Dosing regimen optimisation of vedolizumab during pregnancy through physiologically-motivated sequential population modelling of albumin trends and vedolizumab pharmacokinetics
Zrinka Duvnjak1,2, Robin Michelet1, Casper Steenholdt3,4, Ella S.K. Widigson1,2, João A. Abrantes5, Wilhelm Huisinga2,6, Mette Julsgaard7,8, Charlotte Kloft1,2
1Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universität Berlin, 2Graduate Research Training program PharMetrX, 3Department of Medical Gastrointestinal Diseases S, Odense University Hospital, 4Research Unit of Medical Gastroenterology, Department of Clinical Research, University of Southern Denmark, 5Roche Pharma Research and Early Development, Pharmaceutical Sciences, Roche Innovation Center Basel, 6Institute of Mathematics, Universität Potsdam, 7Department of Hepatology and Gastroenterology, Aarhus University Hospital, 8Institute of Clinical Medicine, Health, Aarhus University
Introduction Physiological changes in pregnancy may significantly impact monoclonal antibody (mAb) pharmacokinetics (PK) [1–3]. Nevertheless, a quantitative understanding of these changes such as plasma volume expansion and transplacental transfer is lacking. It has been hypothesised that a decline in albumin (ALB) during pregnancy can be predominantly explained by plasma volume expansion [2,4]. However, pregnancy impact in mAb population PK (popPK) models was so far described only as a categorical covariate effect on clearance (CL) [5,6]. Moreover, the PK of most mAbs, e.g. vedolizumab (VDZ) - approved for inflammatory bowel diseases (IBD) - remains uncharacterised during gestation [7]. Developing a popPK model incorporating pregnancy effects is challenging due to the common lack of pre-pregnancy drug measurements. Additionally, proposing dosing adjustments is complicated by pregnancy starting at different times within an 8-week dosing interval for different patients. This results in dosing scheduled for different gestational age (GA), each requiring a different optimised regimen. Nevertheless, keeping the disease in remission is crucial for reducing the risk of adverse pregnancy outcomes [8]. Objectives 1. To develop the 1st popPK model of VDZ during pregnancy by: -implementing a sequential covariate-trend/PK modelling approach -linking individual %decrease in ALB concentrations (from the covariate-trend model) to individual %increase in VDZ central volume of distribution (Vc), as it carries information about plasma volume expansion -quantifying remaining changes in VDZ pregnancy PK 2. To optimise VDZ dosing regimens during pregnancy: -for 3 common pre-pregnancy i.v. maintenance dosing regimens -independent of pregnancy start relative to the last dose Methods Data from 39 pregnant IBD patients receiving VDZ (300 mg, i.v., every 8 weeks (Q8W), Q6W or Q4W) in a prospective, observational, multicenter study were analysed, including in total 136 VDZ concentrations at trough levels (pregnancy-only) and at delivery. Patient demographics, disease characteristics, and laboratory data were documented pre- and monitored throughout pregnancy. A sequential covariate-trend/PK modelling approach was applied in NONMEM v7.5.1. First, ALB concentrations were modelled using a mixed-effects second-order polynomial function of fertilisation age (FA, time from conception, true start of pregnancy), with interindividual variability (IIV) in linear coefficient and y-intercept (pre-pregnancy ALB). Bayesian maximum a posteriori estimates of ALB-model parameters were used to compute individual ALB %change from pre-pregnancy ALB for any GA (time from the last menstrual period, commonly used in practice). The pregnancy VDZ PK model was built as an extension of a previously published 2-compartment disposition model of VDZ, with parallel linear and nonlinear elimination and 17 covariate effects [9]. All covariate effects were retained from the published model as baseline effects, while structural and stochastic components were re-estimated or fixed if not supported by the data. Individual ALB %change (derived from the ALB-model) was used as a time-varying covariate effect on Vc and CL. In addition, the remaining GA effects on CL and Vc were tested using an exponential model with varying GA cut-off values. The optimised dosing times (expressed as the patient’s GA at these times) were derived using stepwise simulations from the developed pregnancy PK model, administering the next dose when VDZ concentrations dropped below pre-pregnancy values. This was done separately for 8 scenarios, differing in the time of pregnancy onset relative to the last dose. To allow for derivation of optimised dosing times for any time of pregnancy start without additional simulations, optimised dosing times found for 8 scenarios were modelled together as a polynomial function of the corresponding scheduled dosing times (assuming pre-pregnancy dosing intervals were maintained throughout pregnancy). Results Both the ALB- and the pregnancy VDZ PK model described the data well. The mixed-effects polynomial ALB-model captured individual trends well (total decrease: 13.5%–41.6%, 5th–95th simulated percentile) and facilitated model-based imputation of missing values. Incorporating individual ALB %change as a time-varying covariate enabled estimation of pre-pregnancy CL (0.157 L/day) and Vc (4.43 L) despite the absence of pre-pregnancy VDZ data. Vc was estimated to increase typically by 43.6% (5th-95th simulated percentile: 21.4%-67.5%) by the end of pregnancy, aligning with reported values for plasma volume expansion [1,2,10]. CL was estimated to also increase due to ALB %change, but with a lower magnitude than Vc (76.5% of Vc increase) - a trend also observed when applying allometric scaling [11]. After accounting for ALB %change, the estimated remaining effect of GA on CL (an additional 25% decrease in trough concentrations) was found to start at the beginning of the 3rd trimester, coinciding with the endogenous IgG decline due to transplacental transfer. Given that VDZ is IgG-based and also crosses the placenta, this remaining GA effect may reflect the transplacental transfer [12]. Overall, VDZ trough concentrations declined by 43.5% by the end of pregnancy, indicating the need for dosing adjustments. To maintain pre-pregnancy VDZ trough levels during gestation, the dosing intervals gradually shortened throughout pregnancy, reaching intervals of 5.9, 4.3, and 2.8 weeks for patients on pre-pregnancy Q8W, Q6W, and Q4W regimens, respectively. For any scheduled dosing time (if the pre-pregnancy dosing interval would be maintained), the optimised dosing time can be calculated using the following polynomials for Q8W, Q6W and Q4W, respectively: GA_optimised = -0.00297 × GA_scheduled,Q8W² + 0.998 × GA_scheduled,Q8W - 0.157 GA_optimised = -0.00300 × GA_scheduled,Q6W² + 0.988 × GA_scheduled,Q6W - 0.106 GA_optimised = -0.00310 × GA_scheduled,Q4W² + 0.980 × GA_scheduled,Q4W - 0.0696 In this way, clinicians can derive optimised dosing times requiring only a basic calculator, and without using dedicated modelling software. Conclusions This study presents the first popPK model of VDZ in pregnancy. By linking VDZ concentration declines to plasma volume expansion via ALB decrease and quantifying the remaining GA effect, the model advances the quantitative understanding of mAb PK in pregnancy. The developed framework effectively addressed common data limitations such as missing pre-pregnancy drug concentrations and time-varying covariates, while accounting for more realistic, individualised magnitudes of pregnancy effects. Finally, optimised dosing times for commonly used VDZ maintenance dosing regimens were derived. By translating the findings into easy-to-use equations applicable to any pregnancy onset time, we provide an accessible, model-informed strategy to ensure adequate therapeutic exposure during pregnancy and ultimately decrease the risk of flares.