Development of a model to predict lean liver volume (LLV) for use in scaling drug clearance

Introduction: Clearance (CL) is the most important parameter to describe the relationship between dose and exposure for drugs dosed chronically. CL, for hepatically-cleared drugs, is known to correlate with liver size [1]. Theoretically, lean liver volume (LLV), the liver volume that excludes all fat, represents the size of the metabolically active part and may scale better to CL than total liver volume. Currently, there is no model available for prediction of LLV. The aim of this work was to develop a predictive model for LLV from measurable patient characteristics along with clinical chemistry measures.

Method: Total liver volume and liver fat measurements obtained through computed tomography (CT) for 100 adult Indian healthy subjects were available for analysis. Covariate data included age, sex, weight, height, and various other clinical and laboratory measurements. Subjects were stratified according to BMI (< 25, 25-30, and > 30 kg/m2). Lean liver volume (LLV) was measured by subtracting liver fat from the total liver volume. A model was developed for LLV

using non-linear mixed effects modelling using NONMEM v7.3. Stepwise covariate modelling (SCM) was performed by forward selection (p value < 0.05) followed by backward deletion (p value < 0.01). Relevant size descriptors i.e. total body weight (WT), fat-free mass (FFM), and various clinical markers for lipid profile (e.g. serum triglycerides), liver enzymes (e.g. serum transaminases), and diabetes status (e.g. fasting blood glucose) were tested as potential covariates. The final model was evaluated using the likelihood ratio test and visual predictive checks.

Results: The model with best fit for LLV was a linear function of body size being scaled allometrically, and no other clinical markers were found statistically significant. FFM, in conjunction with sex described the inter-individual variability to a similar extent as shown by WT. FFM was still selected for the final model with an estimated allometric exponent of about 1 as it was preferred statistically over WT in the obese stratum. The mean error (ME) and the root mean squared error (RMSE) of the final model predictions were <2% and <20% of the median observed LLV respectively.

Conclusion: A model to predict LLV from patient characteristics was developed and evaluated. FFM and sex were identified as the predictors of LLV. The LLV model should be evaluated as a potential scaler for drug dose individualisation.


  1. Murry DJ, Crom WR, Reddick WE, Bhargava R, Evans WE. Liver volume as a determinant of drug clearance in children and adolescents. Drug metabolism and disposition. 1995;23(10):1110-6.