Kidney injury, dialysis, and mortality with vancomycin plus piperacillin-tazobactam or cefepime

Retrospective studies have consistently reported higher rates of acute kidney injury (AKI) when vancomycin is combined with piperacillin-tazobactam compared with other β-lactams, raising concerns about nephrotoxicity. However, recent prospective studies and randomized trials have not demonstrated increased risk of clinically meaningful kidney injury, suggesting that creatinine-based endpoints may overestimate harm. We applied a multistate modelling approach to evaluate whether vancomycin-piperacillin/tazobactam is associated with progression to severe renal outcomes beyond creatinine-defined AKI.

We conducted a retrospective cohort study using the MIMIC-IV database, including adult hospitalizations receiving vancomycin with either piperacillin-tazobactam or cefepime. A multistate model was used to evaluate transitions from antibiotic initiation to creatinine-defined AKI, dialysis, and in-hospital death. Cox proportional hazards models were fitted for each transition, with inverse probability of treatment weighting to adjust for confounding.

Among 10 490 hospitalizations, creatinine-defined AKI occurred more frequently with vancomycin-piperacillin/tazobactam than with vancomycin-cefepime (23.1% vs. 15.2%). In multistate models, piperacillin-tazobactam was associated with a higher hazard of AKI (inverse probability of treatment weighting-adjusted hazard ratio 1.63, 95% confidence interval 1.48-1.80) but not with progression to dialysis (hazard ratio 0.77, 95% confidence interval 0.50-1.19). Risks of death were similar between groups, while death following AKI was less frequent among piperacillin recipients.

Although vancomycin-piperacillin/tazobactam was associated with increased creatinine-defined AKI, this did not translate into higher risks of dialysis or death. These findings support the hypothesis that observed creatinine elevations are consistent with pseudo-nephrotoxicity rather than clinically meaningful kidney injury and highlight the value of multistate modelling for interpreting drug safety signals.