4 Oliceridine respiratory effects estimation using stochastic approximation expectation maximization, objective function evaluation using importance sampling, and a final No U-Turn sampling Bayesian step using noninformative priors to visualize and quantify parameter uncertainty. The early samples at 2 min after infusion showed considerable variability. Because infusion was done manually for 1 min, it was hypothesized that this could be caused, at least in part, by variability of the infusion duration. Therefore, NONMEM’s parameter of the infusion duration (D1) was set up to be an estimable parameter. Body weight and, for oliceridine, the metabolizer status based on the genotype of the CYP2D6 gene, were incorporated as covariates in the pharmacokinetic analyses. The change in NONMEM’s objective function value was tested to assess whether weight via allometric scaling improved the fit (because this requires no extra parameters, incorporating allometric scaling would be preferable with any decrease in the objective function value). For metabolizer status, the clearance for each nonnormal status was tested for statistically significant difference from the clearance for the normal status (change in objective function value of at least 6.63; p <0.01). Allometric scaling using standard powers of weight (1 for volumes and 0.75 for clearances) was assumed a priori and implemented in the pharmacokinetic models.22 During model evaluation, it was checked that incorporating allometric scaling indeed reduced NONMEM’s objective function value and that it decreased the dependence of interindividual variability terms on weight. To quantify the hysteresis between the arterial drug concentration and effect, an effect site is postulated characterized by a first-order process with rate constant ke0 and half-life t1/2ke0 (= ln2/ke0). The ventilatory effects of oliceridine and morphine were modeled using an inhibitory sigmoid EMAX model. Ventilation at an extrapolated isohypercapnic level of 55 mmHg ( ˙VE55) was modeled as follows: ˙V E55(t) = ˙VE55 at baseline – ˙VE55 at baseline × CE(t) γ C γ 50 1+ CE(t)γ C γ 50 where baseline is the value before any drug administration, CE(t) is the effectsite concentration at time t, C50 the effect-site concentration causing a 50% depression of ˙VE55, and γ a shape parameter, which was fixed to 1 in the analyses. The same estimation steps were followed as was done for the pharmacokinetic analyses. To determine whether the models adequately described the data, Goodness-of-Fit plots were created and inspected. To allow a visual predictive check of the final pharmacokinetic or pharmacodynamic models, the normalized prediction discrepancies were estimated. Parameter estimates are reported as median ±standard error of the estimate; p <0.01 was considered 71
RkJQdWJsaXNoZXIy MTk4NDMw