Novel endoscopic techniques for the diagnosis of gastric Helicobacter pylori infection: a systematic review and network meta-analysis

Objective This study aimed to conduct a network meta-analysis to compare the diagnostic efficacy of diverse novel endoscopic techniques for detecting gastric Helicobacter pylori infection. Methods From inception to August 2023, literature was systematically searched across Pubmed, Embase, and Web of Science databases. Cochrane’s risk of bias tool assessed the methodological quality of the included studies. Data analysis was conducted using the R software, employing a ranking chart to determine the most effective diagnostic method comprehensively. Convergence analysis was performed to assess the stability of the results. Results The study encompassed 36 articles comprising 54 observational studies, investigating 14 novel endoscopic techniques and involving 7,230 patients diagnosed with gastric H. pylori infection. Compared with the gold standard, the comprehensive network meta-analysis revealed the superior diagnostic performance of two new endoscopic techniques, Magnifying blue laser imaging endoscopy (M-BLI) and high-definition magnifying endoscopy with i-scan (M-I-SCAN). Specifically, M-BLI demonstrated the highest ranking in both sensitivity (SE) and positive predictive value (PPV), ranking second in negative predictive value (NPV) and fourth in specificity (SP). M-I-SCAN secured the top position in NPV, third in SE and SP, and fifth in PPV. Conclusion After thoroughly analyzing the ranking chart, we conclude that M-BLI and M-I-SCAN stand out as the most suitable new endoscopic techniques for diagnosing gastric H. pylori infection. Systematic review registration https://inplasy.com/inplasy-2023-11-0051/, identifier INPLASY2023110051.


Introduction
Helicobacter pylori is a Gram-negative bacterium infecting the epithelial layer of the human stomach, capable of colonizing and persisting in a unique biological niche within the gastric lumen (Correa, 1988;Dunn et al., 1997).In 1994, the International Agency for Research on Cancer (IARC) classified H. pylori as a Group I carcinogen (IARC, 1994).It is linked to chronic gastritis, gastric ulcers, duodenal ulcers, gastric adenocarcinoma, and gastric mucosaassociated lymphoid tissue lymphoma (Suerbaum and Michetti, 2002;Graham, 2015).Over half of the world's population is infected, with prevalence reaching 25 to 50% in developed countries and 70 to 90% in developing nations (Xia and Talley, 1997;Hooi et al., 2017).Postinfection, it sequentially leads to chronic atrophic gastritis, intestinal metaplasia, dysplasia, and gastric cancer (Correa, 1988).Timely diagnosis holds immense significance for H. pylori eradication, preventing diseases such as gastric cancer (Choi et al., 2018).
The diagnosis of gastric H. pylori infection traditionally involves invasive techniques like histological examination, H. pylori culture, and polymerase chain reaction, as well as non-invasive methods such as serological detection, urea breath test (UBT), and fecal antigen detection (Vaira et al., 2002;Wang et al., 2015;Choi et al., 2018).However, the accuracy of invasive detection is affected by factors like biopsy location, size, number, staining methods, and antibiotic use.On the other hand, non-invasive techniques can be influenced by antibiotics, bismuth, and test reaction time (Logan and Walker, 2001).Are there more accurate and intuitive diagnostic options for gastric H. pylori infection?Over the past decade, plain white light imaging endoscopy (WLE) has been utilized as a diagnostic tool for the invasive detection of gastric H. pylori infection.While WLE cannot replace UBT as the diagnostic foundation, it can determine the presence or absence of H. pylori infection during primary disease examination.WLE offers advantages in intuitiveness, immediacy, strong operability, and the potential to avoid biopsy.It guides follow-up examination and treatment, presenting a novel approach to H. pylori diagnosis (Glover et al., 2020b).In this context, endoscopic invasive methods for diagnosing gastric H. pylori infection have emerged as a superior screening tool and research focus.Recent advancements in endoscopic technology have introduced new types of endoscopy, surpassing traditional WLE.These include Magnifying Endoscopy (ME), Narrow Band Imaging Endoscopy (NBI), Linked Color Imaging Endoscopy (LCI), Confocal Laser Endomicroscopy (CLE), Near-Infrared Raman Spectroscopy Endoscopy (NIR), Artificial Intelligence-based Computer-Aided Diagnosis (AI-CAD), and Convolutional Neural Network (AI-CNN; Glover et al., 2020b).Compared to traditional endoscopy, the various images produced by these new endoscopic methods enable better observation of microscopic structures, such as gastric pit patterns, microvessels, cell morphology, and even microorganisms.
Additionally, AI combined with endoscopic images can be trained to determine the presence or absence of infection.The enlargement of gastric pits, disappearance of collecting veins, and the vanishing capillary network increasingly indicate specific endoscopic characteristics of gastric H. pylori infection.This facilitates rapid and minimally invasive endoscopic diagnosis, bringing it closer to pathological diagnosis (Ji and Li, 2014).
A prospective study conducted by Gonen et al. in Turkey, involving 129 patients, affirmed the superiority of high-resolution Magnifying Endoscopy (ME) over White Light Endoscopy (WLE) in diagnosing gastric gastritis associated with H. pylori infection (Gonen et al., 2009).In another prospective study by Ozgur et al. (2013), it was demonstrated that mucosal changes in patients with gastric H. pylori infection were more readily identified using narrow-band imaging (NBI) compared to WLE, with NBI exhibiting a high sensitivity of 92.86% (Özgür et al., 2015).Yagi et al. (2014) compared the diagnostic efficacy of WLE and Magnifying NBI (M-NBI) in patients with post-endoscopic resection.The interobserver agreement for conventional endoscopy was moderate (0.56), with sensitivity and specificity at 79 and 52%, respectively.In contrast, M-NBI demonstrated substantial interobserver agreement (0.77), with sensitivity and specificity reaching 91 and 83% (Yagi et al., 2014).Qi et al. compared ME and M-I-Scan's diagnostic performance and image quality for gastric H. pylori infection.M-I-Scan exhibited high sensitivity and specificity, surpassing ME specificity significantly (sensitivity: 95.45% vs. 95.45%,specificity: 93.55% vs. 80.65%;Qi et al., 2013).In 2017, Shichijo et al. developed an Artificial Intelligence-based Convolutional Neural Network (AI-CNN) capable of diagnosing gastric H. pylori infection through endoscopic images.After learning from 32,208 images across 1750 patients, AI-CNN demonstrated higher accuracy, specificity, and sensitivity than 23 endoscopists.Additionally, the time required for AI-CNN to generate diagnoses was considerably faster than that of the endoscopists (194 s vs. 230 min; Shichijo et al., 2017).Simultaneously, Itoh et al. demonstrated in their study that their AI-CNN deep learning algorithm, trained on 149 endoscopic images under WLE of patients with H. pylori status, achieved diagnostic sensitivity and specificity of 86.7% each when tested on 30 new endoscopic photos (Itoh et al., 2018).In 2023, Zhang et al. published research unveiling AI-WLE, developed using 47,239 images from 1826 patients, which exhibited an accuracy of 91.1% [95% confidence interval (CI): 85.7-94.6].This accuracy was significantly higher than endoscopists (15.5% [95% CI: 9.7-21.3%]).Furthermore, its high sensitivity (0.9290) and specificity (0.8930) were confirmed (Zhang et al., 2023).
In detecting gastric H. pylori infection, different new endoscopes exhibit varying characteristics in terms of sensitivity, specificity, and diagnostic efficiency.Existing systematic reviews or meta-analyses have primarily focused on comparing non-magnifying endoscopy or artificial intelligence for diagnosing human gastric H. pylori images, with a notable absence of comparisons among different new endoscopic techniques.Consequently, evidence-based recommendations regarding the most suitable diagnostic method for gastric H. pylori infection still need to be made (Qi et al., 2016;Bang et al., 2020;Glover et al., 2020a,b).Hence, it is crucial to identify an appropriate technique for diagnosing gastric H. pylori infection among the array of new endoscopic options, particularly when clinicians select different endoscopes for patient diagnosis in clinical practice.
Network meta-analysis, a contemporary evidence-based technique utilizing direct or indirect comparisons, is employed to assess the effects of multiple interventions on disease and estimate the hierarchical order of each intervention (Rouse et al., 2017).In this study, we aggregated existing evidence.We conducted a network meta-analysis to compare novel endoscopic techniques (BLI, LCI, CLE, NBI, ME, AI-CNN, etc.) to evaluate and contrast their diagnostic performance in gastric H. pylori infection patients.This approach aims to furnish patients and clinicians with disease-specific, evidence-based data, facilitating the selection of suitable diagnostic methods for screening and diagnosis.
1.The experimental group employed a novel endoscopic technique as a diagnostic measure for gastric H. pylori infection; 2. The gold standard for diagnosis included the Rapid Urease and Breath Test; 3. Diagnostic techniques comprised novel endoscopic methods and up to two diagnostic approaches; 4. The reported outcome indicators encompassed true positive (TP), true negative (TN), false positive (FP), false negative (FN), sensitivity (Se), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV).When TP, TN, FP, FN, NPV, or PPV were not reported, calculations were derived from known variables such as Se and Sp; 5.The study design adhered to a prospective or retrospective approach.

Exclusion criteria
1. Absence of well-defined inclusion and exclusion criteria; 2. Non-clinical investigations; 3. Excluded document types: guidelines, systematic reviews, meta-analyses, narrative reviews, letters, editorials, research protocols, case reports, short newsletters, etc.; 4. Incomplete research data, duplicated publications, etc.Studies meeting any of the exclusion criteria were excluded from the analysis.

Study selection
The literature was managed using EndNote X9.1 for screening and exclusion.Initially, the two researchers checked for duplications in literature titles, review papers, conference papers, protocols, and short communications.Subsequently, both researchers reviewed the literature's abstracts to determine inclusion and exclusion criteria.The two researchers then comprehensively reviewed the remaining literature to finalize the inclusion scope.The researchers independently screened the literature throughout this process, and the results were compared.In discrepancies, a discussion ensued, and resolution was achieved with the involvement of a third researcher.

Data extraction
Data for inclusion in the study were recorded using a standardized, preselected nine-item data extraction form, categorized under the following headings: 1. Author, 2. Country, 3. Year of publication, 4. Mean age, 5. Total number of individuals and the distribution by sex, 6. Diagnostic methods, 7. Gold standard, 8. Sensitivity, 9. Specificity.

Literature quality evaluation
Two investigators independently conducted a quality assessment using the Quality Assessment of Diagnostic Accuracy Studies Tool (QUADAS-2), and the assessment results were cross-checked (Yang et al., 2021).Any disparities were deliberated upon and resolved by a third investigator.The evaluation scale encompassed the assessment of risk of bias and clinical applicability.The risk of bias evaluation included four sections: case selection, trial under consideration, gold standard, cash flow, and progress.All components underwent assessment for risk of bias, and the initial three components underwent assessment for clinical applicability.The risk of bias was categorized as "low, " "high, " or "uncertain."

Data analysis
We conducted network meta-analysis aggregation and analysis employing Markov chain Monte Carlo simulation chains within a Bayesian framework, utilizing R software version 4.3.1,following the guidelines outlined in the PRISMA network meta-analysis manual (Moher et al., 2015).The resulting network diagram, generated by the R software, illustrates and describes various novel endoscopes.Each node on the network diagram signifies a distinct novel endoscopic technique, while the connecting lines represent direct head-to-head comparisons with the gold standard.The size of each node and the width of the connecting line are proportional to the number of studies conducted (Chaimani et al., 2013).

Quality assessment of included studies
We utilized R software (version 4.3.1) to conduct a Bayesian network meta-analysis involving 36 articles, encompassing 54 observational studies.The quality, risk of bias, and applicability of these 36 articles were assessed using QUADAS-2.Overall, the articles demonstrated satisfactory quality, with 25 rated high quality and 11 as medium quality.Regarding personnel selection, 13 out of 36 articles had an unclear risk of bias, mandating informed consent from patients or their relatives before testing with new endoscopic techniques.Ten articles exhibited an unclear risk of bias in index detection, while 12 had a dark bias in reference standard assessment.The risk of bias in follow-up time was uncertain for 10 articles.Applicability considerations revealed no increased risk of bias in patient selection, reference standards, and index testing (refer to Figure 2 for details).

Network meta-analysis
The full Network meta-analysis figure will be shown in Figures 3A,  4A, 5A, 6A.

Sensitivity
In the results of the network meta-analysis, when compared to the gold standard detection, AI-BLI [MD = 0.966, 95%CI: (0.706, 1.  3. Convergence analysis confirmed the stability of the results, as depicted in Figure 3B.The bar chart illustrates the top five sensitivities in descending order: M-BLI (0.282), AI-BLI (0.237), M-I-SCAN (0.206), OE-ME (0.132), and M-LCI (0.049; Figure 3C).Table 3 presents a comparison between these two distinct detection measures.

Specificity
The network meta-analysis results indicated differences in specificity compared to the gold standard for various endoscopic techniques: AI-BLI [MD = 0.867, 95%CI: (0.697, 1.   4).Convergence analysis demonstrated the stability of the results, as illustrated in Figure 4B.The ranked bar chart revealed the top five specificities in descending order: TXI-IEE (0.275), BLI (0.236), M-I-SCAN (0.178), M-BLI (0.140), and CLE (0.105; Figure 4C).Table 4 provides a comparison between these two distinct measures of detection.

Positive predictive value
Network meta-analysis results revealed differences from the gold standard in terms of positive predictive value for various endoscopic techniques: AI-BLI [MD = 0.879, 95%CI: (0.536, 1.  5).Convergence analysis demonstrated the stability of the results, as illustrated in Figure 5B.The ranked histogram revealed the top five positive predictive values in descending order: M-BLI (0.232), TXI-IEE(0.206),AI-BLI(0.144),BLI(0.122), and M-I-SCAN(0.099; Figure 5C).Table 5 provides a comparison between these two distinct measures of detection.

Negative predictive value
Network Meta-analysis results demonstrated differences in negative predictive value compared to the gold standard for various   6 presents a comparison between these two distinct measures of detection.

Regression analysis
To examine the effect of age as well as the classification of the gold standard on the results, we performed a meta-regression analysis using StataMP 18.

Regression analysis of age
The results of regression analysis showed that the mean age of the study population was not a statistically significant moderator of SE (p > 0.3370), PPV (p > 0.1370), and NPV (p > 0.8860; Table 7).In addition, there is no sufficient reason to deny that the mean age of the population is not a moderator of SP (p > 0.0030).Table 7 provides details on the age regression analysis of the results.

Regression analysis of gold standard classification
Regression analysis showed that the classification of the gold standard was not a statistically significant moderator of SE (p > 0.4280), PPV (p > 0.4280) and NPV (p > 0.0790; Table 8).In addition, there was still no sufficient reason to reject that the classification of the gold standard was not a moderator of SP (p > 0.0330).Digestive endoscopy is a relatively invasive examination, which is the basis of all invasive examination methods of the upper digestive tract.Compared with non-invasive examination, digestive endoscopy may cause certain throat discomfort, nausea or transient digestive discomfort (Liang et al., 2022), but it can more accurately determine the scope of Hp infection and the degree of damage to the gastric mucosa, which is an advantage that other methods do not have.
This study aimed to assess the diagnostic efficacy of various novel endoscopic techniques in screening for gastric H. pylori infection.It encompassed 36 articles with 54 studies, incorporating 14 distinct endoscopic techniques and gold-standard detection methods.The quantitative analysis included a substantial sample size of 7,230 patients.Our findings indicate that M-BLI, M-I-SCAN, AI-BLI, and TXI-IEE exhibit higher diagnostic efficacy than the gold standard.Our study represents the first comprehensive network meta-analysis of diagnostic tests for these novel endoscopic techniques in diagnosing gastric H. pylori infection.
The network meta-analysis results showed that in the ranking chart of new endoscopic techniques, M-BLI ranked first in sensitivity and positive predictive value, second in negative predictive value, and fourth in specificity.
M-BLI is an innovative image enhancement technique integrating ME and BLI.ME, a magnifying endoscopy, enhances resolution by incorporating a zoom lens to aid endoscopists in better observing gastric mucosa details, including pits, collecting venules, and capillary shapes (Bessède et al., 2017).Observing gastric mucosa with H. pylori infection often reveals enlarged pits, irregular or vanished capillary networks, and irregular or absent collecting veins.Conversely, a honeycomblike capillary network, gastric body RAC, and regular round pits frequently indicate the absence of H. pylori infection in the gastric mucosa (Qi et al., 2016).BLI (Blue Laser Imaging Endoscopy) is an advanced contrast imaging technology The large curvature of the mid-upper body of the stomach was meticulously assessed using M-BLI or M-NBI.Small round pits with a regular honeycomb subepithelial capillary network (SECN) regularly scattered in the collecting venules were considered negative for H. pylori infection.Enlarged or extended pits, unclear SECN, or dense, OK, irregular blood vessels indicated H. pylori positivity.The sensitivity, specificity, PPV, and NPV of BLI were 0.98, 0.92, 0.93, and 0.98, respectively, compared with 0.97, 0.81, 0.87, and 0.95 in the NBI group.No significant differences were found between M-BLI and M-NBI groups (all p > 0.2; Tahara et al., 2017).However, with the   inclusion of more recent literature in the mesh meta-analysis, our study shows that M-BLI significantly outperforms M-NBI in SE, SP, PPV, and NPV.The reason may be that M-NBI is not enough to reveal the changes in hemoglobin absorption characteristics, and the contrast and resolution of some diseases with fine structures or similar colors are limited.However, M-BLI is not limited by spectrum compared with M-NBI, and can use the biofilm interference principle to provide a wider range of biomolecular level interaction information.It has higher contrast and resolution to provide sharper images in some cases, and has lower operational dependence.This study demonstrates the high diagnostic accuracy and utility of M-BLI in diagnosing gastric H. pylori infection.Therefore, considering the sensitivity and positive predictive value, M-BLI exhibits superior diagnostic performance and can be recommended as a promising detection tool for gastric H. pylori infection.
The ranking revealed that M-I-SCAN excelled in negative predictive value, ranking first; its sensitivity and specificity were third, and positive predictive value fifth.I-SCAN developed by Pentax Company in Japan, is a computer virtual staining imaging technology with three critical functions for real-time image enhancement: surface enhancement (SE), contrast enhancement (CE), and hue enhancement (TE).The first two enhance lesion identification without significantly altering the color hue and image brightness, often used in tandem.Hue enhancement makes color, hue, and structural changes more apparent after lesion identification.TE includes modes like g for the stomach, c for the intestine, e for the esophagus, b for Barret's esophagus, p for the mucosa, and v for the small blood vessels.Besides microvascular morphology and fine structure observation, I-Scan demonstrates multi-channel and multicolor contrast capabilities, offering unique advantages for determining lesion edges and classifying glandular tube openings (Glover et al., 2020b;Tosun et al., 2022).Sharm et al. conducted a study with 146 patients.WLE's sensitivity, specificity, positive predictive value, negative predictive value, and accuracy in diagnosing H. pylori infection were 59, 100, 100, 69, and 78%, respectively.I-Scan endoscopy exhibited 100, 95, 96, 100, and 97% in the same metrics.I-Scan was superior in observing the fine structure of gastric mucosa, but additional studies are required to understand the H. pylori infection pattern (Sharma et al., 2017).A magnifying endoscopic ME and an i-scan have been developed, providing more explicit images of mucosal and vascular patterns.Qi et al. utilized M-I-Scan and ME to observe H. pylori infection in the gastric mucosa of 84 patients.The accuracy of M-I-Scan in diagnosing H. pylori infection (94.0% vs. 84.5%,p < 0.05, p = 0.046) and specificity (93.5% vs. 80.6%, p = 0.032) were higher than ME (Qi et al., 2013).Therefore, combining ME with I-SCAN testing can uphold a robust negative predictive value for diagnosing H. pylori infection and potentially reduce medical costs.
The ranking chart indicates that TXI-IEE secured the top position in specificity, claimed the second spot in positive predictive value, ranked ninth in sensitivity, and held the tenth position in negative predictive value.TXI-IEE, developed by Olympus Medical Systems (Tokyo, Japan) in 2020, is an    The study also examined the association of endoscopic features with three categories of gastric H. pylori infection status (currently infected, previously infected, and noninfected).Results indicated that TXI-IEE exhibited significantly higher diagnostic accuracy for active gastritis than WLI (85.3% vs. 78.7%;p = 0.034).Odds ratios (ORs) for all endoscopy-specific features related to gastric H. pylori infection status were higher in the TXI-IEE group than in the WLI group.Notably, diffuse redness was the sole observation for current infection (OR, 22.0 and 56.1, respectively).Geographic redness was considered indicative of previous infection (OR 6.3 and 11.0, respectively), while regular alignment of collecting venules (RAC) was associated with an uninfected status (OR 25.2 and 42.3, respectively).All specific endoscopic features linked to gastric H. pylori infection status demonstrated higher ORs in the TXI-IEE group than in the WLI group.TXI-IEE enhanced the visibility of diffuse redness, geographic redness, and RAC by creating more excellent contrast (Kitagawa et al., 2023).Therefore, TXI-IEE, with its highest specificity, can potentially reduce the need for unnecessary gastric biopsy.However, it has limitations due to its low sensitivity and negative predictive value.
The ranking chart indicates that AI-BLI ranks second in sensitivity, third in positive and negative predictive values, and sixth in specificity.Artificial intelligence (AI) is the fastestgrowing field in endoscopic research, which is increasingly applied in clinical practice, particularly for image recognition and classification (Cho and Bang, 2020).In contrast to optical endoscopy, AI-assisted endoscopy exhibits operatorindependent characteristics, ensuring a completely objective diagnostic process.In clinical practice, AI-assisted endoscopy proves valuable for offering second opinions and reducing operator dependence in diagnostic endoscopy (Hoogenboom et al., 2020).Wu et al. demonstrated that AI, coupled with BLI, enhances innovation by identifying the typical structure of the digestive tract.This capability alerts endoscopists to missed sites, significantly reducing the blind spot rate in digestive endoscopy (Wu et al., 2019).Nakashima H et al. developed an artificial intelligence system to predict gastric H. pylori infection status using blue laser imaging (BLI)-bright and linked color imaging (LCI) endoscopic images.Two hundred twenty-two   patients underwent WL, BLI-bright, and LCI to capture three still images of the gastric lesser curvature.Among them, 162 patients constituted the training set, while the remaining 60 patients served as the test set for verification.Results revealed that the area under the curve (AUC) of the receiver operating characteristic analysis for WLI was 0.66.The AUC of BLI-bright and LCI were 0.96 and 0.95, respectively.The AUC of the BLI-bright and LCI groups significantly exceeded that of the WLI group (p < 0.01; Nakashima et al., 2018).A systematic review and network meta-analysis conducted by Bang CS et al. further demonstrated the clinical utility of AI algorithms as an additional tool for predicting gastric H. pylori infection during endoscopic surgery (Bang et al., 2020).The diagnostic indexes of AI-BLI are promising for use as a detection tool.Still, its diagnosis is susceptible to the endoscopic images included in the study, introducing selection bias and, therefore, certain limitations.

Advantages and limitations
Firstly, our study encompassed 36 articles, comprising 54 observational studies, exploring 14 new endoscopic techniques, and involving 7,230 patients undergoing these novel techniques for diagnosing gastric H. pylori infection.The study stands out for its extensive literature coverage, substantial sample size, minimal heterogeneity in results, and rigorous methodology.Secondly, our research and its foundational studies confront certain limitations.Some newly developed endoscopic diagnostic techniques received limited coverage in the literature.The endoscopic operator's experience might influence the efficacy of specific diagnostic methods.The inclusion of patients may have been impacted by factors such as age or medications taken, contributing to diagnostic variations among different gold standards.Readers should exercise caution in interpreting our study's results.For instance, only one report exists on M-BLI, M-I-SCAN, TXI-IEE, and AI-BLI for diagnosing gastric H. pylori infection, indicating a need for further expansion and exploration.

Conclusion
In this study, we comprehensively compared 14 novel endoscopic techniques with the gold standard for diagnosing gastric H. pylori infection, utilizing Bayesian network metaanalysis.The findings indicate that M-BLI and M-I-SCAN exhibit robust diagnostic performance, emerging as particularly suitable endoscopic techniques for diagnosing gastric H. pylori.Despite some limitations, TXI-IEE and AI-BLI serve as valuable tools for early detection and diagnosis of gastric H. pylori infection, holding clinical significance in minimizing unnecessary biopsies and optimizing medical resource utilization.Nevertheless, this conclusion warrants validation through additional literature, and future research demands more meticulously designed, large-scale, and multicenter studies to further elucidate the application value of various new endoscopic techniques in diagnosing patients with gastric H. pylori infection.

FIGURE 1
FIGURE 1Flow diagram of literature selection.
FIGURE 4 (A) Network meta-analysis figure for Specificity.(B) Convergence analysis for Specificity.(C) Ranking chart for Specificity.
FIGURE 5 (A) Network meta-analysis figure for PPV.(B) Convergence analysis for PPV.(C) Ranking chart for PPV.
FIGURE 6 (A) Network meta-analysis figure for NPV.(B) Convergence analysis for NPV.(C) Ranking chart for NPV.

TABLE 1
Search strategy on PubMed.

TABLE 2
Characteristics of the studies included in the meta-analysis.

TABLE 3
League table on sensitivity.

TABLE 4
League table on specificity.

TABLE 5
League table on PPV.