This review concluded that real-time PCR had a high sensitivity and specificity to confirm Clostridium difficile infection. Overall diagnostic accuracy was variable and dependent on the prevalence of C. difficile. The review process was well-conducted but concerns over the methods used for the analysis given the heterogeneity across the studies mean the results should be treated with caution.

To determine the accuracy of real-time polymerase chain reaction (PCR) for the diagnosis of Clostridium difficile in endemic populations.

PubMed, Web of Science, CINAHL, The Cochrane Library and SCOPUS databases and Ovid were searched without language restrictions from 1995 to September 2010; search terms were reported. Reference lists of included studies and relevant reviews were searched. Experts and commercial PCR manufacturers for C. difficile diagnosis were contacted for additional studies.

Studies that evaluated the accuracy of real-time PCR conducted on stool specimens from in-patients or outpatients for diagnosis of C. difficile using cell culture cytotoxicity neutralization assay (CCCNA) or anaerobic toxigenic culture to confirm diagnosis in all patients were eligible for inclusion. Studies had to report sufficient data to construct 2x2 tables of test performance. Studies were excluded if they focused on detecting the virulence of different C. difficile ribotypes using real-time PCR.

All of the studies were conducted in patients suspected of infection rather than screening populations. Most studies used CCCNA as the reference standard, manual extraction and SmartCycler PCR. Prevalence of C. difficile ranged from 5.7% to 51.8%. Several studies (mostly older ones) used in-house assay designs that were not USA Food and Drug Administration approved.

Two reviewers independently selected studies for the review; disagreements were resolved by discussion.

Two reviewers independently assessed study quality using the 11-point adapted version of the QUADAS tool; disagreements were resolved by discussion.

Two reviewers independently extracted data to produce 2x2 tables of test performance on a per sample basis from which sensitivity, specificity, positive and negative likelihood ratios (LR+/-) and diagnostic odds ratios with corresponding 95% confidence intervals (CI) were calculated. Study authors were contacted for additional information. Disagreements were resolved by discussion.

Pooled estimates of sensitivity, specificity LR+ and LR- were calculated using a random-effects model. Heterogeneity was assessed using Χ² and Ι². A summary receiver operating characteristic curve (SROC) was calculated using the Moses-Littenberg model and the area under the curve was calculated. Acceptable test criteria were defined as 90% for sensitivity when estimated at its 25th centile and 3% false positive rate estimated at its 25th centile for specificity. Subgroup analyses were conducted for different reference standards and prevalence of C. difficile.

Nineteen studies met the inclusion criteria (7,392 stool samples, range 56 to 2,296). All studies met eight or more of the quality criteria; approximately 80% recruited a representative patient sample and 60% avoided progression bias. Given the inclusion criteria for the review, all studies avoided verification bias and used an appropriate reference standard. All studies reported on uninterpretable results and withdrawals. The authors stated that in most studies interpretation of the index test and reference test were not performed independently but the graphic indicated that all studies avoided incorporation bias.

Overall sensitivity was 90% (95% CI 88% to 91%), specificity was 96% (95% CI 96% to 97%), LR+ was 26.9 (95% CI 20.8 to 34.7), LR- 0.11 (95% CI 0.08 to 0.15) and the diagnostic odds ratio was 278.2 (95% CI 213.6 to 362.5). The area under the curve was 0.98 (95% CI 0.98 to 0.99). All the analyses showed considerable heterogeneity (Ι²>50%) except the diagnostic odds ratio (Ι²=2.3%).

When anaerobic toxigenic culture was the reference standard (eight studies, 4,833 samples) sensitivity was 90% (95% CI 88% to 92%), specificity was 96% (95% CI 95% to 96%), LR+ was 32.4 (95% CI 20.9 to 50.2), LR- 0.12 (95% CI 0.08 to 0.17), diagnostic odds ratio was 271.2 (95% CI 196 to 375.2) and the area under the curve was 0.98 (95% CI 0.98 to 0.99).

When CCCNA was the reference standard (15 studies, 4,098 samples) sensitivity was 89% (95% CI 87% to 92%), specificity was 97% (95% CI 96% to 97%), LR+ was 24.7 (95% CI 17.2 to 35.5), LR- was 0.10 (95% CI 0.06 to 0.16), diagnostic odds ratio was 323.7 (95% CI 190.2 to 550.6) and the area under the curve was 0.99 (95% CI 0.98 to 1.00).

Real-time PCR had a high sensitivity and specificity to confirm C. difficile. Overall diagnostic accuracy was variable and dependent on the prevalence of C. difficile.

The review addressed a clear question supported by well-defined inclusion criteria. Several relevant sources were searched without language restrictions and with attempts to identify unpublished studies. Diagnostic filters were used during the search so some studies may have been missed. Each stage of the review process was conducted in duplicate, which reduced risks of error and bias. Study quality was assessed using appropriate criteria but the results were neither reported in full nor taken into account in the analyses. All the studies reported results on a per sample basis; there was no indication of the number of patients included in the studies from which the samples were obtained. There were discrepancies between the text and figures/tables. Estimates of sensitivity and specificity were derived from clinically heterogeneous studies using separate frequentist meta-analyses. The Moses-Littenberg model was used for the SROC so estimates of sensitivity and specificity had to be calculated separately using standard meta-analytical techniques on heterogeneous data; more robust SROC models are available.

The review process was well-conducted but concerns over the methods used for the analysis given the heterogeneity across the studies mean the results should be treated with caution.

The authors did not state any implications for practice and research.

None reported.

This is a critical abstract of a systematic review that meets the criteria for inclusion on DARE. Each critical abstract contains a brief summary of the review methods, results and conclusions followed by a detailed critical assessment on the reliability of the review and the conclusions drawn.