The 95% CDIRAs , a Credible Interval Based Method to Capture Uncertainty in Population Modeling: Pharmacokinetics Versus Sum of Exponentials Case Study

Determining the credibility of population PK/PD models is a challenge, particularly when performing uncertainty quantification (UQ) for parameter estimates, which is often represented through relative standard error (RSE) values. It is important to note that RSE values are primarily based on the variance of the parameter distribution and is therefore less sensitive to the overall shape and quantiles of a distribution which also reflects the uncertainty of a parameter. This can lead to unreliable uncertainty results and misleading interpretations. Robust UQ is important for model parameter estimation, as the results from in silico modeling are often used to inform subsequent steps in drug development. To capture the overall parameter uncertainty obtained from parametric approaches, the 95% credible interval ratios (95% CDIRAs) are introduced. The 95% CDIRAs only require the distribution for the individual parameter and consider the shape and quantiles of the distribution in addition to the variance. To showcase the 95% CDIRAs, an exemplar case study comprising simulated plasma concentration data was analyzed to assess whether a one compartment absorption population PK model or a reparameterized sum of exponentials (SOE) model can provide a more credible fit to plasma concentration data. Both metrics identified the population PK model as a more credible fit compared to the reparameterized SOE model. However, on occasions the precision classifications of the RSE values were higher than the adopted 95% CDIRAs precision classifications, which further indicates that uncertainty assessment based on variance alone does not necessarily encompass the full level of uncertainty for a parameter estimate.