Reply to Planche et al

To the Editor—We appreciate the remarks of Planche et al [1] regarding our manuscript [2]. There is no consensus regarding the optimal laboratory method for diagnosis of Clostridium difficile infection (CDI) [3]. A recent systematic review [4] noted high sensitivity and specificity for both multistep testing (incorporates detection of C. difficile glutamate dehydrogenase, toxins A/B, and/or polymerase chain reaction [PCR] for the tcdB gene) and single-step testing using PCR alone. Multiple studies have examined the association between detection of stool toxin and severity/prognosis and most, but not all, have found such an association [5–12].

The results of external studies may not apply locally when practices differ. We conducted a single-center study to internally validate the putative association between fecal toxin detection and adverse outcomes [6]. Although we did not find an association, as we discussed [2] and as Planche et al remarked [1], there was a low prevalence (37.2%) of detectable fecal toxin among cases by enzyme immunoassay (EIA) and of severe CDI (7.9%). Coupled with our institution's policy of collecting stool in Cary–Blair media during the study period, this raised concerns about low sensitivity and, thus, the generalizability of these findings to more sensitive testing strategies. Rather than suggest that toxin detection is not relevant to prognosis, the central message of our analysis is that inherent assay sensitivity is an important consideration, and dilution in Cary–Blair may not be optimal. Furthermore, the low observed severity could have reduced power to detect an effect or even have been a consequence of the testing strategy, which is unable to reject formed stool; our microbiology laboratory has now eliminated collection in Cary–Blair media. As the cytotoxicity assay, used in one of the largest studies to date on the topic [6], can be up to 3 orders of magnitude more sensitive for fecal toxin than EIAs [13], our study suggests that a true association between fecal toxin presence and clinical course necessitates sensitive testing that may or may not be offered by current EIA-based methods.

In regard to their comments about ribotype 027 infection and outcomes [1], we believe Planche et al are conflating our 2 distinct study goals: testing hypotheses and building predictive models. Our findings actually agree with previous ones, affirming that ribotype 027 infection is independently associated with both complicated CDI and 30-day all-cause mortality after adjustment for patient-level variables. However, this association was not enough to contribute meaningfully to the predictive model for complicated CDI. Predictive modeling does not seek to test hypotheses, but rather aims to identify which variables are most important, taken together, in predicting an outcome. Our analysis suggests that even though an association with ribotype 027 is seen, other variables such as age, concurrent antibiotic use, and serum bilirubin are more important for the clinician faced with management decisions regarding the outcome of complicated CDI.

In summary, we believe that our study underscores the need to evaluate the real-world performance of toxin-detection strategies given variable local testing practices. It also affirms that ribotype 027 has an independent association with prognosis in CDI, but suggests that other variables may be of greater importance to clinicians.