgM) did not appreciably increase the sensitivity. The specificities provided by all of the above antigens were high (range, 89% to 100%). However, only 6 (20%) studies involved controls with nontuberculous respiratory disease (23, 35, 37), while 14 (47%) studies involved either healthy volunteers or asymptomatic HIV-infected individuals without TB (35, 37, 103, 108). In 10 (33%) studies, the control group involved HIV-infected individuals whose clinical status ranged from asymptomatic to symptomatic with opportunistic infections other than TB (69).

(ii) Performance of tests with multiple antigens (see Table S5 in the supplemental material).Assays based on multiple antigens provided higher sensitivities (median, 76%; range, 16% to 96% [57 studies]) than assays based on single antigens (median, 53%; range, 2% to 100% [197 studies]), while they maintained high specificities (median, 96%; range, 79% to 100%) (data not shown). The combination of malate synthase plus MPT51 was evaluated in three studies, two studies involving HIV-negative TB patients (2, 108) and one study involving HIV-infected TB patients (108). The sensitivities provided by malate synthase plus MPT51 were similar with HIV-uninfected and -infected TB patients from India: 80% (95% CI, 73 to 87) and 77% (95% CI, 56 to 91), respectively. The specificities were equivalent (97%) whether the comparison group involved HIV-negative healthy volunteers from India or HIV-infected (tuberculin skin test-positive and -negative) asymptomatic individuals from the United States (108). However, this antigen combination yielded a sensitivity of only 55% (95% CI, 36 to 72) with TB patients from the United States (2).

The combination of TbF6 plus DPEP plus malate synthase achieved a sensitivity of approximately 85% with both HIV-negative and HIV-infected TB patients (37). The specificities were high (97%; 95% CI, 83 to 100), even when this antigen combination was evaluated with patients with nontuberculous respiratory disease. In studies in which 38 kDa plus malate synthase was evaluated, the sensitivities reported were 71% (95% CI, 57 to 83) for HIV-negative TB patients and 96% (95% CI, 80 to 100) for HIV-infected TB patients; the specificity was 89% (95% CI, 78 to 96) when it was assessed with healthy volunteers (35). With HIV-negative, sputum smear-positive patients, the combination of 38 kDa plus Ag85B and α-crystallin achieved a sensitivity of 89% (95% CI, 84 to 93) and, with the addition of MPT51, a sensitivity of 91% (95% CI, 86 to 95) (68). With non-HIV-infected, sputum smear-negative patients, the two combinations provided sensitivities of 73% (95% CI, 57 to 86) and 78% (95% CI, 62 to 89), respectively, and a specificity of 87% (95% CI, 75 to 94) when they were assessed with patients with nontuberculous respiratory diseases (68). Only two studies with multiple lipid antigens were identified (see Table S6 in the supplemental material).

Stratification by Ig class showed that in comparison with the results of studies that detected antibodies to IgG (median sensitivity, 61%; range, 8% to 100%) or IgA (median sensitivity, 40%; range, 10% to 90%), studies that detected antibodies to IgM had considerably lower sensitivities (median, 11%; range, 2% to 71%). The median specificities were similar: 96%, 96%, and 98%, respectively. In addition, compared with the results of tests that detected only anti-IgG or anti-IgA antibodies, tests that detected IgG plus IgA showed higher sensitivities (median, 71%; range, 43% to 97%). The inclusion of IgM (IgG plus IgA plus IgM) did not further enhance the sensitivity (median, 71%; range, 60% to 83%).

Principal findings.This systematic review yielded 254 studies evaluating 51 distinct single antigens and 30 multiple-antigen combinations. The performance of these antigens was examined in in-house tests for the serodiagnosis of pulmonary TB. Studies evaluating 13 distinct antigens (recombinant 38 kDa, native 38 kDa, MPT51, malate synthase, CFP-10, TbF6 polyprotein, Ag85B, α-crystallin, DAT, TAT, SL-I, cord factor, and TbF6 plus DPEP [multiple antigen]) were included in the meta-analysis. The results demonstrate that (i) in sputum smear-positive patients, only recombinant malate synthase (sensitivity, 73%; 95% CI, 58 to 85) and TbF6 plus DPEP (sensitivity, 75%; 95% CI, 50 to 91) provided sensitivities significantly ≥50%; (ii) all protein antigens achieved high specificities; (iii) among the lipid antigens, cord factor had the best overall performance (sensitivity, 69% [95% CI, 28 to 94]; specificity, 91% [95% CI, 78 to 97]); (iv) compared with single antigens (median sensitivity, 53%; range, 2% to 100%), multiple antigens yielded higher sensitivities (median sensitivity, 76%; range, 16% to 96%); (v) in HIV-infected patients who are sputum smear positive, antibodies to several single and multiple antigens were detected; and (vi) data on seroreactivity to specific antigens in sputum smear-negative or pediatric patients were insufficient. These results demonstrate that no single antigen provides a sensitivity that is sufficient for a single antigen to be used to devise a serodiagnostic test for TB and that it is unlikely that a single antigen-based serodiagnostic test can be devised. This is not surprising, since the titers of antibodies to each antigen would differ in individuals and the detection of low titers of antibodies would be occluded due to the formation of immune complexes. It is also interesting that while both DPEP and malate synthase are conserved in the M. tuberculosis complex species and in all clinical isolates of M. tuberculosis whose genomes have been sequenced, antibodies to these antigens are not detected in a vast majority of tuberculin skin test-positive individuals with likely latent infection. Proteins that are approximately 50 to 60% homologous to these antigens are present in some other mycobacteria and whether cross-reactive antibodies exist in nontuberculous mycobacterial diseases remains to be reported.

Stratification by Ig class demonstrated that assays for the detection of IgG and/or IgA antibodies provided higher sensitivities than assays for the detection of IgM antibodies. This is not surprising, since IgM antibodies are likely to be expressed early during the onset of infection, with the levels quickly decreasing after this period. By the time that bacteriologically detectable TB manifests, whether it is during primary infection or reactivation, the infection has already progressed for months to years in immunocompetent individuals and weeks to months in immunocompromised patients. Thus, the detection of IgM antibodies may have a role in the identification of early infection, but its value for the serodiagnosis of active TB disease may be limited. Considering that the profile of antigens recognized by antibodies is altered with the progression of M. tuberculosis infection (74), the antigens used in a serodiagnostic test during contact tracing are likely to differ from those used in a test for the diagnosis of clinical TB. To our knowledge, no antigens that can be the basis for an accurate serodiagnostic test for contact tracing have been reported. The discovery, evaluation, and comparison of such tests with gamma interferon release assays need to be considered.

This systematic review and meta-analysis had several strengths. Standard protocols for the conduct of the review (61) and assessment of the quality of the studies (110) were followed. The application of a comprehensive search strategy with various overlapping approaches enabled the retrieval of relevant studies published since 1990. Screening and data extraction were performed independently among three reviewers, and authors were contacted to clarify points and obtain missing data. None of the studies in the review used the result from the antibody test as a reference to confirm TB (incorporation bias). When possible, patients with disease were selected, in preference to healthy controls, to evaluate the performance of antigens with persons in whom TB was initially suspected and subsequently ruled out.

The meta-analysis was limited by the relatively small number of studies investigating the same antigens or antigen combinations. The small number of comparable antigens made it difficult to relate study quality to antigen performance. However, several important deficiencies in study design and quality were noted. Only 20 (7%) studies recruited participants in a random or consecutive manner. Therefore, most studies lacked a sound probabilistic sampling framework. The majority of studies used a case-control design with healthy controls. This design has been found to overestimate test sensitivity and specificity (53, 111), although for the four protein antigens in this review for which a comparison was feasible, the specificities were found to be similar with healthy and diseased controls. Few (26%) studies reported the use of the blinded interpretation of test results and a reference standard. This was not unexpected, since the primary aims of in-house studies are the discovery of novel antigens, evaluation of their diagnostic potential, and/or comparison of different antigens. Nonetheless, the lack of blinding and the dearth of data from cross-sectional studies are major shortcomings of the currently available literature and may have resulted in an overestimation of antigen performance (53). Both errors in design and deficiencies in reporting have been noted as concerns in TB diagnostic studies (62, 89).

An additional limitation was the lack of information about sputum microscopy procedures. Smear status may be determined through the examination of unprocessed sputum (direct smear microscopy) or through the more sensitive examination of sputum after its digestion and concentration prior to the inoculation of cultures (concentrated smear microscopy). The former procedure is more common in low- and middle-income countries, where cultures are rarely done, while the latter procedure is more common in high-income countries. Furthermore, fluorescence microscopy, which is commonly used in high-income countries, is associated with a sensitivity higher than that achieved by conventional light microscopy, which is commonly used in low-income countries (96). Conversely, it is widely appreciated that proficiency in sputum smear microscopy requires regular exposure to positive smears, and this experience is more likely to be gained in low- and middle-income countries where TB is endemic. Thus, the site where smear microscopy was performed may have had an undeterminable impact on the assessment of antigen performance distinct from the nature of the patient population. Another limitation may be that the patient population differed between richer and poorer countries, with persons in the latter having more advanced disease and perhaps being more likely to be infected with HIV. In this review, approximately 75% of the studies involved patients who resided in low-income countries. The maximum benefit of improved diagnosis through serology would manifest in these countries. However, the findings of this review suggest that the assays evaluated could not replace sputum microscopy in these countries.

In TB diagnostic trials, culture is considered the “gold standard.” A majority (78%) of studies used culture as the reference standard. Along with studies that used culture, we also included studies from countries where TB is endemic that used sputum smear microscopy, a test with modest sensitivity. The use of an insensitive reference standard may have led to biased estimates of antigen performance (111). Our choice of reference standard (culture and/or smear) may have limited the inclusion of studies involving children (four studies). Pediatric TB is difficult to diagnose on a bacteriological basis because of the paucibacillary nature of the disease (79). Although statistical tests and graphical methods for the detection of potential publication bias in meta-analyses of randomized control trials are available, to our knowledge such techniques have not been adequately evaluated for diagnostic data (98). It is therefore difficult to rule out publication bias in our review. In addition, our search strategy may have missed some relevant studies by excluding non-English-language publications (28% of the citations initially identified).

Conclusion.In summary, despite the limitations discussed above, this systematic review and meta-analysis has identified potential candidate antigens for inclusion in an antibody detection-based diagnostic test for pulmonary TB in HIV-infected and -uninfected individuals. However, none of the antigens achieves sufficient sensitivity to replace sputum smear microscopy. Combinations of select antigens provide higher sensitivities than single antigens. Our findings should be interpreted in the context of the availability and the variability in the design and the quality of studies of antigen performance. Clearly, unpublished studies did not feature in this review, and a number of published studies did not meet the eligibility criteria for inclusion.

Recently, the number of systematic reviews of TB diagnostic tests has been growing (63). The current review adds to this expanding evidence base and may offer guidance to researchers, test developers, and grant-awarding bodies on directions for future research. Activities leading to the evaluation of combinations of existing potential candidate antigens, as well as the discovery of additional antigens with diagnostic potential that complement current antigens, need to be intensified to devise assays that can improve upon microscopy. A focus on antigen discovery for the serodiagnosis of TB in smear-negative and pediatric patients should be encouraged. Biobanks (such as the TB Specimen Bank of the Special Programme for Research and Training in Tropical Diseases [TDR]) may be useful for the rapid evaluation of new antigen combinations prior to the consideration of the performance of field studies. Finally, research is also urgently needed to determine whether combinations of microscopy and antibody detection with multiple antigens could improve case finding in high-prevalence countries and whether antibody detection tests can be delivered in an appropriate format.

Efforts are needed to improve both the methodological quality and reporting of TB diagnostic studies. TDR has developed guidelines (Diagnostics Evaluation Expert Panel) that researchers can use to assess the performance and operational characteristics of diagnostics for infectious diseases (5). Use of the Standards for Reporting of Diagnostic Accuracy may also lead to improvements in the quality of studies (11). Future studies should evaluate outcomes that go beyond the conventional diagnostic accuracy of sensitivity and specificity. These outcomes include the accuracy of diagnostic algorithms (rather than single tests) and their relative contributions to the health care system, the incremental or added value of new tests, the impacts of new tests on clinical decision making and therapeutic choices, the cost-effectiveness of new tests in routine programmatic settings, and the impacts of new tests on patient-centered outcomes (63).