Concordance between criteria for covariate model building

Area of Science:

• Pharmacometrics
• Pharmacokinetic Modeling
• Statistical Analysis

Background:

• Population pharmacokinetic (PopPK) models frequently incorporate covariates to explain parameter variability.
• Model improvement is typically assessed by enhanced goodness-of-fit and reduced unexplained (random) parameter variability.
• Explained parameter variability, representing predictable variation, is less commonly used as a primary model improvement criterion.

Purpose of the Study:

• To investigate the utility of three criteria—goodness-of-fit, decreased unexplained variability, and increased explained variability—for evaluating covariate inclusion in PopPK models.
• To assess the agreement between these criteria across varying strengths and natures of covariate-parameter relationships.
• To examine these criteria in both simulated datasets and real-world PopPK analyses.

Main Methods:

• Stochastic simulations and estimations were performed to generate data for evaluating covariate-parameter relationships.
• Four previously published real-world PopPK datasets were re-analyzed to assess covariate effects.
• The study compared changes in objective function value, unexplained parameter variability, and explained parameter variability following covariate addition.

Main Results:

• Total estimated parameter variability was influenced by the number of covariates included in the model.
• In simulated and some real data examples, parameter variability increased with more covariates; in others, it decreased or showed no systematic change.
• The three evaluation criteria (goodness-of-fit, unexplained variability, explained variability) were highly correlated, with unexplained variability changes more closely linked to objective function values.

Conclusions:

• The assumption that covariate inclusion solely shifts unexplained variability to explained variability may not hold true.
• The relationship between explained parameter variability and model improvement warrants careful consideration.
• These findings have potential implications for decision-making processes in PopPK model development and covariate selection.