A Markov Chain Monte-Carlo Approach to Dose-Response Optimization Using Probabilistic Programming (RStan)

TL;DR

This paper presents a Bayesian hierarchical logistic regression model for dose-response analysis, implemented with RStan, comparing different priors and environments to optimize drug dosage predictions.

Contribution

It introduces a probabilistic programming approach using RStan for dose-response modeling and compares its performance with other platforms like PyMC and AgenaRisk.

Findings

RStan effectively samples from posterior distributions for dose-response models.

Different prior distributions significantly influence model outcomes.

RStan's results are comparable or superior to other probabilistic programming environments.

Abstract

A hierarchical logistic regression Bayesian model is proposed and implemented in R to model the probability of patient improvement corresponding to any given dosage of a certain drug. RStan is used to obtain samples from the posterior distributions via Markov Chain Monte-Carlo (MCMC) sampling. The effects of selecting different families of prior distributions are examined and finally, the posterior distributions are compared across RStan, and two other environments, namely PyMC, and AgenaRisk.