An AI genetic test aims to detect postpartum depression before symptoms


A previous version of this article incorrectly stated that Dionysus Digital Health received a $6 billion grant from the Department of Defense. It received $6 million.

Postpartum depression is a leading cause of maternal death, but its diagnosis and treatment is spotty at best, negligent at worst.

Now San Diego-based start-up Dionysus Digital Health is pitching a blood test to check for the condition, even before symptoms appear. The company says it has pinpointed a gene linking a person’s moods more closely to hormonal changes. The test uses machine learning to compare epigenetics — how genes are expressed — in a blood sample with benchmarks developed during a decade of research into pregnant people who did and didn’t develop postpartum depression.

Researchers at Dionysus’s academic partners, the Royal’s Institute of Mental Health Research and UVA Health, have published peer-reviewed papers affirming their findings, and the company is partnering with the Department of Defense and the National Institutes of Health for clinical trials, with the eventual goal of making the $250 test widely available and covered by insurance. But women’s health experts say better diagnostics for postpartum depression may not help if mothers can’t access treatment and support.

One in 7 mothers experience postpartum depression. When doctors screen for the condition, they typically use a questionnaire that asks patients how much they identify with statements such as “I have looked forward with enjoyment to things as much as I ever did” and “I have blamed myself unnecessarily when things went wrong.”

If properly diagnosed, mothers rarely receive the care they need. In one widely cited study, just one-third of pregnant patients who showed signs of mental disorders received treatment — which most often consisted of verbal “reassurance” from their providers.


Stories to keep you informed

“Our aspiration is you can be in treatment before you ever even experience a symptom,” Dionysus co-founder and chief scientist Vivienne Ming said in an interview with The Washington Post. “Now we can show it’s not just in your head.”

Ming is one of many researchers using artificial intelligence to hunt for new approaches to complicated health concerns. Palo Alto, Calif.-based Delfi Diagnostics has a test that uses artificial intelligence to detect signs of lung cancer. Researchers at Children’s National Hospital in Washington built an AI tool to diagnose rheumatic heart disease in children.

But AI systems can easily exacerbate existing bias or inequity in health care. A 2019 study found that an algorithm making recommendations for C-sections wrongly flagged Black women as high risk. Another algorithm, tasked with predicting health-care needs for a large diverse group of patients, consistently recommended less care for Black patients, another study showed.

Ming acknowledged concerns about bias, cost and effectiveness. It would probably take years for Dionysus to gain approval from the Food and Drug Administration or to get insurers and employers to agree to cover the cost of the test, Ming says. In the meantime, the company says it has received a $6 million grant from the Department of Defense to validate its test in more environments. The Department of Defense didn’t respond to a request for comment.

Dionysus imagines a world where providers administer a blood test between the second and third trimesters of pregnancy that flags women at higher risk of postpartum depression and other perinatal mood disorders. This, combined with other diagnostic methods, could allow health-care systems to funnel vulnerable mothers toward treatment — and even preventive care.

The American College of Obstetricians and Gynecologists recommends providers screen patients for postpartum depression multiple times during and after pregnancy, but that doesn’t always happen, said Elizabeth LaRusso, a psychiatrist specializing in women’s health. Some people make it all the way through their pre- and postnatal checkups without a provider ever mentioning depression. Low-income women and women of color are less likely to be screened than White mothers, LaRusso’s research has found.

LaRusso said she’d welcome any tool that makes it easier to catch postpartum depression before it leads to hospitalizations, job loss or suicide. But identifying at-risk mothers is only the first step: More screening won’t make a difference if patients can’t access the care they need, such as therapy or medication, she said.

How impactful the Dionysus test proves to be will depend in part on its affordability and whether insurance companies are willing to cover its cost. Perinatal mood and anxiety disorders cost $14 billion each year in lost wages and extra expenses, researchers estimate. If flagging more depression cases could reduce subsequent medical spending, insurers might be motivated to pay for the test, Ming said.

But insurers could also view depression diagnostics as a pathway to more medical spending, as patients seek treatment they otherwise wouldn’t have pursued, said Wendell Potter, a former insurance executive who advocates for industry reform. Ultimately, insurers and employers will decide individually what new medical technology to cover. If patients end up paying out of pocket for postpartum depression screening, tests like Dionysus’s could end up exacerbating existing inequities in maternal care, Potter said.

“I doubt the majority of Americans would be able to fork out of their own bank accounts how much [the test] would cost,” he said.

Is this a safe use of AI?

As companies and researchers propose uses for AI in health care, it will be essential to audit those systems for bias, AI experts say. Since machine learning systems are trained to recognize patterns, it’s easy for them to regurgitate any biases that show up in their training data, said Mark Sendak, a data scientist at the Duke Institute for Health Innovation (DIHI).

Critically, Sendak said, an AI model’s training data should reflect the population it’s meant to serve. Dionysus, for its part, says it first validated its test with a cohort of largely White patients at Johns Hopkins Hospital in Baltimore. Its partnerships with Emory University Hospital and the Department of Defense will help it further validate its model with more diverse groups of patients, Ming said.

Without recent advancements in machine learning, Dionysus would never have been able to link a particular gene to postpartum depression, Ming said. Similar discoveries may be close behind as companies rush to apply AI to medical challenges.

But progress could come with drawbacks, said Suresh Balu, program director at DIHI. If only people with disposable income can afford early screening and preventive care, existing gaps in health-care access will get worse. Finding out you’re at risk for an illness you may never get could come with anxiety — even people with a genetic predisposition to postpartum depression may never develop symptoms if that gene isn’t activated by environmental factors, according to Ming.

Ming said Dionysus’s eventual goal is to sell the postpartum depression test directly to consumers, letting people assess their risk years before they even become pregnant. It could change the lives of mothers and children for the better, she said — if mothers can access the care they need.