Date of Award

2019

Degree Type

Thesis

Degree Name

Master of Science in Statistics

Department

Computer Science and Statistics

First Advisor

Gavino Puggioni

Abstract

Published articles constitute a huge repository of population pharmacokinetic (Pop PK) models that can be repurposed to answer dose-related questions. We present a Simulation-Informed-Statistically-Aided (SISA) method that can be used to select population pharmacokinetic models from literature for repurposing.

Seven published melphalan PK models, and 1 reference model from FDA NDA submission documents were selected. A 30 minute IV infusion of 140 mg/m2 of melphalan was simulated using each of the 7 models. The predictive performance of each of the models were evaluated using a proposed method called modified standard normal deviate (mSND). Five models, M1, M3, M4, M5 and M6 performed well predicting the given dose with mSND within 0 -3. A composite dataset was generated from these 5 models and used to develop a composite melphalan PK model using First-order conditional estimation method with interaction (FOCE-I), and the Bayesian Hamiltonian Monte Carlo No-U-Turn Sampler (HMC NUTS) estimation methods. Both informative and weakly informative priors were tested during the Bayesian estimation.

For both Bayesian and Non-Bayesian methods, a two compartmental model adequately fit the composite data. The parameter estimates and the model predictions were in agreement with those from reference model. This confirmed that validity of SISA. Again, the findings suggested that literature data can be repurposed to develop composite PK models.

It is hoped that further refinement of this method will provide a more objective approach for selecting PK models for repurposing.

Available for download on Monday, December 13, 2021

Share

COinS