The role of glutamate dehydrogenase for the detection of Clostridium difficile in faecal samples: a meta-analysis

Summary

Clostridium difficile causes a serious, occasionally fatal, hospital-acquired infection. The laboratory diagnosis of C. difficile infection (CDI) needs to be accurate to ensure optimal patient management, infection control and reliable surveillance. Commercial enzyme-linked immunosorbent assays for C. difficile toxins have poor sensitivity when compared with cell culture cytotoxin assay (CTA) and toxigenic culture (TC). We performed a meta-analysis of the role of glutamate dehydrogenase (GDH) in diagnosis of CDI. We analysed 21 papers, of which eight were excluded. We included publications of original research that used a ‘gold standard’ reference test (either CTA or TC). We also included publications that used culture without toxin testing of the isolate as a reference test even though this is not recognised as a gold standard. Exclusion criteria were failure to use a gold standard reference test and where the index test was used as the gold standard. Significant heterogeneity between study results justified the summary receiver operating characteristic (SROC) analysis. The meta-analysis demonstrated high diagnostic accuracy of GDH for the presence of C. difficile in faeces; when compared with culture it achieved a sensitivity and specificity of >90%. The SROC plot confirmed this finding. As a surrogate for toxigenic strains the GDH yields a specificity of 80–100% with a false positivity rate of ∼20%, as it detects toxigenic and non-toxigenic strains of the organism. However, GDH test has high sensitivity and negative predictive value and would be a powerful test in a dual testing algorithm when combined with a test to detect toxin.

Introduction

Clostridium difficile is the most common cause of antibiotic-associated colitis. The spectrum of disease can vary between a mild, self-limiting diarrhoea to life-threatening pseudomembranous colitis and toxic megacolon.¹ Currently the diagnosis of C. difficile infection (CDI) relies on the demonstration of C. difficile toxins directly in the faeces of an affected patient either by the use of a cell line cytotoxicity assay (CTA) or by enzyme-linked immunosorbent assay (ELISA) or immunochromatography (IC) and the results taken in conjunction with the patient’s clinical history.²

The CTA method is considered the ‘gold standard’ against which all other toxin detection methods are measured.³ However, there are limitations to the use of the CTA which include expense and longer turnaround time (minimum of 24 h). The CTA is now little used by diagnostic laboratories who prefer the cheaper, quicker and less technically demanding ELISA- or IC-based methods.

Toxigenic culture (TC; culture and isolation of C. difficile from the faeces followed by toxin testing of the actual isolate) has been proposed as an alternative gold standard but there are limitations to its use for diagnosis of CDI.⁴ The most important limitation is in patients who harbour a toxigenic strain of C. difficile but whose diarrhoea is caused by another agent. This may lead to overdiagnosis and unnecessary treatment. Furthermore, toxigenic culture will take a minimum of 48 h to complete.

The problem with the use of ELISA- and IC-based methods relates to their relative insensitivity and low positive predictive value.1, 5, 6 Cases of CDI are undoubtedly missed because of false negatives obtained with these tests. False positive results may also occur (although rarer than false negatives).⁷

Recently in an attempt to increase the accuracy of CDI diagnosis, some workers have included a test for C. difficile glutamate dehydrogenase (GDH, also known as C. difficile common antigen) as a preliminary screening test for detecting the presence of the organism in faecal samples and then, consequent on a positive result, following on with further tests for toxin or the presence of a toxigenic isolate. The detection of GDH indicates the presence of the organism in faeces and is not indicative of toxin production. Glutamate dehydrogenase is a constitutive enzyme produced in large amounts by all strains of C. difficile independently of toxigenicity. Glutamate dehydrogenase is, therefore, easily detected in faeces and makes a good screening marker.⁸ Modern test formats use monoclonal antibodies raised against the C. difficile-specific GDH, thus avoiding any cross-reactivity with GDH produced by other anaerobic bacteria.

The GDH test has been investigated by a number of workers and has been shown to be both sensitive and specific for the presence of the organism, with high negative predictive values reported.3, 8 Although the use of tests for GDH has been widely published there is still much confusion among microbiologists and healthcare staff concerning how the test should be used and its interpretation. The presence of a number of papers reporting the use of GDH in CDI diagnosis has stimulated us to perform a meta-analysis concerning the role of GDH in diagnosis of CDI.

We searched the English literature (using PubMed, Medline and the Cochrane Library) citing the terms ‘glutamate dehydrogenase’, ‘Clostridium difficile’, ‘diagnosis’, ‘testing algorithm’.

Inclusion criteria for entry into the meta-analysis were publications of original research on human clinical faeces, that used a gold standard reference test (either CTA or TC). We have also included publications that used culture (culture for C. difficile but without toxin testing of the isolate) as a reference test even though this is not recognised as a gold standard.

Papers were excluded from our analysis if there was no gold standard reference test used, or where the index test was used as the gold standard. Publications were also excluded if they used a reference test that was a composite of more than one test; where not all samples were tested against the gold standard reference test; or if CTA was used as the gold standard, but a neutralisation step for the toxin was not performed.

Sensitivity and specificity data were calculated as the major measures of outcome. These were calculated from the availability in each publication of true positives and negatives, and false positives and negatives, for each test method compared to the declared gold standard used.

In many studies the test for GDH was compared to CTA or TC where GDH was used as a surrogate marker for the presence of toxigenic C. difficile. Other workers compared GDH with culture (C) where the GDH test could be used as a rapid alternative to culture to indicate the presence of the organism.

Stata Statistical Software, release 9.2 (Stata Corp., College Station, TX, USA) was used for all statistical analyses. The performance of the GDH test was assessed against the following reported gold standard test methods: C, CTA and TC, although most studies carried out only one of these. We calculated sensitivity, specificity, positive and negative predictive values as measures of test performance, as well as local prevalence. These univariate trends were plotted as forest plots (along with 95% binomial confidence intervals) and summary measures presented for each type of gold standard used. To test for the validity of these summary measures we used a χ² approximation test of homogeneity as available on the tabodds command of Stata 9.2.

Even when the same technology is employed, the performance of a test can vary from study to study as a result of variations in the local conditions (organism genotype, physical conditions, etc.). The problem is that sensitivity is inversely related to specificity in a typically non-linear fashion. This can lead to gross underestimates of sensitivity and specificity when the overall average is taken across studies. As a solution it is generally recommended that we plot sensitivity against (1 – specificity) as a description of the test’s ability to produce true vs false positives respectively. This plot is known as the summary receiver operating characteristic (SROC), and its shape is a much more reliable indicator of test performance than any one sensitivity/specificity pair. We fitted a parametric model to the SROC curve using estimates of sensitivity and specificity from the studies included in the meta-analysis.⁹ We regressed the difference in the logits of sensitivity and 1 – specificity (outcome) against their sum (dependent variable), and used weighted linear regression to fit the model. As weights we used the total number of tests in each study, and where we had more than one gold standard per study we declared gold standard as nested within study in the regression algorithm. As a measure of model fit we used the regression r² as given in the STATA output.