A Variational Bayesian Approach for Dosage Regimen Individualization

Abstract Clinical trials generate a vast amount of data which is rarely exploited for improving therapeutic treatments. A Bayesian approach has been developed by Lainez et al. (2011) for determining individualized dosage regimens. This method combines knowledge of the population of patients (the Prior, which is constructed from the clinical trials data) with measurements from the patient (the Likelihood) to produce a patient specific probability distribution (the Posterior). This posterior is then used to define dosage regimens for which the drug concentration in the blood is kept within the therapeutic window at a given confidence level. Monte Carlo approaches used to perform this analysis can be computationally demanding. Hence, we propose an alternative optimization procedure based on variational Bayesian inference. A computational performance study comparing these two approaches is reported using pharmacokinetic data from Gabapentin, a therapeutic agent for treating convulsive diseases. Since the Bayesian approach has been also used in developing kinetics models for catalytic and polymerization applications, the conclusions of this study may have general relevance beyond the pharmacokinetic domain.