How AI could help make the IVF process easier

At his fertility clinic in New York City, Dr. Alan Copperman has just finished one of several egg retrievals he’ll do that day. While Copperman has been helping people at all stages of the in vitro fertilization (IVF) process—from preparing for egg retrieval to the implantation of embryos—for more than three decades, he’s now relying on a new partner to help him make decisions: artificial intelligence.

“We've gone from such a macro level, to using information rather than anecdotal experience to help drive decisions and interventions,” said Copperman, who is the founder of fertility clinic RMA of New York and clinical professor of medicine at the Icahn School of Medicine at Mt. Sinai.

Copperman and his team of statisticians use a suite of licensed software created by Alife, a San Francisco–based company that aims to improve IVF outcomes through AI. The company—which raised a $22 million Series A round backed by Lux Capital, Union Square Ventures, and Maveron—said it trains its models on millions of de-identified data points from patient cycles to help clinicians and patients make decisions based on what has worked best for patients who are most similar, taking into account factors such as age, ancestry, weight, and existing diagnoses.

Alife’s suite of AI-enabled tools is designed to cover every step of the fertility process. Its Stim Assist tool uses a machine-learning algorithm trained on 40,000 cycles to analyze a woman’s data. It then delivers recommendations, such as the dosage of Follicle-Stimulating Hormone (FSH) a woman should take to optimize the number of healthy eggs before a retrieval cycle, and predicts the best day for retrieval. The idea is, by relying on the data and finding what worked for a similar patient, women can potentially save money on medicine and unnecessary future rounds. One complete round of IVF, including medicine, costs an average of $23,474, according to an analysis by FertilityIQ.

After the egg is fertilized to create an embryo, Alife's Embryo Assist helps labs streamline the process by grading and ranking embryos to determine the highest likelihood of success. The platform uses historical data from 12,600 embryo transfer outcomes in five U.S. clinics and an algorithm to rank the embryos. While AI is at the core of the software, Alife also includes an option for manual ranking done by humans, which allows clinicians to compare their rankings to the AI to see if they agree or made similar choices.

“I think one of the things we keep hearing is how much there is confusion in the process, a lack of transparency, and not great resources,” said Paxton Maeder-York, founder and CEO of Alife. “A big component of what we're trying to do is work with the clinics to help enable more patients to be able to receive the best quality care possible at a more affordable price, so that more people can just have access to these important areas of care.”

With more millennials delaying their family plans, same-sex couples seeking to expand their families, and insurance plans covering treatments like IVF and egg freezing, visits to fertility clinics are on the rise. In the United States, approximately 2.3% of infants born every year are conceived using assisted reproductive technology, according to the Centers for Disease Control and Prevention, which releases a new report every two to three years about the state of fertility clinics in the United States. By the year 2100, the International Committee for Monitoring Assisted Reproductive Technologies (ICMART) estimates there will be more than 400 million babies born from IVF, accounting for about 3% of the world’s population.

Copperman doesn’t only use Alife’s software to help inform decisions for patients. He also uses it to streamline his staffing and scheduling to ensure he can meet the demand.

“It's nice to know a week in advance that there are going to be 15 patients that are going to be heading toward an egg retrieval on Sunday, because you want to staff the weekend appropriately. It’s also great in visualizing data in advance at various parts because certain people are scheduled depending upon when they get their period,” he said. “But once somebody starts, it actually bakes in metrics and helps divide their predictive algorithms so we can advance where your needs are going to be on a system level.”

Building AI that works well across demographics

According to a study by the Pew Research Center, 42% of adults in the United States say they have used fertility treatments to grow their families or know someone who has, compared to 33% five years ago. Of those polled, 48% of white people were most likely to have received fertility care or know someone who had, compared to 26% of Black people. Even still, there are clear disparities across race and income that need to be closed in order to make fertility care accessible to everyone.

AI has shown tremendous potential across industries, but it can also exacerbate existing biases and disparities due to the lack of holistic training data. (Take for example, various controversies over the last few years regarding technology in health-related wearables.) Maeder-York said his team has been deliberate at building a database of de-identified training data that is “representative of demographics in the United States” to avoid this exact issue.

Working with the diversity and inclusion task force at the American Society for Reproductive Medicine, the Alife team conducted an internal analysis to confirm it had a diverse database that could yield meaningful recommendations. The team looked at ancestry, BMI, types of medical diagnoses that may be relevant, and other factors to ensure the team’s private data was an accurate snapshot of the people in the United States. Measuring for bias and inclusiveness in their models also happens to enable the team to conduct more granular research to help under-served communities.

“We have a really unique data set where a lot of this data hasn’t been collected broadly, so we're able to do really interesting research to specifically tailor to some of the health disparities that we see in IVF,” Maeder-York said. “Unfortunately, we know that minority women, especially Black women, have an 8% lower success rate. We can start trying to answer some of those questions and really tailor the algorithms and the treatment for those patients to help them have the best chance of having a baby.”

Dr. Kylie Dunning, a reproductive biologist and associate professor at the University of Adelaide, who has authored research on the use of AI in IVF, said the challenge over the next few years will be for more peer-reviewed research that validates AI-assisted approaches.

“That’s a real critical need for us to know if these approaches are actually better than standard practice to ensure we aren’t offering false hope,” she said.

In the future, Dunning said she hopes to see more collaborative efforts between clinics, researchers, and other entities to standardize information and build larger datasets, which she said could “make a bigger, better outcome in predicting success.”

As Alife continues to grow its business and make its software available to more clinics, Maeder-York is hopeful that big data coupled with AI can help make that process a little smoother during what can be a financially and emotionally draining journey.

“It can cost as much as $70,000 to have a baby,” he said. “Anything we can do with advanced analytics and to better our software to help more patients get pregnant faster at a lower cost is really the goal of the company.”

This story was originally featured on Fortune.com