Pharmacodynamic modelling approaches in the setting of suboptimal drug adherence

Introduction. Suboptimal adherence may bias pharmacokinetic (PK) parameter estimates.1 Rather than conditioning the PD response on the model predicted PK response, an alternative approach is to fit a kinetic-pharmacodynamic (KPD) model in which the PK data are excluded from the analysis. KPD models may be an advantage in such scenarios because the KPD analysis allows for imputation of the likely time course of drug exposure. To our knowledge, this concept has not been explored previously.

Aim. To compare the performance of PKPD model versus a KPD approach to describe the time course of drug effect in the setting of known or unknown suboptimal drug adherence.

Methods. A stochastic simulation estimation (SSE) study was conducted. The assumed scenario included a hypothetical drug administered by intravenous bolus at a dose of 1mg daily for 7 days. The drug followed a one-compartment PK model with first order elimination. An immediate effect Emax model was used to link the plasma concentrations of the drug to the pharmacodynamic effect. The simulation I/O model was parametrised in terms of clearance (CL), volume of distribution (V), Emax, and C50 (drug concentration where drug effect is equal of 50% of Emax). Intensive sampling was conducted at 0.563, 1.125, 2.25, 4.5, 9, 18, 36, 72 h post the final dose. Two scenarios were considered for each simulated virtual patient: one with perfect adherence and the other with imperfect adherence. The imperfect adherence was generated such that every virtual patient randomly missed 3 doses except the first and last dose. A total of 100 virtual patients were simulated under each scenario and each scenario was replicated 3 times. The simulated data was fit to (1) a simultaneous PKPD model, (2) a sequential IPP model,2 and (3) KPD model using the A50 parametrisation.3 Visual predictive checks (VPCs) were used to evaluate the performance of models. Simulation was implemented in MATLAB® (vR2021a) and estimation was performed in NONMEM® version 7.3, using the first-order conditional estimation method with interaction. Visual Predictive Checks (VPCs) were created using the Xpose4 package in R (v 4.0.2).

Results. For all three models,  the imperfect adherence scenario resulted in a lower magnitude of drug response compared to the perfect adherence scenario. When comparing the VPCs for the three models, no substantial differences in the model fit was obvious. This was consistent across the three replicates.

Conclusion. The performance of simultaneous and sequential PKPD models and the KPD model in describing the time course of drug effect was similar during imperfect adherence, suggesting that including the PK data, even when the actual dosing history was unknown, does not correct for imperfect adherence.  This should not be construed to mean that the models would predict future PD data appropriately.


  1. Girard P, Sheiner LB, Kastrissios H, Blaschke TF. Do We Need Full Compliance Data for Population Pharmacokinetic Analysis? J Pharmacokinet Biopharm. 1996;24(3).
  2. Zhang L, Beal SL, Sheiner LB. Simultaneous vs. Sequential Analysis for Population PK/PD Data I: Best-case Performance. J Pharmacokinet Pharmacodyn. 2003;30(6):387-404.
  3. Ooi QX, Hasegawa C, Duffull SB, Wright DFB. Kinetic-pharmacodynamic model for drugs with non-linear elimination: Parameterisation matters. Br J Clin Pharmacol. 2020;86(2):196-198.