HormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics

Abstract

The hypothalamus is the central regulator of reproductive hormone secretion. Pulsatile secretion of gonadotropin-releasing hormone (GnRH) is fundamental to physiological stimulation of the pituitary gland to release luteinizing hormone (LH) and follicle-stimulating hormone (FSH). Furthermore, GnRH pulsatility is altered in common reproductive disorders such as polycystic ovary syndrome (PCOS) and hypothalamic amenorrhea (HA). LH is measured routinely in clinical practice using an automated chemiluminescent immunoassay method and is the gold standard surrogate marker of GnRH. LH can be measured at frequent intervals (e.g., 10 minutely) to assess GnRH/LH pulsatility. However, this is rarely done in clinical practice because it is resource-intensive, and there is no user-friendly and accessible method for computational analysis of the LH data available to clinicians. Here we present hormoneBayes, a novel open-access Bayesian framework that can be easily applied to reliably analyze serial LH measurements to assess LH pulsatility. The framework utilizes parsimonious models to simulate hypothalamic signals that drive LH dynamics, together with state-of-the-art (sequential) Monte-Carlo methods to infer key parameters and latent hypothalamic dynamics. We show that this method provides estimates for key pulse parameters including inter-pulse interval, secretion and clearance rates and identifies LH pulses in line with the current gold-standard deconvolution method. We show that these parameters can distinguish LH pulsatility in different clinical contexts including in reproductive health and disease in men and women (e.g., healthy men, healthy women before and after menopause, women with HA or PCOS). A further advantage of hormoneBayes is that our mathematical approach provides a quantified estimation of uncertainty. Our framework will complement methods enabling real-time in-vivo hormone monitoring and therefore has the potential to assist the translation of personalized, data-driven, clinical care of patients presenting with conditions of reproductive hormone dysfunction.

Quantitative approaches in clinical reproductive endocrinology

“…A key strength of the BSA method is that frequency estimates come in the form of Bayesian posterior distributions, facilitating the estimation of uncertainty and hypothesis testing. Finally, Bayesian extensions to the deconvolution method [ 13 , 16 , 17 , 18 ] as well as a recently proposed Bayesian framework for inference of LH dynamics based on mechanistic models of pulse generation [ 19 ] enable uncertainty estimation of parameters as well as estimation of latent hypothalamic dynamics.…”