Mutaz M. Jaber (1), Richard C. Brundage (1)
(1) Department of Experimental and Clinical Pharmacology, University of Minnesota
Objectives: In nonlinear mixed-effects (NLME) modeling of pharmacokinetic (PK) and pharmacodynamic (PD) data, we refer to two levels of random effects: between-subject variability (BSV), and residual unexplained variability (RUV). RUV quantifies the residual uncertainty in the model after accounting for the variability due to the structural, covariate, and BSV models. Various sources are expected to contribute to RUV; imprecision of our analytical techniques, misspecification of the PK, PD, covariate, BSV, or RUV models, errors in recording times of blood sampling or dose administration, variability in formulations, and an uncertain prior dosing history. The goal of this study was to investigate the extent to which PK and RUV model misspecification, errors in recording dosing and sampling times, and variability in drug content uniformity contribute to the estimated magnitude of RUV.
Methods: BASE MODEL. A two-compartment with first-order absorption and linear elimination was simulated. PK parameters included clearance 5.0 L/hr, central volume of distribution 35 L, inter-compartmental clearance 50 L/hr, peripheral volume of distribution 50 L, and absorption rate constant of 0.7 1/hr. All parameters were assumed to have a 30% BSV coefficient of variation (CV). The simulated dose was 120 mg. Intense (INT) and sparse (SPA) sampling designs were investigated in simulations of 100 subjects. For the INT, sampling times were 0.5,1, 2, 4, 6, 8, 12, 24, and 48 hours (9 samples); and at 2, 12, 24, and 48 hours (4 samples) for the SPA. A 5% proportional error model was assumed as an optimistic baseline RUV. This base model was simulated and fit 100 times.
PERTURBATIONS. The base model was perturbed in several ways to investigate changes in RUV: 1) PK model misspecification where the data from above were fit to a one-compartment model. New datasets were simulated as above to explore additional perturbations: 2) RUV model misspecification using a 5% proportional error with 0.01, 0.02, and 0.04 mg/L additive error; 3) dosing time misspecification was achieved by perturbing the time of dose with either a normal distribution (mean of 0 and standard deviation of 5 min), or uniform distribution (between -5 and 5 min); 4) sampling time misspecification was achieved by perturbing the time of sample using the same distributions as dose time misspecification; 5) dose variability misspecification was accomplished by corrupting the dose with a normal distribution (mean of 120 mg and standard deviation of 12 mg); and 6) combined PK model, RUV misspecifications with an additive component of 0.04, dose and sampling time misspecification using the specified normal distribution, and dose misspecification.
In all cases, the perturbed simulated concentrations were fit assuming the baseline nominal dose, times, and RUV. Fractional deviations from nominal RUV were reported as the median RUV [min, max] from the 100 runs for each scenario.
Results: INTENSE MISSPECIFICATIONS. 1) PK model: 1.38 [1.19, 1.58]; 2) RUV model: 1.33 [1.12 – 1.87], 1.78 [1.33 – 2.3], and 2.25 [1.92 – 2.92] for the levels of 0.01, 0.02, and 0.04 mg/L additive error, respectively; 3) dosing time: 1.06 [0.93 – 1.31] and 1.03 [0.76 – 1.22] for the normal and uniform distributions, respectively; 4) sampling time: 1.10 [1.00 – 1.30] and 1.04 [ 0.93 – 1.27] for the normal and uniform distributions, respectively; 5) dose variability: 1.01 [0.94 – 1.33]; and 6) combination: 5.32 [2.15 – 16.53].
SPARSE MISSPECIFICATIONS. 1) PK model: 1.92 [1.00 – 6.25]; 2) RUV model: 3.35 [1.48 – 11.69], 5.84 [2.65 – 14.21], and 13.8 [4.88 – 25.92] for the levels of 0.01, 0.02, and 0.04 mg/L additive error, respectively; 3) dosing time: 0.98 [0.72 – 1.34] and 0.98 [0.70 – 1.39] for the normal and uniform distributions, respectively; 4) sampling time: 1.00 [0.69 – 1.34] and 0.99 [0.72 – 1.42] for the normal and uniform distributions, respectively; 5) dose variability: 1.00 [0.75 – 1.29]; and 6) combination: 19.87 [12.54 – 33.86].
Conclusions: PK and RUV model misspecification were associated with relatively large increases in the magnitude of RUV compared to other sources for INT and SPA designs. As expected, SPA sampling has higher deviations than INT designs. The contribution of dose misspecification, and dosing and sampling time misspecifications have negligible effects on RUV.
Reference: PAGE 29 (2021) Abstr 9688 [www.page-meeting.org/?abstract=9688]
Poster: Methodology - New Modelling Approaches