What Is The Relevance Of Estimating Inter-Study Variability In Population Pharmacokinetic Meta-Analysis ?

Simultaneous estimation of inter-study, inter-individual and residual distribution variability requires 3 nested levels of random effects. None of the traditional methods used for population PK implements such a (complex) model. However the flexibility of NONMEM software allows one to define one random effect h per individual, and then to constrain all h?s to have the same distribution: this constitutes the « true » IIV level, while ISV level is assimilated to the « usual » IIV level implemented in the software (Karlsson MO, Sheiner LB. J Pharmacokin Biopharm. 21:735-750, 1993).

We simulate data from 4, 10 and 20 studies with 12 patients/study assuming a one-compartment model with bolus input, log-normally distributed parameters, ISV affecting clearance (IIV-CL) and volume (IIV-V), or residual variability (RESV), and with different levels of ISV. For analysis we used a standard random effects model that ignores ISV and then the full model, with 3 nested levels of randomness (ISV, IIV and RESV).

Ignoring ISV in the analysis does not affect fixed effect parameter estimation but results in a biased estimation of IIV variances. When the analysis model allows ISV to act on IIV and RESV simultaneously, (i) bias on IIV disappears (ii) the correct model of ISV is selected and (iii) imprecise estimates of ISV are obtained.

If it is believed that this increase of variability is purely due to study effects rather than to true IIV, we suggest to decompose the IIV in true IIV and ISV using a 3 nested level random effect model. These results have to be confirmed further by exploring the effect of heterogeneity between studies like different numbers of subject per study, different sampling designs or different experimental designs.