Which diagnostic is commonly used to assess population PK model fit?

• A. Posterior predictive checks
• B. Visual Predictive Checks (VPC)
• C. Shrinkage
• D. Goodness-of-fit plots

Answer: (B)

Visual Predictive Checks focus on how well the model reproduces what you actually see in the data across time and across individuals. In a VPC, you simulate many datasets from the fitted population PK model and then compare the observed concentrations to prediction intervals (often the 5th and 95th percentiles) within time bins. If the observed data mostly fall inside these intervals and follow the same time-course pattern as the simulations, the model is capturing both the typical trajectory and the variability in the population.

This approach is particularly informative for population PK because it directly shows predictive performance across dosing regimens, sampling times, and subject-to-subject variability. It helps you spot misspecifications such as incorrect absorption kinetics, dosing effects, or error structure, in a way that’s easy to visualize and interpret.

Other diagnostics exist, like posterior predictive checks (more common in Bayesian contexts), shrinkage (which describes how much individual estimates borrow from the population rather than a fit issue), or goodness-of-fit plots (which are useful but can be less intuitive for judging population-level predictive accuracy). Visual Predictive Checks provide a clear, intuitive assessment of model fit at the population level.