Background: This analysis evaluates approaches for handling erroneous dosing data in a population pharmacokinetics (POPPK) dataset due to missed subject dosing records from medication non-adherence. Methods: Data simulated from pre-existing POPPK models were modified with the following imputation methods in the attached table. These datasets were subsequently re-estimated with various degrees of missing dosing records and half lives under three case scenarios. In case 1, only the counts of drug tablets distributed and returned during a dosing period are available. In case 2, one of two doses following a twice daily dosing regimen is missed, but it is unclear whether it is the morning or evening dose. In case 3, the last dose of a twice daily dosing regimen before PK sample collection may be missing or unknown. Estimation precision was assessed using percent estimation error for key parameters, and numerical assessments of Between Subject Variability (BSV) and covariate effects. Results: Overall, the Omit, Prescribed Dose, and Mix methods showed low error estimations for key POPPK parameters, including clearance, volume of distribution, and steady state concentrations. BSV and covariate effects were also well-captured. In case 1, the end period and compartment initialization methods yielded a high number of outliers. For case 2, all methods, with exception of omitting the evening dose (subcategory of the omit method), yielded lower error estimations. In case 3, the Prescribed Dose Method achieved the lowest estimation error in model parameters. Conclusion: With exception of case 3, the Omit, Mix, and Prescribed Dose methods adequately recovered parameter estimates, covariate effects, and BSV distributions across scenarios and compound properties. In case 3, the prescribed dose method is recommended as it achieved the lowest estimation errors.
