Introduction

Tuberculosis (TB) is a serious infectious disease caused by Mycobacterium tuberculosis (MTB). The Global Tuberculosis Report 2019 stated that in 2018, about 1.5 million people worldwide died of TB and nearly 10 million people died from MTB, of which only 6.4 million were diagnosed and officially reported. An estimated 1.7 billion people worldwide are infected with MTB (LTBI) during the incubation period without any obvious symptoms [1]. TB mainly damages the lungs, causing lung disease or pulmonary tuberculosis, but it can also damage other organs, causing bone tuberculosis, nerve tuberculosis, skin tuberculosis, kidney tuberculosis, and other infections [2].

The incubation period of TB is related to the immune status of the person, and there is no clinical, radiological, or microbiological evidence of active TB disease during the incubation period [3]. The typical symptoms of active TB are chronic cough, bloody sputum, night sweats, fever, and weight loss and various symptoms can be observed in extrapulmonary cases [4]. The conventional technique for detecting MTB in an analytical sample (such as pus, sputum, or tissue biopsy) takes two to 6 weeks. So far, for the rapid detection of MTB, many techniques have been developed, such as ELISA (enzyme-linked immunosorbent assay), real-time polymerase chain reaction (PCR), latex agglutination, Gen-Probe amplified M. Tuberculosis direct test, and flow cytometry [5]. Compared to traditional microbial culture techniques, these methods exhibit higher sensitivity in a shorter time, but this requires advanced laboratories and technicians, which is the main limitation of these methods. Therefore, it is essential to develop a real-time, portable, and sensitive technology that can quickly detect MTB at an affordable cost.

MPT64, which is a 24-kDa protein of MTB and an important secretory protein of pathogenic bacteria, is often used as a candidate protein for diagnosis and in vaccines [6, 7]. At present, there are many ways to detect the MPT64 protein, such as immunochromatography (ICT), ELISA, SD Bioline, and Capilia TB [8,9,10,11].

To date, many studies have evaluated the diagnostic accuracy of MPT64 for MTB. In 2013, a systematic review evaluated the diagnostic accuracy of commercial MPT64-based tests for MTB [12]. Our purpose was to evaluate the efficacy of MPT64 protein as a target for detection of Mycobacterium tuberculosis infection. What’s more, we also evaluated the diagnostic efficacy of three common commercial kits relying on MPT64 antigen assay. Our study was more comprehensively than the study by Yin et al [12]

Methods

Research identification and selection

Three independent reviewers (XJ Cao, YP Li, JY Wang) searched four online electronic databases up to July 15, 2020. The databases searched included Embase, Cochrane Library, PubMed, and Web of Science. Finally, we retrieved 1222 articles. After deleting the repetitive articles, 521 were left; 64 studies were left after eliminating unrelated studies and reviews. We included articles that met the expected requirements: (1) The data was provided as two-by-two tables and (2) full text publications and (3) used at least one accepted reference standard (biochemical method or molecular methods). The exclusion criteria consisted of the following: (1) studies whose samples were less than 10 to avoid selection bias, (2) meta-analyses, meeting summaries, and systematic reviews, and (3) animal research. There were 49 studies that successfully extracted the two-by-two tables.

Quality assessment and data extraction

For each eligible article, two investigators (XJ Cao and YP Li) independently extracted the following information: the first author, year of publication, MPT64 detection method, reference standard used, methodological quality, and data for the two-by two tables. Any disagreements were resolved via discussion with the third investigator (JY Wang).

According to the Quality Assessment of Diagnostic Accuracy Studies tool-2 (QUADAS-2), recommended by the Cochrane Collaboration, two investigators independently reviewed the methodological quality of the eligible articles [13]. Disagreements were resolved by consensus. Revman 5.3 was used to perform the quality assessment.

Statistical analysis

In order to analyze the summary estimation of MPT64, we constructed the MPT64 test to cross-classify the two-by-two tables. True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) were directly extracted from the original research or obtained by calculation. The forest plots were used to evaluate the sensitivity and specificity of each study, with a 95% confidence interval (95% CIs). The summary receiver operating characteristic (SROC) curve was established to summarize the combined distribution of sensitivity and specificity. The area under the SROC curve (AUC) was used to evaluate the accuracy of the overall test. Moreover, the combined SPE and SEN were also used to calculate the negative likelihood ratio (NLR) and positive likelihood ratio (PLR). The calculation method of NLR is false negative rate (1 sensitivity) divided by true negative rate (specificity). When a test finding is negative, the NLR is used to determine the degree of decreasing false-negative risk for the test, and evaluate the commercial kits diagnostic accuracy [14]. The diagnostic odds ratio (DOR) was also used for analysis which was an easily comparable measure to get the tool validity. DOR not only combines the advantages of SPE and SEN, but also has superior accuracy as a single indicator [15]. The Fagan plot was constructed to show the relationship between the pre-probability, likelihood ratio, and post-probability. The Deek’s funnel plot was constructed to visually check any potential publication bias. The Fagan plot was constructed to show the relationship between the former probability, likelihood ratio, and latter probability. Moreover, in order to perform heterogeneity testing, a bivariate boxplot was constructed.

To explore the reasons for the heterogeneity and the accuracy of the detection of the three kits, we conducted a subgroup analysis of the studies in which the detection method was SD Bioline, Capilia TB, or BD MGIT TBcID. First, we divided the research that used the three kits into one subgroup and those that used other detection methods into another subgroup. Then, we divided “the three-kits group” into three groups: SD Bioline, Capilia TB, and BD MGIT TBcID. Furthermore, the bivariate boxplot was also drawn to assess the overall heterogeneity. Publication bias was tested using the funnel plot.

The analyses were performed using the Stata statistical software package, version 12.0 (Stata Corp LP, College Station, U.S.A.), Review Manager 5.3, and Meta-DiSc 1.4.

Results

Inclusion and exclusion criteria and quality assessment

We searched a total of 1240 records identified through the database searches. After removing duplicate records, we obtained 521 records. Then 441 were excluded; these consisted of two meta-analyses or reviews, thirty-five conference summaries, two case reports, two animal-based research, and four hundred irrelevant studies. We screened 80 records. After excluding 27 full-text articles for reasons, we assessed 53 good-quality full-text articles for eligibility. Finally, data was extracted from 46 articles analysis. The flow diagram is shown in Fig. 1. The characteristics of the studies included in the articles are shown in Table 1. The quality assessment of the included studies is shown in Fig. 2.

Fig. 1
figure 1

Flow diagram of study identification and inclusion

Full size image
Table 1 Characteristics of the studies included in the articles
Full size table
Fig. 2
figure 2

Quality assessment of the included studies. a. Overall quality assessment of the included studies, b. Quality assessment of the individual studies

Full size image

Overall accuracy of MPT64

To explore the diagnostic accuracy of MPT64 for MTB, we adopted a random-effects model. MPT64 showed good diagnostic performance for MTB. However, there was obvious heterogeneity among the 46 studies. The SEN and SPE and associated 95% CIs were 0.92 (0.91–0.93) and 0.95 (0.94–0.95), respectively (Fig. 3). The NLR and PLR were 0.04 (95% CI 0.03–0.07) and 25.32 (95% CI 12.38–51.78), respectively (Fig. 4). DOR was 639.60 (243.04–1683.18) (Fig. 5). The AUC was 0.99 (Fig. 5), indicating that the diagnostic accuracy of the MPT64 test was very high. The result of overall accuracy of MPT64 was shown in Table 2.

Fig. 3
figure 3

Forest plots of sensitivity and specificity. a. sensitivity, b. specificity

Full size image
Fig. 4
figure 4

Forest plots of positive LR and negative LR. a. positive LR, b. negative LR

Full size image
Fig. 5
figure 5

Overall diagnostic efficacy of MPT64 assays for Mycobacterium tuberculosis. a. diagnostic OR for the diagnosis of Mycobacterium tuberculosis infection, b. SROC curve

Full size image
Table 2 Overall Accuracy of MPT64
Full size table

According to the Fagan plot (Fig. 6), the pre-test probability was 50% and the post-test probability was 99%. The post-test probability significantly improved.

Fig. 6
figure 6

Fagan plot of disease probabilities based on Bayes’ theorem

Full size image

Subgroup analysis of the three commercial kits

The results of the subgroup analyses of the three kits are shown in Table 3, Fig. 7 and Fig. 8. SD Bioline had high pooled specificity and sensitivity for MPT64 detection. There was no significant change in SEN and SPE, indicating that the accuracy of the diagnosis did not depend on the kit.

Table 3 Subgroup analyses for three commercial kits
Full size table
Fig. 7
figure 7

The results of subgroup analysis between “three commercial kits group” and other detection methods. a. the result of “three commercial kits group”, b. the result of other detection methods group

Full size image
Fig. 8
figure 8

The results of subgroup analysis for the three commercial kits. a. the result of BD MGIT TBcID kit, b. the result of Capilia TB kit, c. the result of SD Bioline kit

Full size image

Heterogeneity and publication Bias

As shown by the results of subgroup analyses, the heterogeneity of “the three-kits group” was high. However, when we reviewed the full text and eliminated the research of Kumar et al. and Gomathi et al., the heterogeneity was significantly reduced (less than 50%). According to the bivariate boxplot (Fig. 9b), there were seven sets of data outside the circle, which also showed that there was significant heterogeneity in the overall research.

Fig. 9
figure 9

Publication bias for MPT64 detection for MTB. a. Deeks’ funnel plot asymmetry test to assess the publication bias for MPT64 detection for MTB; b. Bivariate boxplot

Full size image

As shown in Fig. 9a, publication bias existed, with a p value of 0.012.

Discussion

TB is a serious infectious disease and every year, millions of people worldwide contract MTB. Moreover, a large number of people die from TB [1]. Thus, there is an urgent and essential need to develop real-time, portable, and sensitive techniques to detect MTB and its drug-resistant mutations. This study evaluated the accuracy of the diagnosis of MTB by using various MPT64-detecting methods.

Although Yin et al [12] conducted similar research in 2013, new articles have been published since then. Therefore, we have updated their research. Our study analyzed more articles than theirs, which included only 28 articles. Therefore, for now, our research is more comprehensive. Moreover, we added a Fagan plot, which verified the clinical application value of MPT64. After using the MPT64 test, the post-test probability significantly improved. Moreover, when analyzing the heterogeneity, we came to the opposite conclusion as Yin et al. Their research showed that except for the comprehensive sensitivity of the MGIT TBc ID test and the pooled specificity of the SD Bioline Ag MPT64 rapid determination, all statistical indicators had considerable heterogeneity. However, our research found that after excluding the two articles that had problems in sample handling, there was no significant heterogeneity (I2 < 50%) between the three commercial kits.

The overall result showed that MPT64 had a good test performance. In the subgroup analyses, we eliminated two articles because one article mixed weak positives with positives and the samples of another article were partially contaminated. Finally, the results of the subgroup analyses showed that the diagnostic accuracy of MPT64 did not depend on the kit. In addition, there was no obvious heterogeneity between the three commercial kits. Therefore, when resources are insufficient, cheaper kits can be used.

In our study, we only analyzed the impact of the kit on the diagnostic accuracy and did not analyze whether other factors, such as sample type, affect it. In addition, the diagnostic efficacy of MPT64 for different types of tuberculosis is worth investigating. The diagnosis of MPT64 in different populations remains to be studied. For instance, Jorstad et al [21] analyzed the influence of age on diagnostic accuracy and found that the sensitivity of the MPT64 test was significantly higher in children than in adults. Due to insufficient extracted data, we were unable to analyze and verify this.

Conclusion

In conclusion, the MPT64 test shows a good diagnostic performance for MTB; it has high sensitivity and specificity as well as clinical application value. Moreover, the three commercial kits, SD Bioline, Capilia TB, and BD MGIT TBcID, are not heterogeneous. Therefore, when resources are insufficient, cheaper kits can be used.