Methods to Detect Non-Compliance and Minimize its Impact on Population PK Parameter Estimates
Leonid Gibiansky (1), Ekaterina Gibiansky (1), Valerie Cosson (2), Nicolas Frey (2), Franziska Schaedeli Stark (2)
(1) QuantPharm LLC, North Potomac, MD, US; (2) F. Hoffmann-La Roche Ltd, Basel, Switzerland
Objectives: To develop and evaluate methods to detect non-compliance and obtain unbiased parameter estimates in a population pharmacokinetic (PK) analysis.
Methods: Data sets emulating clinical studies with different duration, sampling schemes and levels of compliance to a once daily oral dosing regimen were simulated using a 2-compartment model with first-order absorption and elimination and significant drug accumulation. Non-compliance was simulated as drug holidays preceding some observations in 20 to 80% of subjects. For each dataset, the original model was fit assuming full compliance to evaluate precision and bias on the parameter estimates. Two methods (CM1, CM2) to account for non-compliance were tested. CM1 introduced a random effect (ETAerr) on the magnitude of the residual error and re-estimated PK parameters with increasing fractions of subjects with high ETAerr removed from the data set. CM2 is the generalization of the idea proposed in . It relied on rich data obtained immediately before and after an observed dose in the clinic, while trough PK samples related to unobserved doses outside the clinic (outpatient doses) were ignored. To account for possible non-compliance, individual relative bioavailability of the outpatient doses was introduced, estimated, and associated to individual compliance.
Results: When assuming full compliance, the PK parameter estimates were significantly biased. By introducing ETAerr in CM1 the bias was reduced and non-compliant subjects could be associated with a high ETAerr. Incremental removal of subjects with high ETAerr further reduced the bias until the parameter estimates converged to the true values, while the variance of the ETAerr decreased towards zero. However, precision of the obtained parameter estimates decreased with increasing number of subjects removed to obtain unbiased parameter estimates. CM2 yielded unbiased PK parameter estimates for the datasets with any fraction of non-compliant subjects. Non-compliant subjects could be associated with a low bioavailability estimate for the outpatient doses. However, the method heavily relies on the availability of rich data following an observed dose in the clinic.
Conclusions: The proposed methods offer ways to identify subjects with non-compliance and reduce or eliminate bias on PK parameter estimates based on rich or sparse PK sampling data in populations with prevalent non-compliance.
 Gupta P, Hutmacher MM, Frame B, Miller R, An alternative method for population pharmacokinetic data analysis under noncompliance. J Pharmacokinet Pharmacodyn. 2008;35(2):219-33.