VAREQUE
VAREQUE
Every referral that arrives by fax is retyped by someone before it can be worked — patient name, date of birth, MRN, referring physician. V-FaX Gold reads the page, extracts those fields, and files the ones it can stand behind. The rest go to a reviewer with the reason attached.
Pilot stage. Not generally available. We sign a BAA before the first page moves.
The fax is rasterised and read with an OCR engine tuned on degraded pages — skewed, speckled, light-on-grey documents of the kind a fax machine actually produces.
Where OCR returns too little text, the page is sent once to a vision model that reads the image itself. It is asked for that sector’s fields only, and the call is cost-capped per document.
Medical: patient name, date of birth, medical record number, referring physician.
Legal: case title, county, docket number.
A document is filed automatically only when every condition below holds. Otherwise it lands in a review queue with the page image, the extracted fields and the specific reason it stopped.
This is the part we would want to read if we were buying it. A confidence score cannot tell a correct reading from a wrong one — we proved that to ourselves on a document where the OCR and the model agreed, confidently, on the wrong patient name. So the system stops at specific points, deliberately.
A misread name looks exactly like a new patient. Auto-filing requires the patient to already exist on your account — by name or by MRN. First-time patients go to a person, every time.
A subpoena legitimately arrives with a docket number nobody has seen, and a mistyped one is indistinguishable from it. Court documents are reviewed by a person, full stop.
Two different dates of birth on one page, or a page carrying another sector’s vocabulary, stops the document. Every decision is recorded with its reason — the log is what an audit reads.
These are our own numbers, from our own test documents — not a customer’s traffic. They are here so you can judge the starting point, and so the pilot has something to be measured against.
| Test set | Field accuracy | What it is |
|---|---|---|
| Clean scans | 100% | Text-layer documents, the easy case |
| Realistic faxes | 93.9% | Skew, speckle, contrast loss — readable with effort |
| Court documents | 89.3% | Subpoenas, notices, affidavits |
| Near-unreadable | 51.7% | The edge of legibility — and what we say about it |
Two honesty notes. The 51.7% is not a failure of the pipeline: on those pages nothing can be read reliably, and the system routes them to a person rather than guessing. And our auto-filing rate on the test set is low because the test set invents new patients on every page — a real practice sees its own patients repeatedly.
We have measured this on documents we made ourselves. The numbers that decide whether it works for you — how often it can file without a human, and how long a review really takes — only exist in your traffic. That is what the pilot is for.
We sign a Business Associate Agreement before any document moves, and the vision model we use runs on a provider that can sign one too. We do not claim a certification, because there is no such certification — a BAA is the instrument, and ours is available for review.
It goes to the review queue, not into the record. The reviewer sees the original page with the extracted values highlighted on it, and can correct or add any field. Nothing is filed on a guess.
No. It removes the retyping. A person still decides on anything uncertain, and on everything legal. The claim we are willing to make is narrower than “no human touches your faxes” — it is “no fax reaches your chart without either a human or a patient we already know”.
Yes. Send de-identified pages and we will run them and show you what was extracted, what was refused, and why — before anything is connected to your fax line.
Tell us the volume and the document types. We will tell you honestly whether the pilot is likely to work for you — including if the answer is no.
|
Terms and Conditions |
Privacy Policy |
V-DoX™ 2002-2026 © Vareque
|