TL;DR: You automate filing echo, Holter, and cath reports by inventorying every document type that hits your fax line, defining chart-match rules on name plus date of birth plus MRN, setting a confidence threshold that sends uncertain documents to a human exception queue, and mapping each document type to a specific destination in the EHR. A cardiology fax triage with OCR extraction tool does the reading, classifying, and matching. Your configuration decides where each document lands and who reviews the leftovers.
What a cardiology fax triage with OCR extraction tool actually has to handle
A general internal medicine practice receives roughly three things by fax: records requests, referral paperwork, and lab results. A cardiology practice receives at least eight, and several of them look nearly identical on a 200-dpi page.
In one month, a single post-MI patient can generate a hospital discharge summary, a cath and PCI report from the cath lab, a follow-up transthoracic echo read by an outside hospital's service, a 14-day event monitor summary from a monitoring vendor, and a remote pacemaker interrogation. Five documents, five senders, five different destinations in the chart. One of them — the abnormal echo — needs a physician to see it today.
The volume isn't going away. Around 56% of referrals still move by fax, and health care's fax problem persists because the sending hospital's EHR and yours don't exchange data directly. Even after a decade of federal push, ONC's interoperability data shows a meaningful share of hospitals still aren't exchanging records routinely. Fax is the compatibility layer nobody chose.
Meanwhile the administrative load keeps climbing. The MGMA 2026 regulatory burden report found roughly 95% of medical groups saw burden increase over the past three years, with many calling current demands unsustainable. Filing faxes is a small slice of that, but it's a slice you can actually automate this quarter.
Step one: inventory your document types before you touch a single setting
Every configuration decision downstream depends on knowing what actually arrives. Pull two weeks of inbound faxes — a full 10 business days, not a convenient week — and sort them into piles by type. Count each pile.
A typical cardiology inbound mix breaks down like this:
- Echocardiogram reports — TTE, TEE, stress echo, usually from a hospital reading service or outside imaging center.
- Stress test results — treadmill, nuclear, stress echo, often bundled with the echo report from the same sender.
- Holter and event monitor summaries — from monitoring vendors, arriving 3 to 30 days after the device was placed.
- Cath and PCI reports — from the hospital cath lab, frequently paired with a procedure note and angiography summary.
- Device interrogation reports — pacemaker, ICD, and loop recorder checks from remote monitoring services. Often the highest-volume recurring type in the mix.
- Hospital discharge summaries — the document your care team needs fastest and receives latest.
- Referrals and consult requests — from primary care, with or without supporting records.
- Prior authorization determinations — approvals, denials, and requests for additional clinical information.
For each type, write down three things: how many arrive per week, which senders produce them, and where the document needs to end up. That third column is the actual specification you hand to whoever configures the tool.
Two rules make the inventory usable. If a type is under about 2% of volume, leave it out of automation for the first 90 days and let it land in the manual queue — the tuning effort isn't worth it. And split types by sender when the layouts differ. An echo report from your main hospital partner and an echo report from a mobile imaging vendor may need to be two separate classes, because the OCR extraction is reading two very different page structures.
Skipping this step is the most common way these projects go sideways. Teams configure for the documents they remember, then discover in week three that 18% of inbound volume is device interrogations nobody accounted for.
How does chart matching work, and why does name plus DOB plus MRN beat any single identifier?
Chart matching is the step where the tool decides which patient a document belongs to. It reads identifiers off the page, compares them against your patient index, and returns a match with a confidence score.
Single-identifier matching fails in predictable ways, and cardiology hits every one:
- MRN alone. Fax transmission at 200 dpi turns 8s into 3s and 1s into 7s. A single-digit OCR misread produces a confident match to the wrong chart, which is worse than no match at all.
- Name alone. Cardiology panels skew older, which means more suffixes, more maiden-versus-married mismatches, and more shared surnames inside families who all see the same practice.
- Date of birth alone. Not usable by itself in any panel over a few thousand patients.
The reliable approach is composite matching with tiers. Require at least two of three identifiers — name, date of birth, MRN — to agree. Treat three-of-three as high confidence, two-of-three as medium, and anything less as an exception. Then add secondary identifiers where you have them: the ordering provider's name, the accession number on an imaging report, or the device serial number on an interrogation report all narrow a candidate list quickly.
Matching is only half the job. The document still has to be written into the chart, and that's where a lot of fax tools stop and hand the work back to your staff. A parsed blob sitting in a vendor portal doesn't help the medical assistant who still has to open the EHR and re-key it. Honey Health's Fax Triage agent files into your existing EHR — chart match, document attachment, and structured field write-back all happen in the system your team already works in, not in a parallel interface they have to remember to check.
Set one hard rule before go-live: no document files automatically on a name-only match. Ever. The 20 seconds a staffer spends confirming a borderline match is cheaper than any conversation that opens with "the echo went to the wrong chart."
Media tab, discrete result, or provider task: deciding where each document lands
Filing a document isn't one action. There are three destinations, and most documents need two of them.
The media tab or scanned documents section holds the source PDF. Every inbound document goes here — it's the record of what the sender actually transmitted. Filing here alone is right for narrative documents nobody will trend: discharge summaries, cath and PCI procedure notes, consult letters.
A discrete result is a structured value written into a field the EHR can chart, trend, and report on. This is where cardiology gets the most out of OCR extraction, because several of your highest-volume documents contain exactly two or three numbers anyone cares about:
- Echo reports: LVEF, valve gradients, chamber dimensions
- Holter and event monitors: AF burden, longest pause, PVC percentage
- Device interrogations: battery status, lead impedance, arrhythmia episode counts
- Stress tests: METs achieved, ischemic changes, TID ratio
Pulling those into discrete fields means your quality reporting, device clinic follow-up lists, and heart failure registry stop depending on someone opening a PDF and reading it.
A provider task or in-basket message is the third destination, and the one teams routinely under-configure. A document that files perfectly into the chart but sits unseen for six days hasn't been handled. Route to a task when the document demands a decision: abnormal findings flagged by the reading physician, prior auth denials, discharge summaries for patients owed a 7-day follow-up call, device alerts.
Write the routing matrix down before configuration — one row per document type, columns for media tab, discrete fields, and task recipient. Then hand it to whoever is configuring. Verbal descriptions of routing rules do not survive contact with a configuration screen.
Setting the confidence threshold and staffing the exception queue
Two thresholds matter, and they get conflated constantly. Classification confidence is how sure the tool is that a document is an echo report. Chart-match confidence is how sure it is that the document belongs to Maria Delgado, DOB 3/14/1948. Set them separately.
Start conservative. For the first 30 days, hold both thresholds high — a 95% floor on chart matching is a sensible starting point — and accept a high exception rate. You're not optimizing for automation percentage yet. You're building a sample of auto-filed documents you can audit. Pull 100 at day 30 and check every one against the source fax.
If that audit comes back clean, drop the chart-match threshold to roughly 90% and re-audit at day 60. Most cardiology groups settle between 88% and 92%, with 70 to 85% of documents auto-filing. Any vendor promising 99% auto-file with no exceptions is either not counting the errors or auto-filing documents that should have been escalated.
The exception queue needs an owner, not a committee. Assign one person with a named backup and give them two scheduled blocks a day, 30 to 45 minutes each, mid-morning and late afternoon. Set a service level: nothing sits past the next business day, and anything classified as a discharge summary or an abnormal result gets worked within two hours.
Track exception reasons, not just exception counts. A queue running at 20% is fine if the reasons are spread across genuinely ambiguous documents. A queue running at 20% where 14 points trace to one monitoring vendor's cover page is a configuration problem you can fix in an afternoon. Most of the tuning work in the first quarter is reading exception reasons and adjusting one sender's template at a time.
How do you measure whether automated filing is actually working?
Take the baseline before go-live. Almost everyone skips this, and without it you're left arguing about whether things feel better.
Measure four things across a normal two-week stretch while filing is still manual:
- Filing accuracy. Pull 100 recently filed documents at random and check each — right patient, right document type, right location in the chart. Manual accuracy in most practices runs in the mid-90s, not 100%. Knowing your real number keeps the comparison honest.
- Turnaround time. Median and 90th-percentile hours from fax receipt to the document being visible in the chart. Track discharge summaries separately; they're the clinically urgent tail.
- Staff hours. Person-hours per week spent classifying, indexing, and filing inbound faxes. Time it, don't estimate it.
- Rework rate. How often a filed document has to be moved, re-indexed, or corrected.
After go-live, track the same four plus two more: auto-file rate (documents filed with no human touch) and exception aging (share of exception items older than one business day).
Reasonable 90-day targets for a mid-sized cardiology group: median turnaround under two hours, auto-file rate above 70%, filing accuracy at or above your manual baseline, and staff hours cut roughly in half. If accuracy drops below baseline, you tuned thresholds too aggressively — raise them and take the lower auto-file rate. Speed that costs accuracy isn't a win in a chart.
Frequently Asked Questions
Can OCR read a low-quality fax of an echo report accurately?
Usually yes for the narrative text, less reliably for handwritten annotations and small-font measurement tables. Healthcare-tuned OCR handles fax artifacts better than generic document OCR trained on invoices. The practical safeguard is the confidence threshold — low-quality pages score lower and route to the exception queue rather than filing on a guess.
What happens to documents the tool can't classify?
They go to an unclassified bucket in the exception queue with the page image attached, so a staffer can identify and file them manually. Nothing should ever be discarded or filed to a generic folder. Review unclassified volume weekly — a recurring unclassified type is usually a sender you missed during the inventory step.
Do we need outside hospitals to change how they send us reports?
No. That's the point of automating on the fax side. Asking a hospital cath lab or a monitoring vendor to change their output format is a multi-month negotiation with a low success rate. Configure your triage tool around the documents they already send, one sender template at a time.
How long does configuration take for a cardiology practice?
Plan on two weeks for the document inventory and routing matrix, one to two weeks for configuration and sender-template setup, then a 30-day conservative run before you loosen thresholds. Practices that rush the inventory usually spend the time back in month two chasing exception reasons.
Should Holter and device data be discrete results or scanned documents?
Both. File the source PDF to the media tab as the record of what was sent, and extract the two or three values you actually trend — AF burden, longest pause, battery status, lead impedance — into discrete fields. Document-only filing makes device clinic follow-up and quality reporting a manual chart-review exercise.
What's the biggest risk in automating fax filing?
Filing a document to the wrong patient's chart with no human ever looking at it. That risk is controlled by composite chart matching, a conservative confidence threshold, and a staffed exception queue — not by vendor accuracy claims. Audit a random sample of auto-filed documents monthly for the first quarter.
