Subgroup Analyses in Randomized Trials of Clostridioides difficile–Associated Disease: Recurring Challenges With Analyzing Recurrence

To the  Editor—The trial comparing ridinilazole to vancomycin in treatment of symptomatic Clostridioides difficile–associated disease (CDAD) failed to demonstrate superiority of ridinilazole on sustained clinical response (SCR)—a combination of initial response with no recurrence. Yet a subgroup analysis claims benefit on disease recurrence prevention [1]. Analysis of secondary endpoints is challenging with failure on primary endpoints due to increased rate of false-positive results from multiple hypothesis testing [2]. The recurrence analysis also highlights potential biases in postrandomization subgroups presented in several CDAD trials.

Yusuf et al. define an improper subgroup as trial participants classified by factors measured after randomization [3]. Postrandomization subgroups are particularly prone to bias because their composition can be influenced by effects of interventions being tested, causing imbalances between groups on patient prognostic characteristics. Patients are no longer randomly distributed between test and control groups, meaning that subgroup analyses may be misleading. Observed results may be attributable to confounding rather than the interventions.

While initial resolution and disease recurrence are different outcomes, they are intrinsically linked. Patients cannot experience recurrence unless they survive and experience initial symptom resolution, analogous to competing risks in time-to-event analyses, seen commonly in oncology and cardiology trials [4, 5]. Since survival and initial success occur after randomization and are influenced by the tested interventions, recurrence analyses meet the definition of postrandomization subgroup analyses.

Postrandomization recurrence subgroups are particularly problematic when there are differences in initial response [3]. In the ridinilazole trial, as with past trials of tolevamer and metronidazole, there were numerical differences in initial success [1, 6]. Patients with poorer baseline prognosis are less likely to achieve initial success, meaning more patients with better baseline prognosis likely remain in recurrence analyses for the test groups. A control drug with better initial efficacy includes patients with better and worse prognoses in the recurrence subgroup. As an extreme example, drugs that cure no one initially would have a zero-recurrence rate.

This is not a new issue—several trials in treatment of symptomatic CDAD spanning decades have presented postrandomization recurrence subgroup analyses including those evaluating metronidazole, tolevamer, fidaxomicin, and ridinilazole [1, 6–8]. The tolevamer trial did point out potential selection bias in the analysis [6]. The fidaxomicin trial used “global response” (analogous to SCR) as a hypothesis-generating secondary endpoint, showing superiority to vancomycin [8]. Superiority on sustained response was confirmed in a second trial [9]. There is a lack of independent confirmation when pooling trials as done with ridinilazole [1]. On the other hand, the US Veterans Administration–funded OpTION trial comparing fidaxomicin to short- and long-course vancomycin appropriately uses the primary endpoint of SCR and does not discuss postrandomization subgroup recurrence analysis in the protocol [10].

Future trials in symptomatic CDAD should use SCR as the primary endpoint and avoid invalid postrandomization subgroup analyses. SCR includes patients who do not achieve earlier response along with those who have later recurrence and does not bias results. This endpoint is clinically relevant as patients want optimal results—to survive, achieve symptom improvement, and avoid recurrence. Unbiased analyses help both patients and clinicians best apply trial results to clinical practice.

Notes

Acknowledgments. R. E. S. and J. H. P. authored the manuscript, and it was reviewed and edited significantly by S. R. I. and I. C. D.

Financial support. This work was supported (at least in part) with federal funds from the Intramural Research Program of the National Institute of Allergy and Infectious Diseases and the National Institutes of Health (NIH) Clinical Center (to R. E. S., S. R. I., and I. C. D.) and by the NIH Clinical Center and the National Cancer Institute, NIH (contract number 75N91019D00024, task order number 75N91019F00130 to J. H. P.).