Digital health · Obstetric risk prediction
Elythea
A Y Combinator company that set out to flag life threatening pregnancy complications at the first prenatal visit, and now sells voice AI outreach to Medicaid and Medicare Advantage plans. The obstetric models have one published result and no registered trial.
The number that started it
Most American labor and delivery units decide who might bleed to death using a paper checklist. A nurse scores prior cesareans, multiple gestation, placenta problems and a few other factors, usually once the patient is in the room, and sorts her into low, medium or high risk.
In a 2024 conference abstract, Reetam Ganguli and Stephen M. Wagner gave its yield. The tool from the California Maternal Quality Care Collaborative, they wrote, “only classifes 22% of severe PPH patients as high risk and includes variables only accessible during the labor period.” The misspelling is in the published abstract.
Four out of five women who hemorrhage severely enough to need a transfusion are never flagged. Everything Elythea built sits in that gap.
What the country's numbers look like
In 2024, 649 women in the United States died of maternal causes, a rate of 17.9 per 100,000 live births, per the National Center for Health Statistics, not significantly lower than 2023. The rate for Black non-Hispanic women was 44.8, against 14.2 for White non-Hispanic, 12.1 for Hispanic and 18.1 for Asian women.
Those disparities are why obstetric prediction has to be handled carefully. For fourteen years the standard tool for counselling women about vaginal birth after cesarean asked two yes or no questions, “African-American?” and “Hispanic?”, and returned lower success odds when the answer was yes. In 2021 the Maternal-Fetal Medicine Units Network replaced it with a version that drops race and ethnicity and adds treated chronic hypertension instead. A model that scores pregnant women has to answer for what it feeds on.
A medical student who left
Ganguli was accepted to Brown University's medical school at 17. He read the obstetrics textbooks, interviewed more than 80 providers, and heard the same two answers back: postpartum hemorrhage and hypertensive disease. He left medical school for the company. The name is his shortening of Eileithyia, the Greek goddess of childbirth.
His co-founder Rishik Lad, a computer science student at Dartmouth, had known him since middle school. In March 2023 the pair won first place and $25,000 at Brown's sixth Venture Prize. The student newspaper's account names a third co-founder, Stephen Wagner, then an assistant professor of obstetrics and gynecology at Brown. Conflict of interest statements in journals Wagner published in during 2025 and 2026 describe him instead as a consultant and an advisor. One, in Frontiers in Public Health, calls the company “an artificial intelligence medical startup examining risk factors for postpartum hemorrhage.”
From the exam room to the member list
The original product plugged into electronic health records so obstetricians saw a risk score at the first visit. The market did not want it. Ganguli's explanation to the Jewish Healthcare Foundation in 2025 is blunt: obstetrics is often cash flow negative for a hospital, and a solution that takes nine to twelve months to pay back is a hard sell.
So the company followed the money to the entity that carries the risk. By January 2025 the website described a digital care manager called Clara texting Medicaid members. By August 2025 the same domain was headed “Voice AI for Member Outbound”, an AI call centre for managed care: HEDIS and STARS gap closure, prescription reminders, welcome calls. By June 2026 the title read Voice AI for Medicaid and Medicare Advantage. The pitch that begins with hemorrhage now ends with member engagement.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Published model | SMFM 2024 abstract 171, in the AJOG supplement. Retrospective cohort from the CDC Vital Statistics System, 2018 to 2020 plus hemorrhage cases from 2014, 2016 and 2017. Trained on 7,728,892 patients; severe hemorrhage needing transfusion occurred in 1.2%, or 89,819 cases. An extreme gradient boosting model across 39 variables reached AUC 0.75, 78.2% accuracy and sensitivity 0.65. Top factors: body mass index and maternal age. | Public record |
| How it was validated | Validated on CDC data from 2021, a later year of the same birth certificate file. Temporal validation inside one dataset, not validation in an independent health system. | Public record |
| What the inputs were | Training variables were demographics, medical history and prenatal data; variables above 50% missingness were omitted. The 39 are not listed. Whether race is among them is not stated. | Not stated |
| Patent application | US 2025/0069754 A1, filed Aug 22, 2023, published Feb 27, 2025, assigned to Elythea, Inc. by Ganguli, then abandoned for failure to respond to an office action. Its training source's demographic fields are listed as “race, age, marital status, educational attainment, live birth order, birth interval, etc.” and it describes synthetic records for “other minority classes like race”. That is the 2023 draft, not what any shipped model uses. | Public record |
| Registered trials | A ClinicalTrials.gov query for Elythea returned zero studies on Sep 24, 2026, and no registration for the randomized trial described since 2023. | Not found |
| FDA status | No 510(k) and no PMA record. A score a clinician can independently review may sit outside the device definition under the Cures Act criteria in FDA's Clinical Decision Support Software guidance, reissued January 2026. No determination was published. | Not found |
| Prospective studies, as described | In October 2023 the company said it had “prospectively tested our models on 2,000+ women enrolled from 13 sites across Cameroon, Nigeria, Uganda, Texas, and Rhode Island” and had begun a randomized trial across more than 1,500 women in four African countries. No results, registration or protocol found. | Company-stated |
| Outcome claims | A technology directory reports 80% reduction in preeclampsia, 68% in hemorrhage, 45% fewer C-sections and about $863 saved per patient. No supporting study was found. | Company-stated |
| Accuracy claims | In 2023 Ganguli told a student newspaper the startup had a 90% predictive accuracy rate against 40% for competitors; an AI directory that year reported accuracy up to 85% and AUC up to 0.85. The abstract reports 78.2% and 0.75. | Differs from the record |
| Publication count | The January 2025 site claimed “over a dozen peer-reviewed studies accepted for publication in the top obstetric journal (AJOG)”. PubMed indexes no journal article carrying an Elythea affiliation, and SMFM abstracts in the AJOG supplement are conference abstracts. | Differs from the record |
| Category evidence | Hemorrhage models travel badly. A 2022 external validation found models reported at C-statistic 0.93 fell to 0.57 in a record system measuring blood loss quantitatively. A 2026 study of CDC natality data found eclampsia model discrimination was lower for Black, American Indian or Alaska Native and Native Hawaiian patients, and for those born abroad or on lower incomes. | Public record |
Read plainly: one abstract showing a birth certificate model that beats a paper checklist on sensitivity and lands at AUC 0.75. A real result, and a modest one. Everything larger exists only as the company's description of itself. The literature says such models lose accuracy when they move, which is why external validation matters more here than in most categories, and it is the piece that has not been shown.
What to watch
- A registered trial. The randomized study has been described as under way since 2023 with no registry entry found.
- A peer reviewed paper with an Elythea affiliation, reporting the African and US cohorts in full.
- Disclosure of the 39 model variables, and whether race is one.
- Any named health plan customer. The Fortune 25 Medicaid plan is unidentified.
In their words
“what if we could use more advanced machine learning frameworks to detect and forecast your risk as soon as the first day of your pregnancy, or as soon as your first prenatal visit?”
Reetam Ganguli, founder and CEO, National Patient Safety Board podcast, 2024 · Interview
“It was frustrating to see how clinical care was fundamentally reactive to these morbidities instead of being proactive.”
Reetam Ganguli, Jewish Healthcare Foundation Q and A, 2025 · Interview
“There is limited financial incentive for proactive and preventive care as it limits the charge for services rendered and can actually be counterintuitive to top-line profit, even in value-based systems.”
Reetam Ganguli, Jewish Healthcare Foundation Q and A, 2025 · Interview
“platform that obstetricians use in any setting to predict risk for complications of pregnancy”
Reetam Ganguli, quoted in The Brown Daily Herald, March 2023 · Independent
“Any venture using AI will hit the class imbalance issue: low positives/low proportion of minority patients will lead to poor AUC/accuracy.”
Elythea, MIT Solve CURE Challenge application, 2023 · Company-stated
“It’s case in point to me that our clinical tools can remain scientifically rigorous and still clinically useful without race correction.”
Darshali Vyas, resident, Massachusetts General Hospital, on the VBAC calculator, STAT, 2021 · Independent
Related companies
Sources
- Public recordAbstract 171, machine learning for severe postpartum hemorrhage
- Public recordUS patent application 2025/0069754 A1
- Public recordMaternal Mortality Rates in the United States, 2024
- Public recordClinical Decision Support Software guidance
- Public recordClinicalTrials.gov query for Elythea, zero studies
- Public recordExternal validation of hemorrhage prediction models
- Public recordEclampsia risk prediction across diverse U.S. populations
- Public recordA VBAC calculator without race and ethnicity
- Public recordDisclosures naming Stephen Wagner as advisor to Elythea
- IndependentResearchers remove race from a calculator for childbirth
- IndependentStudent startups vie for slice of $50,000 prize pool
- IndependentPatient safety competition winners at CES 2025
- InterviewBeyond the Prize: Reetam Ganguli and Elythea
- InterviewEpisode 22: Meeting the Challenge
- IndependentElythea company and funding profile
- CompanyCompany page and Launch YC post
- CompanyElythea CURE Challenge application
- Companyelythea.org archived captures
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
