This substudy used data from the cluster-randomized MedSafer trial to evaluate whether stopping outpatient proton pump inhibitors (PPIs) at hospital discharge affected the 30-day risk of Clostridioides difficile infection (CDI). Observational evidence has suggested PPIs are a risk factor for initial and recurrent CDI, but whether deprescribing PPIs reduces incident CDI was unknown. The authors report an unexpected association between PPI deprescription and higher odds of 30-day CDI.
The analysis was a secondary, quasi-interventional substudy using the MedSafer clinical trial dataset. The parent trial enrolled hospitalized older adults prescribed five or more medications and randomized participants to usual care (a medication review) or electronic deprescribing decision support prior to hospital discharge. For this substudy, all trial participants were included irrespective of randomization arm. The total sample comprised 4,990 patients.
Participants were categorized on the basis of PPI use at hospital admission and at discharge. Two key exposure groups are reported:
All other patients formed comparator categories based on their PPI status at admission and discharge, as defined by the authors.
The primary outcome for this substudy was occurrence of Clostridioides difficile infection (CDI) within 30 days following hospitalization/discharge. Across the full cohort, 32 patients experienced CDI within 30 days (0.6%).
The association between PPI exposure status and 30-day CDI was modeled using a Bayesian mixed-effect logistic regression adjusted for prespecified confounders. The model produced adjusted odds ratios (aORs) with 95% credible intervals (95% CrI) and reported the posterior probability that deprescription was associated with increased odds of CDI.
PPI deprescription (stopping a home PPI by discharge) was associated with an adjusted odds ratio for 30-day CDI of 3.66 (95% CrI 1.25–10.11). The analysis yielded a 99.1% probability that deprescription increased the odds of CDI.
New PPI use initiated during hospitalization was not associated with a probable increase in 30-day CDI: adjusted odds ratio 1.14 (95% CrI 0.20–4.67).
Overall, 32 of 4,990 patients (0.6%) developed CDI within 30 days.
The principal finding—an increased odds of CDI after PPI deprescription—was unexpected because clinical rationale for deprescribing PPIs often includes reducing CDI risk based on prior observational associations. The authors highlight that this result may not reflect a causal protective effect of ongoing PPI therapy. They explicitly raise confounding by indication as a plausible explanation for the association: patients selected for deprescription or those in whom PPIs were stopped may differ systematically from those who continued therapy in ways not fully accounted for by adjustment.
The authors call for mechanistic investigations and confirmatory studies to determine whether the observed association is real, causal, or artefactual.
The report emphasizes the unexpected nature of the association and the possibility that results are artefactual due to residual or unmeasured confounding. As a secondary analysis of trial data and a preprint report, the findings require replication. The authors recommend further mechanistic and confirmatory research to clarify whether PPI deprescription influences CDI risk and to identify underlying explanations if the association is confirmed.
This preprint reports that the Research Ethics Board at the Research Institute of the McGill University Health Centre provided ethical approval. The parent MedSafer trial is registered (NCT03272607). Funding was declared from the Canadian Institutes of Health Research. Two authors (Drs. Lee and McDonald) hold copyright to MedSafer in conjunction with McGill University. The authors state that all data produced are available upon reasonable request.
Notes and caution: This manuscript is a medRxiv preprint and has not been peer reviewed; the authors and platform note it should not be used to guide clinical practice. The article reports the results and statistical outputs described above; the source did not provide additional mechanistic data, subgroup analyses, or other numeric detail beyond what is summarized here.