Everlywell's $249 AI Test Reads Mammograms for Cancer Risk
Update, Oct. 3: two follow-ups since launch day. First, the guideline hedge in this article is now retired — the 2026 NCCN Clinical Practice Guidelines for Breast Cancer Screening and Diagnosis include AI-based mammogram risk assessment as an option for identifying increased risk from age 35 (confirmed by Techstrong.ai and the launch announcement's guideline citation). Second, new practical details from Inc.com's Oct. 2 interview: the $249 test is HSA/FSA-eligible, Everlywell pulls your mammogram imagery electronically so there's no CD-chasing, and eligibility excludes women with a known BRCA mutation or those currently in cancer treatment. Worth knowing for the intervention question: Lehman says risk scores "change meaningfully over time," and the team is studying whether intervention can bring scores down — with low-dose Tamoxifen lowering estrogen-receptor-positive risk by up to 50% in some women.
Everlywell's $249 AI Test Reads Mammograms for Cancer Risk
On the first day of Breast Cancer Awareness Month, Everlywell put the first FDA-authorised AI for predicting breast-cancer risk from a routine mammogram on its consumer platform — a $249 test, not covered by insurance, that reads the mammogram you already had. The science is real. The care pathway it drops you into is not.
Photo: A modern 3D mammography machine, shown for illustration. Clairity Breast estimates five-year breast-cancer risk from a standard screening mammogram — no new imaging needed. (Aboutmovies / Wikimedia Commons, CC BY-SA 4.0)
Your mammogram knows something about your future that your radiologist cannot see. That is the pitch, anyway — and as of Thursday, it is a product you can buy. Everlywell — yes, the at-home testing company — put Clairity Breast on its consumer platform on October 1, which is not a coincidence: it's the first day of Breast Cancer Awareness Month. Here's the product: an AI that estimates a woman's five-year risk of developing breast cancer from a standard screening mammogram. The one she already had. No new imaging. Every result gets reviewed by a licensed provider before it's released.
The New York Times covered the launch the same day — Roni Caryn Rabin's piece, "Women Can Now Assess Their Breast Cancer Risk With The Help Of A.I." The price, per the Times: $249, not covered by insurance. We went looking for that price on Everlywell's own product page and couldn't confirm it, so take the number as reported, not verified.
Marketing aside, this is a real regulatory milestone in a consumer-product costume. Clairity Breast is the first AI tool with FDA De Novo authorisation for predicting five-year breast-cancer risk from a routine mammogram. De Novo is the pathway for novel devices with no predicate — nothing like this existed before, which means every future competitor has to clear the bar Clairity just set. But keep the qualifiers on "first of its kind", because the mammography-AI field is crowded, and the gap between detecting cancer and predicting it is the entire story.
What the AI actually does — and doesn't
OK, so what does the thing actually do? Strip away the launch-day language and it's refreshingly un-mysterious: software. Not a scanner. Nobody's getting a new mammogram for this. The qualifying setup is a screening mammogram from within the last 12 months — the one she already had — fed through Everlywell's platform. The AI reads the standard 2-D images and returns a five-year risk estimate. Three buckets — low, intermediate, high — plus a percentage probability. Here's the part the marketing buries slightly: a licensed provider in Everlywell's clinician network orders the test, reviews the result, and returns it with next steps drawn from national guidelines. Prescription-gated, in other words, not a pure DIY readout. She can take the result to her own doctor, and Everlywell says its clinical team answers questions.
The model, developed by Boston-based Clairity, Inc. (founded 2020 by Dr. Constance "Connie" Lehman, Harvard Medical School professor of radiology and former chief of breast imaging at Massachusetts General Hospital), is a deep convolutional neural network trained on mammograms paired with at least five years of follow-up outcomes. It looks for micro-patterns of tissue architecture that are, in Lehman's phrase, invisible to the human eye. "The information we need to understand a woman's breast cancer risk has been in her mammogram all along – we just haven't been able to interpret it until Clairity Breast," Lehman said in the launch announcement. "Now, it can also inform conversations with your healthcare provider about the best screening and risk reduction options for you."
The don'ts matter as much as the dos, and the FDA was blunt about them: this device does not diagnose breast cancer. It does not detect it. It does not make care recommendations, and it does not tell anyone how to interpret an imaging exam. I keep coming back to a mental image — call it the sorting hat for screening, if you'll forgive the metaphor. It puts you in a lane. It doesn't drive the car.
Everlywell's founder and chief executive Julia Cheek framed the launch as access: "Women deserve clear answers about their health, and access to the most advanced tools in health care, not years from now, but today… a woman can learn more about her future breast cancer risk from the mammogram she already gets, and walk into her doctor's office with information she's never had before."
Clairity Breast FAQ: who qualifies, HSA/FSA, and how Everlywell gets your mammogram
Do I qualify? You need a mammogram from within the last 12 months, no known BRCA mutation, and you can't be currently undergoing cancer treatment.
Is it covered by insurance or HSA/FSA? It's not insurance-covered — but it IS HSA/FSA eligible.
Do I need to chase my imaging center for the files? No. Everlywell pulls the imagery electronically: "You do not lift a finger after you hit 'order'" (Julia Cheek, via Inc.com).
Who actually reads my result? A licensed provider in Everlywell's clinician network orders, reviews, and releases it with next steps.
The FDA authorisation — and what De Novo really means
We read the FDA's De Novo decision summary ourselves — it's public, and the limitations section is the most important paragraph in the whole story. The details: De Novo request DEN240047, device name Allix5 (Clairity Breast's regulatory name), sponsor Clairity, Inc., final FDA decision May 30, 2025, Class II device under 21 CFR 892.8500, product code SEZ. Sale and use are restricted to prescription use.
De Novo is the key term and it is worth understanding, because it is doing a lot of work in the marketing. Most medical AI reaches the market through the 510(k) pathway, which requires showing your device is "substantially equivalent" to an already-cleared predicate. De Novo is for novel devices with no predicate — the FDA's way of saying nothing like this existed before. That is a real regulatory bar, and it is also a moat: future competitors in five-year mammogram risk prediction will have to meet or exceed Clairity's evidence package.
The company is backed by Santé Ventures and ACE Global Equity and raised a $43 million Series B in November 2025. Until this week, you could only get Clairity Breast through health systems — Beth Israel Deaconess in Massachusetts, Invision Sally Jobe in Colorado. Thursday's launch is the first time it's been sold directly to consumers. The product moved from a hospital's risk-stratification workflow into a shopping cart.
Photo: A mammography technician demonstrates a mammogram machine during Breast Cancer Awareness Month in 2016. The Everlywell launch was timed to the start of this year's awareness month. (US Air Force / Staff Sgt. Teresa Cleveland, public domain — file photo, not connected to Everlywell or Clairity)
The NCCN guideline update — now confirmed
The hedge in the original version of this article can now be retired. The 2026 NCCN Clinical Practice Guidelines for Breast Cancer Screening and Diagnosis include AI-based mammogram risk assessment as an option for identifying increased risk from age 35 — confirmed by Techstrong.ai and the launch announcement's own guideline citation.
The numbers: why I'm taking this one seriously
I'll say this plainly: the evidence package here is serious by medical-AI standards, and I've read enough AI-health press releases to know the difference between a real dossier and a vibe. For the FDA submission, Clairity put up more than 77,000 screening mammograms from five geographically diverse US sites — all with five-year outcome data, which is the part that actually costs money and time. Then, after authorisation, came real-world testing: more than 120,000 mammograms across 10 US facilities, presented at RSNA 2025, landing at an AUC of 0.72 with good calibration. For the non-statisticians: the scores didn't just rank women in the right order, they tracked actual outcomes. That's the difference between a clever demo and a clinical tool.
And the paper trail goes back further than the product does. Lehman's team published in Radiology back in 2019 — 88,994 mammograms from 39,571 women — with image-only deep learning at AUC 0.68 against 0.62 for Tyrer-Cuzick, the questionnaire model everyone actually uses in clinics. A hybrid version hit 0.70. A 2022 Journal of the National Cancer Institute study had the model beating traditional risk tools again. Then the RSNA 2025 data — 421,499 mammograms, 27 facilities, Europe, South America and the US — with the model picking out women at two to four times higher risk. Including women with non-dense breasts, which is the exact group density-based assessments keep missing. That last detail is doing a lot of quiet work.
Here's the clinical logic, and why it lands even if you're skeptical of AI health products generally. Most women who get breast cancer — 85%, no family history. Roughly half have no identifiable risk factors at all. The WHO's 2022 count was 2.3 million new cases worldwide. The old models, Tyrer-Cuzick and Gail, run on family history and questionnaires — which is precisely why they miss most of the women who actually get sick. An image-only model throws the questionnaire away entirely. And in the pivotal data, 37% of women in their 40s came back intermediate-risk, 16% high-risk — distributions you'd expect from women a decade older. So: either a screening revolution, or a recall-rate time bomb. Which one depends entirely on what happens after the score lands — and nobody, not Clairity, not Everlywell, has figured that part out yet.
The oncologists are, predictably, enthusiastic. Dr. Larry Norton of Memorial Sloan Kettering — he's the founding scientific director of the Breast Cancer Research Foundation — called the FDA authorisation "a turning point for more women to access the scientific advances of AI-driven cancer risk prediction." His case, in his words: "Breast cancer is rising, especially among younger women, yet most risk models often miss those who will develop the disease. Now we can ensure more women get the right care at the right time." And Robert A. Smith of the American Cancer Society was blunter about the screening logic: "Personalized, risk-based screening is critical to improving breast cancer outcomes, and AI tools offer us the best opportunity to fulfill that potential."
The "first" claim, honestly framed
Now the necessary pedantry, because "first FDA-authorised AI" is doing heavy lifting in every headline on this story — including, potentially, this one's. The claim is corroborated by multiple independent outlets in its precise form: the first AI tool to receive FDA De Novo authorisation for predicting a woman's five-year breast-cancer risk from a standard screening mammogram (MedTech Dive, Applied Radiology, June 2025).
Drop any of those qualifiers and the claim breaks. iCAD markets its ProFound AI Risk as "the world's first" clinical decision-support tool for two-year breast-cancer risk estimation from a screening mammogram — a different time horizon and a different regulatory pathway. And the mammography-AI field is full of FDA-cleared tools for detection, triage and density assessment — Therapixel's MammoScreen, iCAD's ProFound AI, ScreenPoint's Transpara, Lunit's INSIGHT — with computer-aided detection dating back to 1998. As Lehman herself told MedTech Dive: "most of these tools are used to detect cancer, not predict future risk."
The honest frame: Clairity is first-in-class for future risk prediction, not for mammography AI generally. Detection AI finds the tumour that's there. Prediction AI finds the woman who needs watching. Those are different products, different evidence bars, and — crucially — different failure modes.
Clairity Breast: what a high risk score actually buys you
Which brings us to the $249 question — the one the press release doesn't ask. A high five-year risk score arrives with "next steps based on national guidelines," reviewed by a licensed provider. But what is the next step for a 38-year-old flagged high-risk? Supplemental MRI and ultrasound coverage is inconsistent across insurers; the 2026 NCCN guideline update that reportedly adds AI-based mammogram risk assessment as an option from age 35 — a company claim we could not independently verify against the NCCN text — is new and untested in practice; and the test itself isn't insurance-covered. The product hands a woman a number and then drops her into a care system that hasn't finished building the road the number points down.
The radiology literature has a name for the downstream risk: automation bias. Chen et al., writing in Diagnostic Imaging in September 2025 on AI risk-flagging in screening, warned that "recall rates would increase markedly if interpreting radiologists were to feel compelled to recall all examinations flagged as intermediate risk by AI" — and that the human tendency to defer to the computer "could contribute to many potentially unnecessary recalls." A September 2026 systematic review in MDPI, covering ten years of screening-mammography AI, was blunter still: "interval-cancer and mortality outcomes are still awaited in every trial."
The FDA's own position, quoted in October coverage of mammography AI, is that these tools belong in adjunctive roles rather than as wholesale replacement for standard interpretation — and that "more technology doesn't automatically mean lower rates of breast cancer overdiagnosis and fewer false positives." Overdiagnosis is the shadow side of every screening advance: finding risk, and then finding disease, that would never have harmed the patient, and treating it anyway.
None of this is Clairity-specific evidence — the bias and validation critiques apply to the field. But they apply to a field that has just gone direct-to-consumer, which changes the stakes. In a hospital workflow, a risk score lands in front of a clinician who ordered it. On a consumer platform, it lands in front of a woman who paid $249 for answers and gets a probability instead.
Our take: a real milestone wearing a costume
Let's give credit where it's due, because the milestone is genuinely rare. A De Novo authorisation is not a rubber stamp — it is the FDA saying this device had no predicate, reviewing a 77,000-mammogram evidence package, and imposing prescription-use restrictions with explicit "not for diagnosis" limitations. In a year when AI health products routinely launch first and seek permission later — or never — Clairity did the slow, expensive, regulated thing. That deserves to be said plainly.
But the launch deserves scrutiny in equal measure, because what Everlywell is selling is not the FDA's careful, clinician-mediated risk tool. It is that tool repackaged as a consumer product, at a price the uninsured pay out of pocket, at the start of the month when pink-ribbon marketing is at its annual peak, with next steps that depend on a care pathway — supplemental screening coverage, guideline adoption, provider familiarity — that is still being built. The science cleared the FDA's bar. The business model cleared no one's.
The question worth asking isn't whether the AI works — the AUCs, the calibration, the 2–4x risk stratification all say it does something real. It's what a "high risk" score buys a woman who isn't at Beth Israel Deaconess, who doesn't have a breast-imaging specialist on speed dial, who paid $249 and now has five years of worry with no guaranteed follow-up. Until the answer to that question is as validated as the model, this is a prediction in search of a plan — and predictions, unlike diagnoses, don't come with a treatment protocol.
Sources: Everlywell Business Wire press release (Oct 1, 2026); FDA De Novo summary DEN240047 (accessdata.fda.gov) and FDA AI-enabled medical devices list; New York Times (Roni Caryn Rabin, Oct 1, 2026, via KFF Health News briefing); AuntMinnie (Oct 1, 2026); Applied Radiology; MedTech Dive (June 2025); Cancer Therapy Advisor/HealthDay (June 2025); News-Medical (Nov 2025, RSNA data); ITNOnline (RSNA 2025); Diagnostic Imaging (Sept 2025); MDPI systematic review (Sept 2026). $249 price and insurance status reported by the NYT; NCCN guideline claims are company claims not independently verified. Internal-link slugs curl-verified (200) before inclusion.
Comments
Post a Comment