Aurevia is an AI platform, built by a physician, for patients between visits and the ER doctors who see them.
It never replaces your doctor, and it never diagnoses.
It began with my mom.
My mom has lupus, Sjogren's, rheumatoid arthritis and myasthenia gravis. Her thyroid was removed for Graves'. She's on immunotherapy.
A few weeks ago she called me in pain. Swelling, new symptoms, scared. Is it the medicine or the disease? Does she go to the ER tonight or wait? Her doctor can't see her for weeks. English is her second language, and getting a ride isn't always easy.
She got through it because she has a doctor in the family. Most patients don't. And she's not unusual. She's the patient ER doctors see every shift, and the one who comes back.
When she does end up in the ER, the doctor has minutes. Her story isn't in the chart. Results keep coming in, and it's not obvious which ones matter for her.
Dr. Ammar Ahmed, DO · Founder and CEO
At home
The app keeps her story in one place: her conditions, her medicines, her past visits, wearables, photos of her papers and pill bottles, voice notes, readings from home.
Works in her language
By voice or by tap.
She decides what to share
One consent. Information flows only where she allows.
In the ER
Her story comes with her. Then the results start landing: labs, vitals, notes, imaging.
Most don't change anything
A few change the whole plan, and it isn't always obvious which ones.
Aurevia works it out as each one arrives
Using what it knows about her, so the doctor doesn't have to dig through the chart.
Diagnoses ranked and re-ranked
As the picture changes, it turns what the doctor does into a finished note.
The doctor makes every decision.
How it connects at NYP/Weill Cornell
Her story is ready at triage, so decisions come faster and ER rooms open sooner.
Fewer patients come back, so inpatient beds free up, boarding drops, and ER rooms open sooner too.
At a full hospital, an open room or bed goes to the next patient, at NYP's normal margin.
That means more patients seen, fewer walkouts, and better scores on the CMS timeliness measure that becomes mandatory in 2028.
Accurate severity runs alongside: correct payment for each admitted stay.
Does a faster ER visit create new money?
Only when the ER is full.
NYP's median ER stay for discharged patients is about 258 minutes, against 162 nationally, which points to an ER short of space.
When a room opens sooner at a full ER, the next patient is seen instead of leaving, being diverted or going elsewhere. That visit is new revenue for NYP.
A published study of three EDs and 170,723 visits valued a 5% throughput gain at a capacity-constrained ED at $66.02 per visit.
When a readmission doesn't happen, who gains?
Most people assume the payer keeps the savings. At NYP, the hospital usually gains.
NYP runs full, so a freed bed goes to the next patient. Medicaid stays are often paid below cost, and New York Medicaid penalizes excess readmissions. Medicare penalizes excess readmissions for six conditions, and many plans deny or bundle readmissions. Uninsured stays are largely unpaid.
The payer keeps the savings in one case: a commercial patient when NYP has open beds. Our model counts only the refilled bed. Penalties, denials and below-cost stays avoided are upside.
The model for one NYP ER
NYP/Weill Cornell Medical Center: one ER, about 96,000 visits a year.
About $4.2M a year conservative, $8.6M center.
Faster decisions
$3.17M to $6.34M
Accurate severity
$0.71M to $1.41M
Fewer returns
$0.32M to $0.89M
Flat fee: $400,000 a year. NYP keeps $3.80M to $8.24M, 90% to 95%.
Payback in 1.3 to 2.7 months. Three-year net present value: $7.8M to $17.2M.
Even if the ER isn't full, modeled value stays above the fee.
Modeled from published studies. The Weill Cornell study replaces each number with NYP's own.
The ER group: randomized by pod and shift
Randomized time blocks, one illustrative week.
Automatic enrollment
Every eligible patient in a block is enrolled automatically in the EHR. No bedside consent.
QR code at discharge
Only ON blocks hand out the QR code for the app at discharge.
No crossover
A physician never switches arms within a block, so what Aurevia shows for one patient can't change care for the next patient in the control group.
Balanced blocks
Blocks are balanced by day and time, and nobody knows a block's arm until it starts. Short blocks add about 15% to the sample size. Week-long blocks would roughly double it.
Triage handoff
When a patient using the app comes back, the triage nurse sees her history on a one-screen handoff card. Triage is faster and more complete, and better context helps move patients along sooner.
Main outcome: Time from arrival to a disposition decision.
Also measured: Time to a treatment space, patients leaving before evaluation, the four parts of the CMS timeliness measure, length of stay, ER returns at 7 and 30 days after discharge, and documentation accuracy on blinded review in both directions. For returning app users: how complete the handoff card is, and time to see a clinician.
Safety: Returns within 72 hours that end in admission, and missed diagnoses on blinded chart review.
About 2,400 patients, after adjusting for clustering.
The readmission group: no inpatient platform needed
Who: higher-risk adults admitted through the ER and discharged home, with at least one need such as new medicines, pending tests, no follow-up booked, or cost, transportation or language barriers.
How
Before hospital discharge, patients are randomized one to one to the app or usual care. Inpatient teams change nothing. On discharge day, the nurse hands out the QR code for the app.
What the app carries
Her full ER visit and hospital stay: what her doctors found, what they did and why, and the plan for going home. It uses that to guide her step by step for 30 days.
Optional unit randomization
If the department wants Aurevia to support the inpatient discharge itself, we randomize by discharging unit instead. That adds about 20% to the sample.
Main outcome: unplanned readmission within 30 days.
Also measured: readmission at 7 days, ER returns without admission, medicines picked up within 3 days, follow-up visits completed, and time from a warning sign to action.
Safety: a blinded panel reviews every return. A return that happened because the app correctly caught a problem doesn't count as a failure.
Sample size: to detect a drop from 20% to 15%, with 90% power, we need 1,212 patients per group. That's about 2,700 in total, allowing for patients lost to follow-up, or about 3,200 if randomized by unit.
How it runs: three tracks from day one in parallel
Weill Cornell
Designs the study, secures approvals and checks Aurevia against its own past records.
NYP procurement
Agrees the price, the shared KPIs and the conversion terms, and completes security review.
Aurevia
Finishes its Epic integration at Yale New Haven.
When all three are done, Aurevia runs silently on live NYP data. No one sees its output. This phase exists to confirm it's safe and accurate before anyone uses it. Randomized enrollment starts only once it passes.
Two campuses: NYP/Weill Cornell Medical Center and NYP Lower Manhattan. About 215 patients a week over about 26 weeks. Final numbers come from 6 to 12 months of NYP's own data.
A day in the study, and what we need
What NYP provides
- A site lead and IRB submission
- Past records for the first accuracy check
- The choice of campuses, eligibility and block schedule
- A team to act on warning signs after discharge, such as virtual care or Community Tele-Paramedicine
- Independent analysis by your research team
What Aurevia provides
- The Epic integration, at Aurevia's cost
- Accuracy and silent-phase safety reports
- Training and support
- The patient app, in multiple languages, with pathways our physicians review
- Funding toward study costs
Staff time: ER doctors review what Aurevia shows. The discharge nurse spends about a minute handing out the QR code for the app.
Connection: standard HL7, FHIR and CDS Hooks interfaces, read-only during the silent phase. Data stays in an NYP-approved environment under a business associate agreement.
Safe, independent, and decided in advance
IRB and governance first
Weill Cornell IRB approval, and NYP AI governance and security review, before any live use.
Independent safety board
An independent data and safety monitoring board, with stopping rules set in advance.
Your research team analyzes the results
Not Aurevia, with full rights to publish. Every patient is analyzed in the group they were assigned to. Returns to other hospitals are captured through the regional health information exchange and claims.
Protocol registered before enrollment
Price and KPIs agreed on day one
If safety holds and the KPIs are met at month 12, NYP converts at that price. If the KPIs aren't met, the fee is revisited against what was measured.
Costs are shared under a research agreement
Aurevia contributes $200,000 in cash toward study costs
We also cover the Epic integration, hosting, the patient app, our engineers, team and support at our own cost.
NYP contributes staff time, data access and analysis in kind
Built into the EHR to keep costs low
Automatic enrollment, no bedside consent, and outcomes drawn from existing records.
We size the final scope together to fit the budget
For example starting at one campus or with the ER group first, and we can pursue outside grant funding together if the full design calls for more.
Most tools hold one end
vs. Epic and MyChart
Epic has MyChart. It's a record and a message box.
- The patient has to know what to look for and what to ask, and a message waits for a busy clinic to answer.
- The patients who need help most, older, sicker, less comfortable in English, tend to use it least.
- Epic is adding AI too, and that AI works from what's already in Epic.
Aurevia is different:
- Starts from the full ER visit, including the doctor's reasoning.
- Walks the patient through what comes next, one step at a time.
- What happens at home comes back into the ER doctor's decisions as each result lands.
Aurevia holds both ends and connects them.
vs. Imaging AI, Scribes, and Documentation Tools
Imaging AI works on what's already in the chart. Scribes help write the note. After-the-fact documentation tools read the chart days later. Patient apps don't reach the ER doctor's decision.
vs. Chronic Care Platforms (ChartSpan, Validic, Current Health)
These platforms aggregate patient data for longitudinal chronic care management, but their output is not surfaced at the ER. Aurevia is purpose-built for the emergency context: urgent, time-compressed, acting on incomplete information. A different buyer and a different moment.
A doctor in the family, for everyone.