IntroBayesNONMEM

Introduction to Bayesian pharmacometric data analysis using NONMEM

This repo contains the developing content for an introductory workshop on Bayesian pharmacometric data analysis using NONMEM.

Objectives

• Introduce principles and methods of Bayesian data analysis for pharmacometric applications.
• Provide a hands-on experience in Bayesian data analysis using NONMEM.

Primary intended audience: Pharmacometricians

Background assumed

• Population PKPD modeling
• Use of R and NONMEM

Workshop outline

• Why Bayesian?
• Introduction to Bayesian statistical principles and methods
  • Bayes Rule
  • Bayesian modeling & inference process
• Computation for Bayesian modeling
  • Maximum a Posteriori (MAP) Bayes
    • Individual: NONMEM POSTHOC
    • Population: Penalized Maximum Likelihood
  • Full Bayesian analysis
    • General computational approach: posterior simulation
    • Brief intro to Markov chain Monte Carlo (MCMC) simulation
      • Gibbs sampling
      • Metropolis-Hastings
      • Hamiltonian Monte Carlo and NUTS
• Overview of NONMEM implementations
  • MAP estimation
    • Using prior distributions with optimization methods
  • MCMC: BAYES and NUTS methods
  • Prior specification in NONMEM
• Hands-on 1: Example illustrating Bayesian data analysis workflow
• Prior distributions
  • Role of a prior distribution
  • Informative, uninformative or weakly informative?
• Hands-on 2: MAP popPK with selective use of informative priors for nuisance parameters: pediatric atorvastatin
• Model evaluation and comparison
• Assessing convergence and choosing numbers of burn-in and post-burn-in samples
• Getting your hands on posterior samples for individual parameters and predictions
• Hands-on 3: Full Bayes popPK with selective use of informative priors for nuisance parameters: pediatric atorvastatin
• When stuff goes wrong
  • Diagnosing and remedying sampling problems encountered with MCMC
  • Reparameterization, e.g., centered vs non-centered parameterizations for hierarchical models
  • Prior distributions as part of the solution
• Hands-on 4: Full Bayes popPKPD using semi-mechanistic model
  • Friberg-Karlsson semi-mechanistic model for drug-induced myelosuppression
  • Informative priors for drug-independent system parameters
• Practical strategies for selecting Bayesian estimation methods for specific types of problems
  • When to go Bayes (and why)?
  • Which method?
  • Which tool?
• Preview of Bayesian data analysis using Stan and Torsten
  • Brief intro with demo
  • Advantages/disadvantages
• What didn't we cover?