P-1253. Validating and Comparing the Predictiveness of Cefepime Pharmacokinetic Models Used in Bayesian Dosing

Abstract Background Bayesian software may enhance beta-lactam dosing. Therefore, we analyzed the predictiveness of cefepime pharmacokinetic (PK) models for use as Bayesian priors. Methods Data on adult patients receiving cefepime intravenous therapy were entered into a Bayesian software at 2 sites. Inclusion criteria encompassed patients with ≥ 1 cefepime drug concentration (TDM), including those undergoing continuous renal replacement therapy. Exclusion criteria were dialysis and outlier TDMs defined as > 3 standard deviations above the mean. Adult parametric PK models were identified and implemented using PKPDsim. Measured TDMs were predicted iteratively, using model priors to predict the first set (“a priori”), with subsequent TDMs predicted using Bayesian estimation (“a posteriori”), to mirror the clinical experience (i.e., initial predictions were made using only demographics; updated predictions were made using each additional TDM event). TDMs were analyzed separately as peaks or as random/trough levels. Mean percent error (MPE) assessed bias and root mean square error (RMSE) assessed imprecision. Statistical significance was determined using overlapping 90% confidence intervals from 1000 bootstrap simulations. Results A total of 114 patients with 200 TDMs for a priori and 35 TDMs for a posteriori predictions were included (Table 1). Five PK models were identified (Table 2). The An et al. and Jonckheere et al. models were most predictive, with acceptable RMSE ≤ 25 mg/L and bias < 25% (Fig 1.). The An et al. model was most precise a priori (RMSEpeak: 23 mg/L, RMSErandom/trough: 19 mg/L versus other models ranging from 26-33 mg/L and 23-30 mg/L) but demonstrated bias in predicting peaks a priori (MPE: 24%) that was not improved a posteriori. The Jonckheere et al. model demonstrated similar precision a priori to the An et al. model and showed better precision in predicting peaks a posteriori (RMSEpeak: 6 mg/L versus other models ranging from 13-25 mg/ L; Fig 1.). Conclusion While many models performed similarly (overlapping 90% CI), the An and Jonckheere models both demonstrated acceptable bias and precision for predicting random/trough levels. Centers using a random/trough sampling approach to estimate cefepime exposures may find these models clinically useful. Disclosures Maria-Stephanie Hughes, PharmD, InsightRX: Employee of company|InsightRX: Stocks/Bonds (Private Company) Nathaniel J. Rhodes, PharmD MS, Apothecademy, LLC: Advisor/Consultant Brandon Smith, MD, PharmD, Melinta Therapeutics: Advisor/Consultant|Shionogi, INC: Advisor/Consultant Ryan K. Shields, PharmD, MS, Allergan: Advisor/Consultant|Cidara: Advisor/Consultant|Entasis: Advisor/Consultant|GSK: Advisor/Consultant|Melinta: Advisor/Consultant|Melinta: Grant/Research Support|Menarini: Advisor/Consultant|Merck: Advisor/Consultant|Merck: Grant/Research Support|Pfizer: Advisor/Consultant|Roche: Grant/Research Support|Shionogi: Advisor/Consultant|Shionogi: Grant/Research Support|Utility: Advisor/Consultant|Venatorx: Advisor/Consultant|Venatorx: Grant/Research Support Jasmine Hughes, PhD, InsightRX: Employee|InsightRX: Stocks/Bonds (Private Company)