Automating dermatopathology report routing means the software reads each inbound path report, pulls the patient identifiers and accession number off it, matches it to the right chart, and files it to the ordering provider's results queue with a priority flag on malignant findings. The manual version of that — open, read, search, match, upload, notify — takes several minutes per report and stalls whenever the person doing it is out. The automated version takes seconds, and your staff's job shifts to reviewing the small share the system isn't confident about.
What automated path routing actually replaces
Watch the current process end to end and the target becomes obvious.
A report arrives from one of your labs. Someone opens it, reads far enough to confirm it's pathology, notes the patient name and date of birth, searches your EHR, resolves any ambiguity between similar names, opens the chart, selects the right document category, uploads the PDF, sets the date of service, and then flags the ordering dermatologist. If the finding is malignant, they also have to notice that and escalate it — which depends on a person reading a report they weren't trained to interpret.
That last part is the real risk. Everything upstream of it is tedious; that step is where a delayed melanoma diagnosis lives.
Automation targets the whole chain. The system extracts identifiers, matches the chart with a confidence score, files the document under the right category and date of service, creates the results task for the ordering provider, and applies an urgency flag based on the diagnosis text. Reports it can't confidently handle go to a person with the best guess already attached.
The point isn't speed for its own sake. The College of American Pathologists standard is that 90% of routine cases sign out within two business days. If your lab hits that and your office then takes another day to get the result in front of the physician, your patients' experience of turnaround time is three days, and one of those days is entirely yours to fix.
Step 1: Inventory every inbound path source
You can't build routing rules against a document mix you haven't measured.
Pull two consecutive weeks of inbound pathology and count four things.
- Volume by lab. Most derm groups send to two or three labs and receive occasional reports from outside dermatopathologists or hospital labs their patients were seen at. The concentration matters — a classifier gets good fast on layouts it sees daily.
- Report layout per lab. Grab a sample from each. Where does the accession number sit? Is the diagnosis in a structured field or a free-text paragraph? Does the report include the original requisition?
- Time from receipt to chart. Have whoever files them time themselves honestly. The number is always higher than the practice's estimate.
- How many arrive by fax versus a lab portal versus an interface. Practices frequently discover a mix, with one lab sending through an HL7 feed and the rest faxing.
That inventory tells you what share of volume automation can realistically clear in the first month, and it gives you the baseline you'll measure against later.
Step 2: Define the routing rules and the urgency flag
Routing rules answer two questions per report: which chart, and who needs to know.
The chart side is mechanical. The notification side is where the design decisions live, and they should be made deliberately rather than left to a vendor default.
Build a short table before you configure anything:
| Report outcome | Routes to | Timing |
|---|---|---|
| Benign, routine | Ordering provider results queue | Standard |
| Atypical / requires re-excision | Ordering provider, flagged | Same day |
| Malignant — BCC/SCC | Ordering provider, flagged + surgical scheduling notified | Same day |
| Melanoma or high-grade | Named person paged, not queued | Immediate |
| Insufficient specimen / needs repeat | Ordering provider + scheduling | Same day |
Notice that the last two rows aren't really routing rules — they're escalation paths, and they need a named human on the receiving end. A queue that gets checked at four o'clock is not an escalation path.
Keep the document taxonomy short. Twelve to twenty categories is workable for a derm practice; forty produces both human misfiling and low machine confidence at the same time.
One more configuration decision belongs here: index by date of service, not date received. It's free to set correctly now and expensive to fix retroactively across thousands of documents.
Step 3: Set the matching threshold and build the exception path
Patient matching should always run at a stricter confidence bar than document classification. Misclassifying a report as a consult note is annoying. Filing it in the wrong patient's chart is a documentation incident.
Open the threshold high, so only near-certain matches file automatically. Expect a large review queue in week one — that's the design working, not failing. Audit that queue for a defined period and track how often the system's suggestion was correct even when it fell below the threshold. Then lower the bar in steps, watching the error rate after each move.
Three configurations that matter more than they sound:
- Duplicate detection. Labs re-send. Without a dedupe rule, automation will file the same report three times faster than a person ever could.
- Accession-number reconciliation. If the system can match an inbound report back to the original order, you get an unmatched-orders report — the list of biopsies you sent that never came back. Practices routinely find things on that list.
- Unknown-patient handling. A report for someone with no chart should route to a human by design, never create a chart automatically.
Honey Health's fax triage agent handles this read-extract-match-route sequence inside the EHR the practice already runs, with the low-confidence cases surfacing as an exception queue rather than a second system your staff has to work alongside the chart.
Who signs off before a malignant result reaches the patient?
Automating the routing does not mean automating the clinical loop, and being clear about that distinction protects both patients and the project's credibility with your physicians.
The software's job is to put the right report in front of the right physician faster, with the urgency correctly flagged. The physician's job — reviewing the finding, deciding on management, and communicating with the patient — doesn't change.
Decide explicitly who confirms that flagged results were acted on. In most practices this is a clinical staff member with a daily worklist of flagged results, checking that each one has a documented physician action. That check takes a few minutes a day and is the single highest-value human step in the whole workflow.
Also decide what happens when the physician is out. A flagged melanoma result that routes to a dermatologist on vacation is a routing rule that needs a coverage fallback, and this is exactly the kind of gap that only shows up if you plan for it.
Step 4: Instrument turnaround time so you can prove it worked
Measure three intervals from day one, ideally starting before you turn anything on.
- Lab sign-out to receipt at your practice. This is the lab's performance plus transmission. You'll use it to hold labs accountable and to separate their delays from yours.
- Receipt to filed in chart. The interval automation attacks directly. This should collapse to near-zero for auto-filed reports.
- Filed to physician acknowledgment. The clinical loop. Automation shortens it by getting the report in front of the physician sooner, but the remaining time belongs to workflow, not software.
Report those three weekly during rollout and monthly after. Also track the review queue as a share of total path volume. A healthy deployment shows that share falling as the system learns your labs' layouts. If it's flat after four to six weeks, something's wrong with the taxonomy, the integration, or the source document quality, and that's a month-two conversation with your vendor rather than a renewal-time one.
What should stay human
Set expectations with staff up front, so exceptions read as designed rather than broken.
- Degraded scans. A report that was printed, faxed, scanned, and faxed again loses enough fidelity that no OCR recovers it cleanly. A permanent minority of volume.
- Handwritten pathologist comments. A margin note about clinical correlation is meaningful and hard to capture as structured data. The document files; the note may not.
- Ambiguous name matches. Two patients, similar names, close dates of birth. The correct system behavior is to stop and ask.
- Anything with an open clinical question. A report recommending correlation with clinical findings needs a physician, not a routing rule.
- First-time patients. No chart, nothing to match against, human by design.
None of these argue against automating. They argue for automating the 85-plus percent that's mechanical and putting your experienced staff on the part that actually requires judgment.
Frequently Asked Questions
How long does it take to automate dermatopathology routing?
Most implementations run several weeks from kickoff to cutover, with the routing-rule design and the shadow-comparison period consuming more time than the technical integration. Practices receiving from a small, stable set of labs move faster. Accuracy typically keeps improving for a month or two after go-live as the system learns each lab's report layout.
Do we need an HL7 interface with our pathology lab?
It helps but isn't required. A direct interface delivers structured data and is the cleanest path when the lab supports it and the cost is reasonable. Fax-based automation reaches most of the same outcome by reading the report image, which is why practices with a mix of interfaced and faxing labs can automate both streams through one workflow.
Can the software tell whether a result is malignant?
Modern systems extract the diagnosis text and apply configurable rules to flag urgency, which is meaningfully different from interpreting the finding. Treat the flag as a routing signal that gets the report to the physician faster, not as a clinical read. Build the escalation path and test it in shadow mode before you rely on it.
What happens to reports the system can't match to a patient?
They route to a human review queue with the extracted identifiers and the closest chart candidates already attached, so the reviewer confirms or corrects rather than starting from scratch. That queue shrinks over the first several weeks as the system learns your recurring senders. It never reaches zero, and it shouldn't.
Will this let us reduce front-office headcount?
Most practices reinvest the recovered hours rather than cutting positions. The staff who were filing documents move to work that's harder to automate and more directly tied to revenue — scheduling, prior auth follow-up, patient callbacks. Frame the business case on recovered capacity, which is both more honest and easier to defend internally than a headcount claim.
