Interventional Radiology Referral Intake Is a Procedure Order

Interventional radiology referral intake starts from a different place than every other specialty queue. The packet does not ask you to see a patient. It asks you to do something. A port placement. A dialysis access intervention. A biopsy. A drainage. That sounds like it should be easier to book than a consult, because the decision has already been made somewhere else. It is harder. A consult can happen with a thin packet. A procedure cannot.

Your intake sits between an office that has already told the patient what is going to happen and a procedure suite that will not run without prior imaging, recent labs, and an authorization. The referral arrives as a PDF. The imaging that justifies it lives at an imaging center you have no login for. The labs are three weeks old at another practice.

So the packet goes into a pile to be worked later, and the referring office finds out where it stands by calling you. Referrals that stall this way rarely fail loudly. They just never turn into a booked date, and nobody in either office is watching the gap.

For a practice of two or three interventionalists, the volume is never big enough to justify a full-time referral coordinator and never small enough to work out of an inbox. It gets done by whoever has an afternoon.

The referral names a procedure, not a visit

This is the difference that changes your whole intake process, and most referral automation ignores it. A packet asking for a vein ablation and a packet asking for a tunneled catheter exchange need different prerequisites, different room time, and different authorization paths. Sorting them by arrival date treats them as the same object.

athenaOne already holds the structure to tell them apart. Referral order types are a reference set, not free text, and referral sources are configured records tied to the offices that send to you. An inbound packet can be matched against both at intake: what is being asked for, and who is asking.

That gives you a queue organized by what each packet needs next. Procedure requests with complete prerequisites go to scheduling. Procedure requests missing a study go to a records-pending bucket with an outbound chase already running against them. Anything the AI cannot map to a configured order type stops and waits for a person, which is the behavior you want.

Referrals do not arrive in one queue

Ask a vascular and interventional practice where referrals land and you will get a list, not an answer. Flat PDFs over fax, labeled as referrals but not actually structured orders. A provider portal for the two or three high-volume offices that want status back. Self-referrals from patients who found you through a screening. Each one lands somewhere different, and the count is usually four or five places.

Fax is a large part of that and it is not going anywhere. A March 2026 MGMA Stat poll found nearly 1 medical practice in 4 (24%) does not have a digital fax solution fully integrated with their EHR and workflows, while 73% do. The gap between those two numbers is the whole problem: an integrated fax is a document that can be classified and attached to a chart without a person opening it first.

There is a failure mode worth naming here because it is quiet. During a fax-processing rollout at one procedural practice, the pipeline only worked documents filed under the main office department. Everything filed under any other department sat unprocessed for weeks, and nothing in the system reported it as a backlog. Sweeping every department bucket, not the default one, is the difference between a working intake queue and one that looks empty because it is blind.

The study that makes the procedure bookable is somewhere else

A referral for an intervention almost always depends on an outside study that is not in your chart. The referring office has it. Sometimes an imaging center has it. Offering a procedure date before it arrives creates a booking you will move.

Referral loops fail quietly at scale. Across more than 103,000 referral scheduling attempts in one large health system, only 34.8% resulted in a documented completed appointment, and 38.9% of attempts lacked an appointment date at all.

This is a paperwork chase, and it runs well without a person. The AI opens a follow-up task against the referral for each missing item, generates the records request through the athenaOne document pipeline, and sends it out by fax because that is what the receiving office will accept. Then it calls the referring office to ask for the study by name rather than sending a generic request and waiting.

You can apply your own lead-time rule here too. If new procedure referrals cannot book inside four business days so records and authorization have room to land, the AI holds the offer until the packet can hold the visit. That is the same discipline as prior authorization windows for IR procedures, applied earlier, before a date has been promised to anyone.

When the referral belongs to someone else

The sharpest question a practice has ever asked us about referral intake was this: if a packet arrives that is really somebody else’s case, will the system notice, or will it just tell the patient we have a referral on file?

The honest answer has a hard boundary in it. The AI does not decide what is wrong with a patient and does not rank one referral as more urgent than another. Those are clinical calls and they stay with your clinicians. What it can do is administrative and still useful: compare the referral order type on the packet against the procedure catalog your practice is actually configured to perform, and notice when there is no match.

When there is no match, the packet does not get a slot and it does not get a reassuring phone call. It routes to a named person with the mismatch flagged and the sending office attached, so somebody qualified can look at it the same day and either redirect it or keep it. The value is that it surfaces in hours instead of surfacing when the patient calls to ask why nobody has scheduled them.

One referral often means two bookings

At one vascular group roughly 9 in 10 visits are an imaging study paired with a provider visit, booked together with specific spacing between them. Nothing in the scheduling template enforces that pairing. It exists as convention, in the heads of two schedulers.

The failure mode is not dramatic. Someone books one leg and not the other, and the second half of the workflow silently does not exist until a human catches it in review. Native reminders make it worse, because they fire on the chronologically first appointment, so a patient with a morning study and a mid-morning provider visit gets reminded about one of them.

Intake is where that gets fixed, because intake is where you can still see that the referral implies two bookings rather than one. More than three medical groups in four (76%) manage referrals in their EHR (66%) or in referral management software (10%), so the system of record is already there. What is usually missing is anything that treats a referral as a sequence instead of a row. The same gap shows up in vascular follow-up tasks and the missing second visit.

Where the work hands back to a person

Judge referral automation by what it does with the packets it cannot finish, not by the ones it can. The clean ones were never the problem.

Four things should always land on a human desk, with the context already attached. A referral order type that does not map to your catalog. A packet where the referring office has refused or failed to send a study after the chase has run its course. Anything where the authorization path and the requested date cannot both be true. And any self-referral that arrives without a sending provider at all, because that one needs a decision about whether you take it.

Everything else, which is most of it, is data entry with a phone call attached. Chart built, documents filed under the right class, referral order created against the sending office, prerequisites requested, status sent back to the referring practice. Your coordinator spends the day on the four exceptions instead of the sixty routine packets, and the referring office gets an answer the next morning instead of the next week. That is the same shape as vascular surgery referral intake across five inboxes, on a queue where every row is a procedure.

Key Takeaways

Sort inbound referrals by the procedure being requested, not by arrival date, because prerequisites differ by order type.
Match packets against configured athenaOne referral order types and referral sources so the queue organizes itself at intake.
Sweep every department bucket for inbound referral documents, not the main office alone, or the backlog stays invisible.
Chase the outside study with a follow-up task and a generated records request before any procedure date is offered.
Treat a referral that implies a study plus a provider visit as two linked bookings at intake, not one row.
Route order types your practice does not perform to a named person with the mismatch flagged, and never let the AI make that call clinically.

Interventional radiology intake is not hard because the volume is high. It is hard because every packet is a commitment to do something, and the things that make it doable live at other offices. Sort the queue by what is being asked for, sweep every inbox and every department bucket, chase the missing study automatically, book the study and the visit together, and hand your staff the four packets a day that genuinely need a person. That is a coordinator’s week back, and a referring office that stops calling to ask where things stand.

Sources