671 - Predicting Individual Volume of Distribution and Clearance of Gentamicin in Neonates

Poster Number: 671
Publication Number: 671.241

Mane Sargsyan, Brookdale University Hospital Medical Center, BROOKLYN, NY, United States; Dushyant Mukkamala, Yale School of Medicine, Milford, CT, United States; Min Ji Kim, Brookdale Hospital Medical Center, Brooklyn, NY, United States; Roger Kim, Brookdale University Hospital and medical Center, Brooklyn, NY, United States; Fernanda E. Kupferman,, One Brooklyn Health @ Brookdale Hospital Medical Center, Brooklyn, NY, United States

Background: Gentamicin is an antibiotic used as an antimicrobial treatment in neonatal units. Nonlinear mixed-effect pharmacokinetic (PK) models for gentamicin concentration rely on patient-specific, but directly non-measurable variables, such as volume of distribution (Vd) and gentamicin clearance (Cl). For gentamicin dosing decisions, one needs to proactively estimate Vd and Cl using direct measurements of covariates. Empirical models disagree in the estimation of Vd and Cl, leading to varying estimates of gentamicin concentration and consequently biasing the dosage decisions.

Objective: To infer Vd and Cl from retrospective data. To compare established empirical models from literature. To develop linear regression model between measured covariates and the patient-specific Vd and Cl.

Design/Methods: We used retrospective data of 516 neonatal patients who received gentamicin between 2016 and 2021 at the Brookdale NICU. We employed Python to implement alternative empirical models from literature and a one-compartment, first-order PK model with its fitting procedure. The empirical models were compared to the inferred values of Vd and Cl using R-squared values. We also developed a linear regression model to predict Vd and Cl using the measured predictor variables and statistical libraries in Python. Figure 1 demonstrates the overall workflow.

Results: The fitted solution curves of a standard one-compartment first-order PK model are shown in Figure 2(a). Each of the curves corresponds to an individualized Vd and Cl. The mean and standard deviation values for the inferred variables are Vd = 0.90+-0.37 L/kg, and Cl = 0.1+-0.03 L/kg/hr. We then compared seven alternative empirical models, listed in the review paper [Crcek et al., 2019] with the inferred Vd and Cl values. Further, we developed a linear regression model using six covariates to predict Vd and Cl with R-squared values 0.46 and 0.67, respectively. Figure 2(b) shows the comparison of our linear regression model with the best performing empirical model from [Botha et al, 2003]. As evident by the R-squared values, the linear regression model outperformed Botha model and consequently all empirical models for our dataset. The linear regression model also allowed a comparison of the impact of the covariates on the gentamicin concentration level curves. Impact was measured by weighted regression coefficients, listed in Table 1.

Conclusion(s): Linear regression model using six covariates outperformed established empirical models. The most substantial impacts on gentamicin concentration curves are attributed to Gestational Age, Birth Weight, and Apgar5 score.