Introduction/Objectives: Controlled ovarian hyperstimulation (COH) in women undergoing in vitro fertilisation (IVF) involves daily administration of recombinant follicle-stimulating hormone (FSH; follitropin alfa) to stimulate follicular maturation. As follicles grow, they initially secrete inhibin-B [1], making inhibin‑B a sensitive biomarker of early ovarian response. Substantial inter‑individual variability exists in FSH pharmacokinetics, meaning optimising the number of retrieved oocytes is difficult [2]. Population pharmacokinetic/pharmacodynamic (popPK/PD) modelling could enable us to predict response to FSH, supporting individualised dosing to optimise oocyte retrieval. This study aimed to (i) update and extend a previously published popPK/PD model of FSH and inhibin‑B, and (ii) assess model performance using real‑world Australian IVF data to inform individualised FSH dosing practices.
Methods: Data were obtained from 81 Australian women undergoing IVF. Repeated FSH and inhibin‑B measurements were collected before the first dose (baseline) and approximately 12 hours after last dose (Ctmid) intermittently. The a priori model [3] comprised of a one‑compartment PK model with first‑order absorption and linear elimination, an ODE turnover model for endogenous FSH, linked to a PD turnover model for inhibin‑B production stimulated by FSH. Endogenous FSH was converted from μg/L to IU/L to reflect the Australian data. The parameter estimates of the a priori model were re-estimated using the dataset using non-linear mixed-effects modelling through the stochastic approximation expectation-maximization algorithm (Monolix, 2024R1), followed by covariate evaluation. Covariates evaluated included baseline age, total body weight, anti-Müllerian hormone (AMH), COH cycle number, oestradiol, and luteinizing hormone (LH). Covariate evaluation used conditional sampling use for stepwise approach based on correlation tests [4] for automated correlation‑driven selection, assessing for physiological plausibility and improvements in model fit. Parameter precision was assessed via relative standard errors. Model performance was assessed using standard diagnostic plots and visual predictive checks (VPC; 90% prediction interval).
Results: Of 81 patients, all contributed ≥2 FSH concentrations and 24 contributed ≥2 inhibin-B concentrations. The typical patient was 37 (range 25-45) years old, weighing 65 (44-100) kg, was 1.65 (1.48-1.75) m tall, and taking 250 (125-450) IU/d of follitropin alfa. The median number of oocytes retrieved was 10 (0-28). The PK and endogenous FSH a priori models demonstrated consistent structural performance, while IC50 was omitted from the PD a priori model as it introduced model instability and confounded EC50. The re-estimated population parameters were similar to the a priori model. The final covariate relationships were as follows: V/F=6.86 ast(Weight/80)^2.41 L/h; Baseline [FSH]=0.44 * e^((-0.014 * baseline [AMH]))* e^((0.045 * baseline [LH])) IU/L. Including these covariates improved model fit (BICc delta -30.88; −2LL delta -45.35). Only 1.94% and 2.27% of the observation-prediction FSH and inhibin-B data, respectively, were outside of the 90% prediction interval. VPCs indicated adequate description of data despite sparse clinical sampling schedules typical of routine IVF practice.
Conclusions: The integration of covariates improved the predictive performance of the popPK/PD model of FSH and inhibin-B. FSH is a large and hydrophilic glycoprotein [5], meaning weight only slightly informs volume of distribution. AMH and LH indirectly influence FSH through endocrine feedback loops [6]. Next, the inclusion of ovarian response biomarkers as time-varying covariates using in-cycle data to further inform FSH dose adjustments will be explored. Overall, the updated model supports future development of model‑informed precision dosing tools for FSH, particularly for early identification and management of poor or hyper responders. Despite sparse data, the model robustly characterised patient trajectories, demonstrating feasibility for real‑world clinical application. Planned in silico simulations will evaluate personalised dosing strategies and assess potential improvements in COH outcomes.
References: [1] Eldar-Geva, J Clin Endocrinol Metab, 2000; [2] Bahadur, BMJ Open, 2023; [3] Ebid, J Clin Pharmacol, 2021; [4] Ayral, CPT Pharmacometrics Syst Pharmacol, 2021; [5] Bousfield, Endocrinol, 2019; [6] Raju, J Hum Reprod Sci, 2013.