Referral Orders: Creating the Order, Not Just the Message

Most ENT practices receive referrals as documents. A fax with a scribbled reason, an email from a primary care office, a portal submission, a phone call from a coordinator. Each one is a message about a referral order rather than the order itself, and the difference decides whether anyone can tell you what happened to it three weeks later.

A message has no status. It arrives, someone reads it, and from that point forward the only record of what happened is a person’s memory, a note, or a scheduled appointment that may or may not connect back to why the patient was sent.

So referral tracking gets rebuilt by hand at every practice that tries. A spreadsheet, a shared inbox with folders, a coordinator who knows which primary care offices need chasing. It works until volume rises or the coordinator takes a week off, and then the practice discovers it cannot answer a basic question: of the referrals we received last month, which ones turned into a patient in a chair?

The referral loop is measured, and the measurements are not flattering

This is a documented problem with a national quality measure attached to it, which is worth knowing before treating it as a local process failure.

CMS maintains an electronic clinical quality measure, Closing the Referral Loop: Receipt of Specialist Report, defined as the percentage of patients with referrals for which the referring clinician receives a report from the clinician the patient was referred to. The measure exists because breakdowns in referral communication lead to delays and worse outcomes.

The rationale published with that measure cites a 2018 analysis of primary care referrals to specialists in which, of 103,737 referral scheduling attempts, only 36,072 resulted in documented complete appointments. That study attributes much of the gap to low documented appointment scheduling rates rather than to patients refusing care.

Read that from the receiving end and it is a demand problem, not just a safety problem. A meaningful share of referrals sent toward specialty practices never becomes a documented, completed visit, and the specialty practice usually never learns which ones.

An order has a status, a document does not

The practical distinction is that a referral order is a structured record in the chart with a type, a recipient, and a lifecycle, while a faxed page is an image with a date.

athenaOne carries a reference list of referral order types, readable through GET /reference/order/referral, and referral orders are created against an encounter through POST /chart/encounter/{encounterid}/orders/referral. That structure is what a tracking process needs. A record that can be queried, aged, and reported on beats a folder of documents that can only be counted.

For an incoming ENT referral, the work is matching the message to the structure. Which referral order type does this correspond to, which provider or provider group can take it, what has to be true before it can be scheduled, and what is missing right now. Every one of those questions is administrative, and every one of them is being answered by a human today by reading a fax.

Once the referral is a record rather than a document, the questions the practice could never answer become queries. How many referrals from each source are still unscheduled. How old the oldest one is. Which sources send referrals that consistently arrive without the information needed to book them.

The complication: the order has to be created inside the encounter

This is the constraint that surprises people who assume any of this can be fully automated end to end, and it is a good constraint.

Building a state-mandated pediatric health form for a multi-site group, the core technical limitation turned out to be that you have to be inside the encounter in the EHR to generate the document at all. Referral orders carry the same shape. The order is created against an encounter, by the clinician conducting it, as part of the clinical record.

That is exactly where the line belongs. Deciding that a patient needs a specialist, and which one, is a call only a clinician makes, inside a visit. The fact that the system requires an encounter enforces something that would otherwise depend on discipline.

So the automation works on both sides of that moment rather than through it. Before: assembling what the encounter will need, confirming which referral order types the practice actually uses, and making sure the patient’s coverage and prior records are in place so the order is not written against missing information. After: taking the created order and running everything administrative that follows, which is where the loop is actually lost.

The handoff is the order itself. When an incoming referral does not map cleanly to a type the practice uses, or when the reason for referral is ambiguous enough that the appointment type is genuinely unclear, the automation stops and routes it to a coordinator with a specific question rather than booking the closest-looking slot. An ENT practice booking a complex airway evaluation into a fifteen-minute follow-up because the fax said ‘ENT eval’ has not saved anyone time.

What the front office can actually run after the order exists

The administrative half of a referral is larger than the clinical half, and it is almost entirely chase work.

The outbound side is contacting the patient to schedule, and doing it more than once. AHRQ’s patient safety network summarizes an expert-panel review of the ambulatory referral process that lays out its steps from the clinician ordering the referral through communication of the plan back to patients and families, and recommends clear standards of accountability for patient follow-up. Accountability for follow-up is a staffing question in most practices, which is another way of saying nobody owns it after hours.

The inbound side is the paperwork the visit requires: prior records from the referring office, imaging, audiology results, authorization where the plan requires one. A referral that arrives without those and gets booked anyway produces a visit where the clinician cannot do anything useful, which is worse than a delayed visit because it consumes a slot.

The third piece is the report going back. The measure CMS defines is about the referring clinician receiving a report, and a specialty practice that reliably closes that loop is a practice referring offices keep sending to. That is a growth argument, not just a compliance one.

All three are administrative, repetitive, and time-bound, which makes them the right work to automate. None of them require deciding anything about the patient’s care.

Measure the referral funnel by source, not in aggregate

An aggregate referral conversion rate is nearly useless, because it averages together sources that behave completely differently.

Split it by referring office. One practice sends complete referrals that book in two days. Another sends a fax with a name and a phone number, and every one of those takes four calls. Those two look identical in a monthly total and require opposite responses: the first needs nothing, the second needs a conversation about what to include.

Then track time from receipt to scheduled, and the share still unscheduled at seven, fourteen, and thirty days. Aging is where referral leakage actually shows up, and it shows up long before the patient is formally lost. A referral sitting at day twenty-one has usually already been solved somewhere else.

The last number is the honest one: referrals received that became a completed visit. Practices avoid it because it is lower than they expect, which is exactly why it belongs on the report. The national picture behind the CMS measure suggests the gap is large and mostly structural rather than a reflection of any single practice’s effort.

Key Takeaways

Convert incoming referrals into structured records rather than filed documents. A record has a status you can query and age; a fax has a date.
Accept that the referral order is created by a clinician inside an encounter, and design the automation to work before and after that moment rather than through it.
Do the matching work up front: referral order type, eligible provider, what has to be true before booking, and what is missing right now.
Escalate ambiguous referrals to a coordinator with a specific question. Booking a complex evaluation into a short follow-up slot because the fax was vague costs more than the delay.
Chase the paperwork the visit needs, not just the appointment. A booked referral with no prior records consumes a slot and produces nothing.
Close the loop back to the referring office. The report going back is what keeps referral sources sending, and it is what the CMS measure is built around.
Break referral conversion out by referring office. An aggregate rate averages together sources that need completely different responses.
Watch referrals aging at seven, fourteen, and thirty days. Leakage shows up in the aging report well before anyone records the patient as lost.

An ENT practice that treats referrals as documents will always be reconstructing its own funnel by hand, and will always be one absent coordinator away from losing track of it. Treating the referral as a record with a status changes what the practice can see and what it can automate, while leaving that call exactly where it belongs, with the clinician inside the encounter. An AI team working referral intake inside athenaOne can match incoming referrals to the practice’s own order types, chase the records and authorizations the visit needs, work the scheduling until the patient is booked, and hand a coordinator the ones that need a human to decide.

Sources