Abstract Study question Can key elements of ovarian physiology be accurately modelled, thereby providing a tool to generate novel insights and guide development of new therapeutics? Summary answer A physiological computational model was developed to describe follicle growth and follicle size distributions observed in agonist and antagonist ovarian stimulation (OS) protocols. What is known already The outcome of OS remains unpredictable despite identification of several prognostic factors. The average first cycle pregnancy rate is 30-40% across fertility centres, highlighting the need for improved stimulation protocols. This is an active area of clinical research; however, clinical trials are costly, time-consuming, and frequently have outcomes that are unexpected or difficult to interpret. We integrated available physiological knowledge of human follicle growth and development and clinical trial data into a computational model to enable in silico simulations with the aim to optimise clinical trial design. Study design, size, duration Computational model building was performed using a mechanistic approach aimed at capturing well-understood physiology that is relevant in OS protocols. Model inputs included patient characteristics (such as anti-Müllerian hormone (AMH) level), OS regimen and pituitary downregulation with gonadotrophin releasing hormone (GnRH) agonist or antagonist. Key outputs of the model included follicle number and mean size, duration of stimulation and serum hormone concentrations (oestradiol, testosterone, inhibin B and androstenedione). Participants/materials, setting, methods Published physiological data on regulation of follicle development and hormone production informed model design. Following implementation of the model in the Python platform, data from two OS clinical trials were used to calibrate (partially censored) and test the model. The two trials provided data on the effects of the recombinant follicle stimulating hormone preparation follitropin delta over a range of fixed doses and compared the effects of follitropin delta in agonist and antagonist OS protocols. Main results and the role of chance The computational model describes the interplay between pituitary gonadotrophins ovarian steroid hormones and follicle growth and development. Processes spanning the cellular level (such as steroidogenesis and receptor dynamics in theca and granulosa cells), the ovaries (follicle number and size) and the organism (pharmacokinetics and hypothalamic-pituitary-ovarian axis) are included as a system of coupled differential equations. Approximately 70 compartments, 500+ parameters and 1700+ reactions constitute the model framework. Follicles are recruited in the model at variable times, starting from 7 days prior to the start of stimulation. Follicles are initialised with individual sensitivity to gonadotrophins and growth rates. The fate of each follicle (dominance or atresia) is governed by these properties, which determine whether the balance shifts towards proliferation or atresia under the influence of gonadotrophins. Model-simulated numbers of follicles, their average size and mean serum hormone concentrations match closely with observed values in the two clinical trials. For each of these variables, their dependence on time, the dose of follitropin delta, AMH level and downregulation protocol were described with good precision. The inter-patient variability in the simulated number of follicles and in serum hormone levels matched closely with the observed variability based on comparison of the interquartile range. Limitations, reasons for caution The computational model currently describes follicle development during treatment with follitropin delta. We envision extending the model to encompass other treatment outcomes and additional preparations of exogenous gonadotrophins. While the model incorporates our current understanding of mechanisms important for follicle development, ongoing research in the field will likely drive improvements. Wider implications of the findings The computational model is hypothesis-generating and can improve the design of future clinical trials through in silico trial simulation. For instance, the effects of changing subject inclusion/exclusion criteria or of novel dosing regimens can be investigated. Trial registration number NCT01426386, NCT03809429
Read full abstract