Activity Number:
|
399
- Recent Developments in Precision Medicine
|
Type:
|
Topic Contributed
|
Date/Time:
|
Wednesday, August 5, 2020 : 1:00 PM to 2:50 PM
|
Sponsor:
|
Biopharmaceutical Section
|
Abstract #313501
|
|
Title:
|
Bayesian Pharmacokinetic Models for Dose Personalization
|
Author(s):
|
Daniel Lizotte* and Demetri Pananos
|
Companies:
|
University of Western Ontario and University of Western Ontario
|
Keywords:
|
bayes;
MCMC;
pharmacokinetics;
decision support
|
Abstract:
|
The ability to incorporate information prior to seeing data via bayesian models is valuable in applications like precision medicine where observations are difficult or expensive to obtain and accuracy of estimates is at a premium. Summarizing Bayesian model posteriors via maximum a posteriori (MAP) is a common method of using the Bayesian models for prediction in pharmacokinetics, however it has been noted that when the model is high dimensional maximum a posteriori can lead to less accurate predictions as compared to full Bayesian inference. We demonstrate this deficiency in the context of pharmacokinetics by first fitting a Bayesian model via Hamiltonian Monte Carlo (HMC) and then using the posterior to generate pseudo data for a simulation study. We refit the model using HMC and MAP on the pseudo data and then compare each model’s prediction accuracy on latent drug concentrations. We show that the normal approximation made by MAP results in dramatically different estimates of uncertainty for latent concentration in some patients. This result has significant implications for decision support derived from pharmacokinetic models.
|
Authors who are presenting talks have a * after their name.