A study led by researchers at Alife Health, a fertility technology company building artificial intelligence (AI) tools designed to improve in-vitro fertilization (IVF) outcomes, found that a machine learning model may help doctors optimize the starting dose of follicle stimulating hormone (FSH) to improve patient outcomes for a significant number of patients.

When undergoing ovarian stimulation during IVF, patients are given FSH to promote multifollicular development. In this process, a key decision for physicians is what starting dose to use. Too little starting FSH may lead to inadequate follicle recruitment and cycle cancellation, while too much FSH may lead to excessive response and ovarian hyperstimulation syndrome. The choice of starting FSH is often individualized to patients based on their age and ovarian reserve, as well as the observed response to FSH in previous cycles. Ultimately, the goal is to select a starting dose that will yield an optimal number of good quality oocytes while reducing the risks of overstimulation.

The study, published online in Reproductive BioMedicine Online, is one of the first to utilize machine learning for optimization of starting FSH dose, which has historically been a subjective decision that can vary widely between different doctors, clinics, and countries. Adoption of the published machine learning tool would represent a step towards a more quantitative and standardized approach for the selection of FSH starting dose.

For their analysis, conducted with collaborators at RMA New York, Boston IVF, and RSC Bay Area, the researchers drew from over 18,500 historical IVF cycles that were performed at multiple centers between 2014 and 2020.

Study results suggest that if Alife's machine learning model had been used in the cycles studied, it could have helped doctors retrieve an average of 1.5 more mature oocytes (eggs), 1.2 more fertilized oocytes (eggs fertilized by sperm), and 0.6 more usable blastocysts (embryo) for dose-responsive patients that were given a non-optimal dose of FSH. For non-responsive patients given a high dose of FSH (i.e. those where there is no improvement by changing the FSH dose), comparable egg outcomes could have been achieved with significantly less FSH than what was prescribed, reducing the cost of an often prohibitively expensive procedure. In the U.S. market, this could lead to savings of up to $2,100 on average for this patient group, assuming an average cost of $1.50/IU of FSH.

“This retrospective study indicates that our machine learning model may help physicians identify an optimal dose of starting FSH that is often lower than what is currently used. A reduction of FSH dose has the added benefit of cost savings for patients,” says the study’s senior author, Kevin Loewke, head of data science at Alife. “We look forward to entering the clinic and performing prospective studies in the near future to confirm these retrospective findings.”

“These promising results further indicate that we are on the right path towards utilizing AI to improve the effectiveness of IVF for our patients,” says the study’s co-author, Eduardo Hariton, MD, MBA. “As we aim to leverage technology to not only improve the outcomes for our patients, but also increase the efficiencies and expand access to care, clinical decision support tools like this one will be a key piece of the puzzle.”

The study, titled “An interpretable machine learning model for individualized gonadotropin starting dose selection during ovarian stimulation,” was led by Michael Fanton, PhD, senior data scientist at Alife and co-authored by:

Paxton Maeder-York, MS, MBA, CEO of Alife Health
Eduardo Hariton, MD, MBA, reproductive endocrinologist at RSC Bay Area later this year
Oleksii Barash, PhD, HCLD, IVF laboratory director at RSC Bay Area
Louis Weckstein, MD, reproductive endocrinologist at RSC Bay Area
Denny Sakkas, PhD, CSO of Boston IVF
Alan Copperman, MD, FACOG, reproductive endocrinologist at RMA New York
Kevin Loewke, PhD, head of data science at Alife Health

To read the RBMO article describing this study, please click here.