Symptom Screening After Hours Without Crossing the Line

A large share of after-hours calls to an urgent care group are not scheduling calls. They are symptom screening calls: someone describing what is happening and asking, in effect, where they should go tonight. The practice already has an answer to that question, written down, signed off by a clinician, and sitting in a binder or a protocol subscription. The protocol was never the problem. The problem is that at 9pm there is nobody available to walk through it.

What most groups buy instead is a relay. An answering service takes a name and a number, writes down a sentence or two of what the caller said, and either pages the on-call clinician or leaves the message for the morning. The clinician gets paged for calls that a scheduled appointment would have solved, and the calls that needed a clinician arrive as a paraphrase written by someone who was not following the protocol.

Both failure directions cost the practice. A caller with a question that the practice could have answered inside its own hours ends up at an emergency department, and that visit does not come back as revenue or as continuity. A caller who could have waited until 8am gets a page sent to a clinician who is asleep. Neither one is a screening failure. It is a coverage failure, and the practice pays for it every night.

The question worth asking is narrower than it looks. Not whether AI should screen patients, which is the wrong question, but whether the questions the practice already wrote can be asked reliably at 9pm and the answers put in front of the right person fast enough to matter.

The protocol belongs to the practice, and it stays that way

This is the distinction that decides whether an after-hours program is defensible or not.

The protocol set is clinical policy. A clinician selected it, a clinician configured the routing rules inside it, and a clinician owns any change to it. That does not move. What moves is who is available to run it at 9pm on a Sunday.

The AI’s job is narrow and worth stating in exactly these terms: ask the questions in the order the protocol specifies, record the answers in the patient’s own words, and apply the routing the practice already wrote against those answers. It does not compose a new question, it does not soften one, and it does not interpret an answer that the protocol did not anticipate. Anything outside the written path goes to a person, immediately and with the transcript attached.

That boundary is also what makes the program auditable. Every call has a protocol path, a recorded set of answers, and a routing outcome that can be traced back to the rule that produced it. An answering service message that reads patient says she has been feeling bad since dinner cannot be audited against anything.

One configuration detail matters more than it sounds like it should: the immediate-escalation phrases have to interrupt the flow rather than wait for the end of it. If a caller opens with something the practice has flagged for instant connection, the correct behavior is to stop asking questions and connect, in the first ten seconds, not to finish the script first.

Three exits, decided before go-live

Every screening call ends in one of three places, and the practice has to map protocol outcomes to those three before anything is turned on.

The first is handled during the call. The caller gets the practice’s own pre-visit instructions, gets booked into the right appointment type for tomorrow, or gets told what the practice’s posted policy says about walk-in hours. Nothing here is improvised; each is a piece of text or a scheduling rule the practice supplied.

The second is a case created and routed for the morning. It carries urgency, the department bucket that owns it, the recorded answers, and a callback number. This is the exit that absorbs most volume, and it is the one that decides whether the morning starts with a clean worklist or an inbox.

The third is live escalation now, to the on-call clinician, with the recorded answers already delivered so the clinician is not starting the conversation from zero.

The mapping between protocol outcome and exit is a clinical policy decision, not a configuration preference, and practices that treat it as configuration end up rewriting it in week three. The useful version of that conversation is per department and per hour block, because the same answer set at 7pm on a Tuesday and 3am on a Saturday often has a different correct exit.

The complications that arrive in the first week

Screening flows break on the calls that are not about symptoms at all, and on the callers who are not the patient.

The caller is frequently a parent, an adult child, or a caregiver speaking for someone else. The chart being screened against and the person on the phone are different people, and the callback number belongs to the caller. Getting that relationship recorded correctly at the start is what makes the morning case actionable.

Identity is the next one. A caller who does not have the patient’s date of birth handy is common at night, and matching to the wrong chart is worse than matching to none. Duplicate charts make it worse. The workable rule is to capture the callback number first, before any screening question, because calls drop and the number is the only thing that makes the call recoverable.

Then there is the mixed call. Someone starts with a symptom question and finishes by asking to move next week’s appointment. Both halves are real work, and a flow that can only do one of them hands the patient back to a queue that is closed.

Language belongs in this list too. A screening protocol asked in English to a caller who is more comfortable in Spanish produces answers nobody should route on. Running the protocol in the caller’s language, with the recorded answers stored in both, is a requirement rather than a feature for most urgent care panels.

What lands in athenaOne when the call ends

The value of the call is what survives it, and that is entirely an integration question.

A completed screening call should leave four things behind. A case in the correct department bucket with urgency set per the protocol outcome. The recorded answers appended as a note in the patient’s own words. An appointment booked with the correct appointment type, when that was the exit. And a timestamped escalation record that says whether the on-call clinician was reached, not merely paged.

That last field is the one practices underestimate. Paged and reached are different events, and only one of them means the patient got a clinician. When they are recorded separately, the gap between them becomes visible and fixable.

The morning artifact is what staff actually touch. Not a call log, but a worklist ordered by urgency, with the calls that already resolved themselves filtered out of it. A front desk arriving at 7:45am to nineteen recordings will listen to four of them.

The numbers that tell you it is working

Start with escalation rate, split by hour block. A program that escalates most of its calls has not reduced anything, and a program that escalates almost none of them deserves a review of its routing rules by the clinician who owns them.

Then time from call start to a live clinician for the calls that escalated. This is the only number that speaks to the risk side of the program, and it is the one to report first when anyone senior asks whether the change was safe.

After that, the access numbers: share of after-hours calls that produced a booked appointment, share that produced a morning case rather than a voicemail, and abandonment. Abandonment before screening even begins is usually a hold-time artifact and gets fixed by capacity rather than protocol.

The honest check is the sample review. Pull a set of recorded calls a week and have the clinician who owns the protocol confirm the path matched the answers. That review is cheap, it is the thing that keeps the program aligned with policy over months, and it is the artifact to have ready when someone asks who is accountable for after-hours screening.

Key Takeaways

The screening protocol stays clinical policy owned by a clinician. The AI asks the written questions, records answers verbatim, and applies the routing the practice already wrote.
Map every protocol outcome to one of three exits before go-live: handled on the call, case routed for the morning, or live escalation now.
Immediate-escalation phrases have to interrupt the flow in the first ten seconds rather than wait for the script to finish.
Capture the callback number before the first screening question. Calls drop, and the number is what makes a dropped call recoverable.
Record the caller’s relationship to the patient. After-hours callers are frequently parents, adult children, and caregivers speaking for someone else.
Log paged and reached as separate events for escalations. Only one of them means a clinician actually spoke to the patient.
Report time to a live clinician first, escalation rate by hour block second, and run a weekly sample review with the clinician who owns the protocol.

After-hours coverage is not a question of whether software should screen patients. It should not, and the version that works never tries to. It is a question of whether the questions your clinicians already wrote can be asked consistently at 9pm, recorded in the chart, and put in front of the right person inside minutes. An AI team running your protocol inside athenaOne does the asking and the routing and the writing back, and hands every call it was not built to hold to a person with the answers already attached.

Sources