Design: Multicentre retrospective observational study
Population: 157 adults on vancomycin, 63 with hepatic dysfunction and albumin below 2 g/dL
Primary outcome: ≥30% deviation between population-model and Bayesian AUC
Effect: Hepatic dysfunction: adjusted OR 2.26 (95% CI 1.10–4.62)

This multicentre retrospective study in Japan included 157 adults on vancomycin with measured trough levels, 63 of whom had hepatic dysfunction (cirrhosis, hepatic failure or other liver disease) with albumin below 2 g/dL. It compared AUC estimated from population pharmacokinetic parameters with AUC estimated by Bayesian methods using the measured troughs.

Hepatic dysfunction was associated with a 30% or greater difference between the two estimates (adjusted OR 2.26, 95% CI 1.10–4.62). Patients with liver dysfunction had higher actual exposure and higher trough ratios than predicted.

The mechanism is familiar: low creatinine production makes kidney function and vancomycin clearance look better than they are. Dosing tools that rely on creatinine will therefore under-predict exposure and risk nephrotoxicity. The study is small and retrospective, but the advice it supports is low-cost and consistent with AUC-guided dosing.

In patients with liver disease and low albumin, measure vancomycin levels early, within the first doses.
Do not rely on population-model AUC estimates alone in hepatic dysfunction.
Expect higher exposure than predicted, and adjust doses downward if levels confirm it.
Monitor creatinine and urine output closely for nephrotoxicity.
Apply the same caution to other renally cleared, narrow-index drugs.

Why it matters

Model-based vancomycin dosing assumes a creatinine that liver disease makes unreliable.

The statistics, in plain English

An adjusted odds ratio of 2.26 means more than double the odds of a large mismatch between predicted and measured exposure. The interval runs from 1.10 to 4.62, so the true effect could be modest or large. With 157 patients from a retrospective review, this is a signal, not a precise estimate.