Retrospective Bayesian Reanalysis of Single Gentamicin Concentrations: A Neonatal Case Series

TL;DR

This study evaluates how well Bayesian software predicts gentamicin levels in neonates using a single blood test, finding it reliable for trough levels but less so for peak levels.

Contribution

The study introduces the use of Bayesian estimation for gentamicin dosing in neonates, comparing its accuracy to traditional nomograms.

Findings

Bayesian estimation reliably predicts trough concentrations (Cmin) when using a single trough level.

Peak concentration (Cmax) predictions are less accurate and tend to overestimate true values.

Using a single peak level leads to overestimation of both Cmax and Cmin.

Abstract

Background/Objectives: Nomograms for adjusting gentamicin therapy in neonates using a single concentration are limited. The Dersch–Mills nomogram is inefficient for short-duration therapies, while the NeoFax nomogram is outdated based on the current American Academy of Pediatrics (AAP) guidelines. Bayesian software has shown accuracy for vancomycin in adults, but its performance for gentamicin in neonates is unclear. This study evaluates the accuracy of Bayesian estimation in predicting peak and trough gentamicin concentrations from a single measured level in neonates. Methods: A single-center, retrospective re-analysis was conducted of gentamicin concentrations in neonates. InsightRx® was used to estimate maximum and minimum concentrations in a dosing interval (Cmax and Cmin) based on a single peak or trough concentration. Bias and accuracy were characterized using the mean difference (MD) between estimated and measured concentrations and the 95% limits of agreement (LOA) for the differences (±1.96 × SD). Results: Fifty-seven neonates (73 peak/trough pairs) were analyzed. Median gestational age was 34 weeks and median postnatal age was 0 days. The MD (LOA) between Cmin estimates and measured troughs was 0.03 mg/L (−0.17 to 0.13) for the trough-only analysis and 0.21 mg/L (−0.38 to 0.8) for the peak-only analysis. The MD (LOA) between Cmax estimates and measured peaks was 0.16 mg/L (−3.2 to 3.3) for the trough-only analysis and 1.2 mg/L (−0.58 to 3.0) for the peak-only analysis. Conclusions: In neonates, a Bayesian analysis of a trough concentration produces reliable Cmin estimates but is not as accurate in estimating Cmax. Analyzing a peak concentration produces Cmax and Cmin values that overestimate true concentrations. If the goal of monitoring is to ensure sufficiently low troughs, a single-level analysis is reasonable if levels are drawn near the end of the dosing interval, but Cmin predictions based on levels drawn early in the dosing interval should be avoided.

Figures