Comparisons Between Hamiltonian Monte Carlo and Maximum A Posteriori For A Bayesian Model For Apixaban Induction Dose & Dose Personalization

A. Demetri Pananos, Daniel J. Lizotte
Proceedings of the 5th Machine Learning for Healthcare Conference, PMLR 126:397-417, 2020.

Abstract

Precision medicine’s slogan is "right drug - right patient - right time." Implicit in the slogan is "right dose"; however, determining the right dose for any one patient can be challenging when dose-response data are limited. Bayesian methods, with their ability to explicitly incorporate prior information to supplement limited data, have been proposed as a solution to this problem. Although Hamiltonian Monte Carlo (HMC) is a leading methodology for inference in Bayesian models because of its ability to capture posterior distributions with high fidelity, dose personalization studies commonly use simpler Maximum A Posteriori (MAP) inference methods. The impact of the choice of inference engine on dose decision-making has not been explored. To better understand this issue, we perform a simulation study characterizing the differences between inferences made via MAP and HMC for personalized dosing strategies. The simulation study uses a new Bayesian pharmacokinetic model for apixaban pharmacokinetics written in an open source Bayesian language; the model code and posterior summaries of all parameters will be publicly available. We demonstrate that the differences between HMC and MAP are non-trivial and can greatly affect the choices surrounding dose selection for personalized medicine.

Cite this Paper

BibTeX
@InProceedings{pmlr-v126-pananos20a, title = {Comparisons Between Hamiltonian Monte Carlo and Maximum A Posteriori For A Bayesian Model For Apixaban Induction Dose & Dose Personalization}, author = {Pananos, A. Demetri and Lizotte, Daniel J.}, booktitle = {Proceedings of the 5th Machine Learning for Healthcare Conference}, pages = {397--417}, year = {2020}, editor = {Doshi-Velez, Finale and Fackler, Jim and Jung, Ken and Kale, David and Ranganath, Rajesh and Wallace, Byron and Wiens, Jenna}, volume = {126}, series = {Proceedings of Machine Learning Research}, month = {07--08 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v126/pananos20a/pananos20a.pdf}, url = {https://proceedings.mlr.press/v126/pananos20a.html}, abstract = {Precision medicine’s slogan is "right drug - right patient - right time." Implicit in the slogan is "right dose"; however, determining the right dose for any one patient can be challenging when dose-response data are limited. Bayesian methods, with their ability to explicitly incorporate prior information to supplement limited data, have been proposed as a solution to this problem. Although Hamiltonian Monte Carlo (HMC) is a leading methodology for inference in Bayesian models because of its ability to capture posterior distributions with high fidelity, dose personalization studies commonly use simpler Maximum A Posteriori (MAP) inference methods. The impact of the choice of inference engine on dose decision-making has not been explored. To better understand this issue, we perform a simulation study characterizing the differences between inferences made via MAP and HMC for personalized dosing strategies. The simulation study uses a new Bayesian pharmacokinetic model for apixaban pharmacokinetics written in an open source Bayesian language; the model code and posterior summaries of all parameters will be publicly available. We demonstrate that the differences between HMC and MAP are non-trivial and can greatly affect the choices surrounding dose selection for personalized medicine.} }
Endnote
%0 Conference Paper %T Comparisons Between Hamiltonian Monte Carlo and Maximum A Posteriori For A Bayesian Model For Apixaban Induction Dose & Dose Personalization %A A. Demetri Pananos %A Daniel J. Lizotte %B Proceedings of the 5th Machine Learning for Healthcare Conference %C Proceedings of Machine Learning Research %D 2020 %E Finale Doshi-Velez %E Jim Fackler %E Ken Jung %E David Kale %E Rajesh Ranganath %E Byron Wallace %E Jenna Wiens %F pmlr-v126-pananos20a %I PMLR %P 397--417 %U https://proceedings.mlr.press/v126/pananos20a.html %V 126 %X Precision medicine’s slogan is "right drug - right patient - right time." Implicit in the slogan is "right dose"; however, determining the right dose for any one patient can be challenging when dose-response data are limited. Bayesian methods, with their ability to explicitly incorporate prior information to supplement limited data, have been proposed as a solution to this problem. Although Hamiltonian Monte Carlo (HMC) is a leading methodology for inference in Bayesian models because of its ability to capture posterior distributions with high fidelity, dose personalization studies commonly use simpler Maximum A Posteriori (MAP) inference methods. The impact of the choice of inference engine on dose decision-making has not been explored. To better understand this issue, we perform a simulation study characterizing the differences between inferences made via MAP and HMC for personalized dosing strategies. The simulation study uses a new Bayesian pharmacokinetic model for apixaban pharmacokinetics written in an open source Bayesian language; the model code and posterior summaries of all parameters will be publicly available. We demonstrate that the differences between HMC and MAP are non-trivial and can greatly affect the choices surrounding dose selection for personalized medicine.
APA
Pananos, A.D. & Lizotte, D.J.. (2020). Comparisons Between Hamiltonian Monte Carlo and Maximum A Posteriori For A Bayesian Model For Apixaban Induction Dose & Dose Personalization. Proceedings of the 5th Machine Learning for Healthcare Conference, in Proceedings of Machine Learning Research 126:397-417 Available from https://proceedings.mlr.press/v126/pananos20a.html.

Related Material