Evaluation of GeneXpert vanA/vanB in the early diagnosis of vancomycin-resistant enterococci infection

Purpose Vancomycin-resistant enterococci infection is a worrying worldwide clinical problem. To evaluate the accuracy of GeneXpert vanA/vanB in the diagnosis of VRE, we conducted a systematic review in the study. Methods Experimental data were extracted from publications until May 03 2021 related to the diagnostic accuracy of GeneXpert vanA/vanB for VRE in PubMed, Embase, Web of Science and the Cochrane Library. The accuracy of GeneXpert vanA/vanB for VRE was evaluated using summary receiver to operate characteristic curve, pooled sensitivity, pooled specificity, positive likelihood ratio, negative likelihood ratio, and diagnostic odds ratio. Results 8 publications were divided into 3 groups according to two golden standard references, vanA and vanB group, vanA group, vanB group, including 6 researches, 5 researches and 5 researches, respectively. The pooled sensitivity and specificity of group vanA and vanB were 0.96 (95% CI, 0.93–0.98) and 0.90 (95% CI, 0.88–0.91) respectively. The DOR was 440.77 (95% CI, 37.92–5123.55). The pooled sensitivity and specificity of group vanA were 0.86 (95% CI, 0.81–0.90) and 0.99 (95% CI, 0.99–0.99) respectively, and those of group vanB were 0.85 (95% CI, 0.63–0.97) and 0.82 (95% CI, 0.80–0.83) respectively. Conclusion GeneXpert vanA/vanB can diagnose VRE with high-accuracy and shows greater accuracy in diagnosing vanA.


Introduction
Since 1988, vancomycin-resistant enterococci (VRE) have been found in patients with critical diseases due to extensive use of antibiotics, prolonged hospital stays and intensive care unit (ICU) admission [1]. They became a type of antimicrobial resistance (AMR) bacteria that most commonly spread in medical institutions, especially in Europe [2], with an incidence of 2-34.9% [3]. At present, VRE is prevalent globally, and its prevalence has increased significantly, which is a worrisome clinical problem worldwide [4].
VRE testing is currently performed mainly by traditional culture and Polymerase Chain Reaction (PCR) detection of the resistance genes vanA and vanB [5,6]. Although culture is the confirmed reference method [7,8], it takes a long time, requires complex extraction and detection steps and has a high economic impact during a VRE outbreak [9]. The U.S. Food and Drug Administration (FDA) approved a rapid molecular assay, the GeneXpert vanA/vanB [8,10], which is a unique and completely automated process that includes deoxyribonucleic acid (DNA) extraction, amplification and detection using real-time PCR. Furthermore, results are usually available in less than one hour [4,5].
It is indicated that GeneXpert vanA/vanB testing is sensitive as well as cost-effective [5,11]. In addition, there are some researches supporting that some indetermination results exist in that of GeneXpert vanA/vanB detecting van B [10,12]. There are few systematic-analyses on the diagnostic accuracy of GeneXpert vanA/vanB for VRE in evidence-based medicine. Therefore, to appraise the accuracy of GeneXpert vanA/vanB in the diagnosis of VRE and distinguish the differences between GeneXpert vanA/vanB detecting vanA and vanB, we conducted data integration and analysis.

Search strategy
A systematic literature search was carried out for publications until May 03, 2021, related to the diagnostic accuracy of GeneXpert vanA/vanB for VRE. Four databases were involved: PubMed, Embase, Web of Science and the Cochrane Library. According to PCIO criteria, the search stratagem utilized was as follows: (((Enterococcus) AND (Vancomycin Resistance)) OR (Vancomycin-Resistant Enterococci)) AND (GeneXpert vanA/vanB). Possible matches were also retrieved from the related references and the language was restricted to English.

Study selection
Inclusion criteria: (i) Each included study used GeneXpert VanA/VanB for detection of VRE. Clinical specimens were identified as VRE or standard strains by reference methods, which were regarded as the gold standards; (ii) Human samples were detected and analyzed; (iii) A 2 × 2 table was constructed with sufficient data to estimate sensitivity, specificity, and the likelihood ratio.
Exclusion criteria: (i) Samples from animals or other species; (ii) Reference standards cannot be found; (iii) Incomplete raw data: when the raw data were unable to construct the 2 × 2 tables, or when raw data were unable to obtained from the authors; (iv) Duplicate publications; (v) Reviews, conference abstracts, case reports and studies that data extraction was impossible to perform.
Two independent reviewers assessed the studies according to the defined criteria above. If the results were found to be inconsistent, the third investigator was consulted and concluded the same.

Data extraction
An Excel spreadsheet was created to collect data, which was extracted by two investigators who scanned the included literature independently. Any disagreements were reconciled by a third team member. The following variables comprise the first author's name, the publication year, the area where the research was implemented, type of study, clinical features and settings, the specimen type, reference standard test, and false and true positives and negatives (TP, TN, FP, FN). When we discuss vanA and vanB simultaneously, named vanA and vanB group, Mycobacterial culture was defined as the gold standard. When we discuss vanA or vanB separately, named vanA group and vanB group, the golden standard was defined as mycobacterial culture and PCR.
In the studied texts, multiple groups and different backgrounds were considered discrete units of analysis comprising a single study.

Quality assessment
The quality of the publications were assessed using Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) [12]. There are four key domains that compose the tool, patient selection, the index test, reference standard and flow and timing, that evaluates bias and utility of the reviewed studies. Values of high, unclear, or low risk were assigned to grade each group of data conducted by different researchers independently figuring out the questions of the four domains. When a divergence appeared, a third investigator was invited to make the final decision.

Statistical analysis
(1) Statistical testing. The pooled sensitivity, specificity, positive likelihood ratios (PLR), negative likelihood ratios (NLR), diagnostic odds ratio (DOR) and 95% confidence intervals (95% CI) were analyzed based on the data provided in the article and evaluated by forest plots, adopting a random-effects model. A value of 0.5 was added to studies with zero values to correct for continuity. A Fagan's nomogram was facilitated to estimate the clinical application of GeneXpert vanA/ vanB for the clinical diagnosing of VRE [13] by calculating the pre-test and post-test probabilities.
(2) Analysis of heterogeneity. In diagnostic experiments, the threshold effect or nonthreshold effect might be the primary cause of heterogeneity [14]. We gave priority to ensure whether the threshold effect exists by plotting summary receiver operator characteristic (SROC) curve and further calculating the Spearman correlation coefficient (R). An SROC space shows a typical "shoulder arm" pattern, suggesting the presence of a threshold effect. An R � 0.6 revealed a threshold effect, which manifests a rapid increase of the logit of sensitivity with the logit of 1-specificity adding [15].
Several reasons other than threshold have contributed to the appearance of correlation between sensitivity and specificity [16]. Cochran's Q test and the inconsistence index (I2) were facilitated to evaluate heterogeneity. When I2<50%, evidence shows no significant heterogeneity, use fixed-effects model. On the contrary, the random-effects model is adopted [17]. We performed meta regression and the sensitivity analysis to investigate potential sources of heterogeneity. AUC (the area under the SROC curve) takes values between 0 and 1, presenting an overall summary performance of studies [18]. To analyze publication bias, Deeks' funnel plot was applied; P > 0.05 showed that this meta-analysis has no publication bias [19].
(3) Tools. Meta-DiSc 1.4 was employed to analyze all data and STATA 12.0 was employed to draw Fagan's nomogram, bivariate box plot, and evaluating publication bias. Review Manager (RevMan) 5.3 software was applied to conduct the quality assessment.

Description of meta-analyzed publications
Of the 8 articles, the publication years range from 2010 to 2019. Two were from the U.S. Three studies were retrospective while the remaining were prospective. The sample size was comprised 3064 subjects in total, 1563 subjects of which were categorized as vanA and vanB group and 2362 subjects were categorized as vanA group, vanB group. Sample types included rectal swabs, blood cultures, perianal swabs, and stool. Except for the articles which did not refer to the patients, three studies introduced patients from ICUs, one study's patients suffered from renal dialysis and another's patients were from hematology or gastroenterology departments. All bacteria were diagnosed as VRE.
Study characteristics in Table 1 show individual studies and their characteristics respectively.

Heterogeneity and publication bias
No "shoulder arm" SROC curve was observed (Fig 1), and the Spearman correlation coefficient (R) was -0.943. In conclusion, there was no evidence of threshold effect. A forest map of DOR (Fig 2A) revealed that Cochran's Q = 32.40, P � 0.01 and I 2 = 84.6%, indicating that significant heterogeneity was observed in the included studies. The result of meta regression ( Table 2) indicated that sample types might be one possible source of heterogeneity. Sensitivity analysis showed that removal of any study did not alter the significance of the pooled effect size except the study of Zabicka (S2 Fig). After excluding this study, the I 2 value for heterogeneity decreased to 45% (Fig 2B). According to the method described above, Deeks' funnel plot showed no substantial asymmetry (P = 0.279). Therefore, publication bias was excluded (Fig 3).

Methodological quality
Using RevMan 5.3, the overall methodological quality of the included studies is shown in Fig  4. Patient selection and the index test mainly contribute to the risk of bias. In patient selection domain, we assessed four studies as taking a high risk for bias, because they didn't enroll participants randomly or consecutively, and one had a case-control design [9]. In the field of the index test, two studies were assessed to be high risk for bias: one index test did not use a prespecified threshold, and the other was explained with prior knowledge of the reference standard results. In the reference standard area, most studies had a low risk of bias, as they stated that the results of the reference standard were interpreted without knowing the index test results. Judging from the index test, the flow and timing of the risk of bias were relatively low.
There was no concern about the assessment of applicability for nine studies in the patient selection, the index test and reference standard domain.

Merge analysis results
The pooled sensitivity, specificity, PLR, NLR and DOR of GeneXpert VanA/VanB of each group were shown in  5). As Fagan' s nomogram showed, when the pre-test probability was set to 50%, the PLR of the upper diagonal was 24 and the post-test probability was 96%. Correspondingly, the NLR of the lower diagonal was 0.01 and the post-test probability was 1% (Fig 6).

Discussion
To the best of our knowledge, this is the first meta-analysis accessing the overall diagnostic accuracy of GeneXpert vanA/vanB. In this study, we did a thorough search using strict

PLOS NEGLECTED TROPICAL DISEASES
screening criteria, and finally, including 8 articles, groups in different reference standards. The results of our study indicate that GeneXpert vanA/vanB assay has a high diagnostic accuracy. Its excellent sensitivity (0.96, 95% CI, 0.93-0.98), specificity (0.90, 95% CI, 0.88-0.91) and DOR (440.77, 95% CI, 37.92-5123.55) made it an attractive option for routine surveillance of VRE in the future. The combined PLR and NLR were 16.44 (95%CI, 3.66-73.86) and 0.04 (95%CI, 0.00-0.32), respectively, suggesting that GeneXpert vanA/vanB has a brilliant capacity to diagnose and exclude a VRE. The SROC AUC was 0.9882, which is close to 1, indicating a high ability for VRE detection. Thus, GeneXpert vanA/vanB showed a very good diagnostic accuracy. Fagan's nomogram showed the clinical application value of GeneXpert vanA/vanB in various situations.
We also conduct a study on GeneXpert vanA/vanB diagnosis discrepancy between vanA and vanB. The combined sensitivity, specificity, PLR, NLR, DOR of the vanA group were higher than those of the vanB group. Furthermore, the pooled NLR was lower, revealing Gen-eXpert vanA/vanB is more accurate in diagnosis on vanA.
That there were more false-positive results in vanB group may be attributed to the presence of genes in several species of aerobic and anaerobic bacteria that were highly similar to the vanB sequences [5,7]. It is inevitable for the reason these bacteria also exist in the human [21]. The culture method for all clinical E. faecium isolates may neither be feasible nor cost-efficient in the setting of every routine lab, which makes it impossible to make a clear decision about the need to isolate the patient. Hence, supplementary tests are needed for further investigating [22,23].
Sensitivity analysis demonstrated that the study of Zabicka contributes to heterogeneity. It could be influenced by the factor that the experiment performed during a VanA E. faecium outbreak, as the report of Dekeyser et al. [24], and none of the patients was colonized with VanB enterococci. Several FP vanB results may be concerned with the specimen type, stool     swabs. Stool and rectal swabs might be the harbors where anaerobic microbes were commonly checked, which increased the risks of detecting false-positive vanB results [11]. The meta regression also confirmed the specimen types might be one of possible sources of heterogeneity. The discrepancies between GeneXpert vanA/vanB detecting vanA and vanB might be a source of heterogeneity. Restrained by only two studies conducting both experiments on vanA and vanB detecting simultaneously or separately, a further analysis is required for more data. There were still other variables that required to be explored, such as relevant description of patients. The sources and characteristics of patients were quite distinguished. However, sources of heterogeneity could not be formally explored for most tests because few studies were available for further evaluation.
The present study has several limitations. First, remarkable heterogeneity was observed in this study. To account for this heterogeneity, a random effects model was used to synthesis the identified studies in our analysis, which potentially increases the probability of type I error. Moreover, the results of meta regression and the sensitivity analysis were attempted to explain that detected sample could partly explain the source of heterogeneity. Subgroup analysis is looking forward to with more updating data. Second, our study also confirmed the observation of other authors that the GeneXpert vanA/vanB test has a low positive predictive value (PPV) for vanB enterococci [25]. Combining additional detection technologies may represent a pragmatic solution to increase VRE detection rates. Finally, we only retrieved published literature from four English databases. Only included studies written in English may have affected our findings. Despite comprehensive searches, the influence of unpublished positive results on the overall results could not be eliminated.

Conclusion
In summary, GeneXpert vanA/vanB has a high accuracy diagnosing VRE. Furthermore, Gen-eXpert vanA/vanB shows more accuracy when diagnosing vanA. Additional test is needed for further detecting VanB.