The relative importance of between subject correlation of clearance (CL) and volume of distribution (V) compared with correlation of estimation error when applied to pharmacokinetic (PK) simulation

Purpose To describe the bias and imprecision of t 1/2 when CL and V are simulated with and without correlations in the random effects (PPV) and standard errors of parameter estimates (SEE). Methods: PK data for a 1‐compartment IV bolus model parameterized with CL (10 L/h, PPV=31.6%) and V (50 L, PPV = 31.6%), with and without correlation (R) between CL and V, and residual error (CV%= 20) were simulated in NONMEM. Parameter estimates obtained in NONMEM using FOCE were used to simulate t 1/2 for 1,000 subjects under 8 scenarios using the Pharsight Trial Simulator ( Table 1). Mean prediction error (%MPE), root mean square error (%RMSE), and standard deviation of prediction error (%SDPE) relative to the true values for t 1/2 were calculated. Results Table 2 shows statistics for bias and imprecision for t 1/2 by scenario. Conclusions Failure to include the covariance between CL and V is an important source of unrealistic results when using parameter estimates for simulation. Incorporation of information about parameter estimation error reduced bias and imprecision, especially when the correlation in PPV is incorrectly not included. Clinical Pharmacology & Therapeutics (2004) 75 , P89–P89; doi: 10.1016/j.clpt.2003.11.341Table 1. Scenario 1 2 3 4 5 6 7 8PPV Simulation R=0 R=.8 R=.8 R=0 R=0 R=.8 R=.8 R=.8Estimation R=0 R=.8 R=0 R=0 R=0 R=.6 R=.6 R=0SEE Simulation None None None R=0 R=‐.2 R=0 R=.1 R=.1Estimation None None None R=.1 R=.1 R=.8 R=.8 R=.8Table 2. Scenario 1 2 3 4 5 6 7 8t 1/2 (h) %MPE ‐36.3 ‐17.7 ‐116.0 ‐27.1 ‐22.9 ‐0.9 ‐5.0 ‐17.9%RMSE 126 51.8 241 99.8 112 28.6 28.6 63.8%SDPE 121 48.7 211 96.0 110 28.6 28.1 61.2