KQ1. Diagnostics

We used an updated rubric for assessing the quality of included studies (QUADAS-2). Overall, 12 of 37 studies were “low risk of bias” in all 4 QUADAS-2 domains (patient selection, index test, reference standard, and flow and timing). In keeping with the previous report, most studies that were included in this report that were not included in the original report enrolled samples from patients at risk for or with symptoms consistent with CDI. However, some studies included enrolled unformed specimens only irrespective of whether testing for CDI was requested by the patient's clinician. The clinical characteristics of the patients from whom fecal samples were obtained for inclusion in the included studies were generally not described, making determination of applicability of findings problematic. While the characteristics of patients from whom fecal specimens were obtained for inclusion in the study were often not described, most studies (26) included only unformed stools samples while two studies contained both formed and unformed specimens and nine studies did not specify whether samples were formed or unformed. Nineteen studies did not include repeat samples from a single patient, but 18 studies included more samples than patients or did not specify the number of patients.

In contrast to the previous report, we included studies that prospectively enrolled samples from a patient population with a “baseline” pre-test probability of CDI without modification of the probability of disease by a screening test. The prevalence of CDI in the studies varied widely, between 6 percent and 48 percent. While this variability may not have an impact on sensitivity and specificity, the positive and negative predictive values of included tests are not applicable to a population with different prevalence than the prevalence of CDI in an included study. Seventeen studies enrolled a random or consecutive sample of samples, 20 studies did not specify if a consecutive or random sample of patients was included, and three studies did not include a random or consecutive sample of specimens. The impact of enrolling nonconsecutive samples on the measured operating characteristics of a certain diagnostic test is unclear. We cannot exclude the possibility that a study that had a nonconsecutive sample of patients could systematically entrain bias if there were characteristics of that led to samples being included and others excluded, such as volume of stool, variability of testing practices in certain wards, or other characteristics.

Similar to the previous report, we found that there were few concerns in the conduct and interpretation of index tests with respect to risk of bias. However, there was significant heterogeneity in the studies and the source of this heterogeneity in observed operating characteristics for the included studies is not completely clear. Many studies did not apply different tests to the exact same number of patients and the reasons for these differences were not often specified. There was some variability in how invalid or inconclusive index test(s) were interpreted and if the index test(s) were repeated on invalid or inconclusive specimens. The previous report included studies with a combination of reference standards including cell cytotoxicity test, cell cytotoxicity test in conjunction with toxigenic culture, one used a toxin immunoassay in conjunction with toxigenic culture, multiple immunoassays for toxins A and B in conjunction with toxigenic culture, and in-house gene detection tests. In the current update report, we used a more stringent reference standard of the cell cytotoxicity assay, toxigenic culture, or a combination thereof. A few studies used enriched toxigenic culture as the reference standard which is likely a more sensitive reference standard that typical toxigenic culture or cytoxicity assay; the logical consequence is that index tests may appear less sensitive when compared against a more sensitive reference standard. Thirty studies used toxigenic culture as the reference standard, five studies used a composite reference standard of cell cytotoxicity assay and/or toxigenic culture, and two studies used cell cytotoxicity assays as the reference standard. Although regarded as an acceptable reference standard, toxigenic culture, cell cytotoxicity assay or a combination thereof are not perfectly accurate. In the majority of included studies the diagnostic tests were performed independently although it was usually not explicitly stated whether or not the tests were evaluated without knowledge of the other tests. However, it was inferred that most index tests (which are more rapid than the reference standards that take 24-48 hours) were interpreted prior to the results of the reference test being available.

Nineteen studies were “high risk of bias” with respect to flow and timing, mostly due to not all samples being included in the analysis. While the number of indeterminate results was generally small, small changes in a 2×2 table for a certain study can have marked changes in the calculated operating characteristics. As in the previous report, the handling of indeterminate or inconclusive results is problematic. One approach many investigators used was to exclude the inconclusive tests from the calculation of the operating characteristics of a certain test, while others repeated the index test and used the second result (if positive or negative) as the result used in the calculation of operating characteristics. The former approach may lead to an overestimation or underestimation of the sensitivity and specificity of a test depending on whether the reference standard result of the excluded samples is positive or negative. Further, this approach also may lead to the body of samples included being no longer consecutive or random. The latter approach may also lead a misestimation of the operating characteristics as the approach to inconclusive results likely varies significantly between laboratories.

Appendix Table F1. Diagnostic study quality

KQ2. Prevention

Appendix Table F2. Prevention study risk of bias

Appendix Table F3. Quality of previous systematic reviews

KQ3. Standard Treatment

Appendix Table F4. Standard treatment study risk of bias

KQ4. Nonstandard Treatment

Appendix Table F5. FMT adjunctive treatments study risk of bias

Appendix Table F6. Probiotic adjunctive treatments study risk of bias

Appendix Table F7. Other adjunctive treatments study risk of bias