METHODS: Using a multicenter retrospective cohort of adults admitted between March 2, 2020 and February 11, 2021 for the treatment of SARS-CoV-2, we used a series of three-level logistic regression models to estimate the probability of receiving each of several antibiotics of interest. For each antibiotic, we then limited our data set to patient-days at intermediate probability of receipt and used augmented inverse-probability weighted models to estimate the average treatment effect of one daily dose-equivalent, compared to management without that daily dose-equivalent, on the probability of hospital-onset CDI. RESULTS: In 24,406 patient-days at intermediate probability of receipt, parenteral vancomycin increased risk of hospital-onset CDI, with an average treatment effect of 0.0096 cases per daily dose-equivalent (95% CI: 0.0053-0.0138). In 38,003 patient-days at intermediate probability of receipt, cefepime also increased subsequent CDI risk, with an estimated effect of 0.0074 more cases per daily dose-equivalent (95% CI: 0.0022-0.0126). CONCLUSIONS: Among common empiric antibiotics, parenteral vancomycin and cefepime appeared to increase risk of hospital-onset CDI. Causal inference observational study designs can be used to estimate patient-level harms of interventions such as empiric antimicrobials.