The ovary is one of the first organs to age in the human body. At birth, ovaries contain all the eggs they will ever have, representing their “ovarian reserve”. The ovarian reserve dictates the reproductive lifespan as it undergoes a natural winnowing over time. Menopause is caused by the depletion of ovarian reserve (Johnson, Emerson and Lawley, 2022; Meng et al, 2018).
The ovaries are both pacemakers of healthy aging in women, as well as early warning systems of future pathologies. Despite great variability in age of natural menopause (ANM), and its importance as a marker for overall health, we are currently not able to predict its onset and our understanding of the correlates of ANM is rudimentary. The factors driving ovarian aging and therefore menopause remain poorly understood. We are interested in understanding the contribution of modifiable versus non-modifiable ones. Modifiable ones include hormonal, metabolic, reproductive and lifestyle factors while non-modifiable ones include genetics and gestational factors.
We will use machine learning to derive various algorithms that will serve as biomarkers of ANM. We will use ML best suited for regression analyses (eg random forest, support vector regression and k-nearest neighbors), prioritizing algorithms that allow for explainability of the model. We will build multiple ML models in stages: 1) blood makers only; 2) blood and questionnaire data on lifestyle factors; 3) blood, lifestyle and physiological data; 4) blood, lifestyle, physiological and genetic data.
We will use the following features:
! Blood markers: CBC (n=31); NMR measures (n=249); proteomics data (n~2,900)
! Reproductive factors: Gynecological and obstetric history
! Lifestyle factors: Smoking, alcohol, diet, exercise
! Physiological (VO2 max) and body composition scores
! Genetic Risk Score: A polygenic risk score (PRS) for ANM.
! Gestational factors: Singleton versus multiple birth; low birth weight.
