Closing the Sleep Medicine Follow Up Loop After the Study

Sleep medicine follow up fails in a specific place, and every practice administrator can name it. The patient completes the home sleep test. The result comes back. Then there is a gap, sometimes two weeks, sometimes two months, before anyone books what comes next. By the time the office calls, the patient has moved on, changed insurance, or decided the whole thing was more trouble than it was worth. The study got done and billed. The episode never finished.

Sleep is a sequence, not an appointment. Home sleep test or in lab study, then a results visit, then a titration or a device setup, then a follow up that has to land inside a specific window. Four bookings, three of which depend on something that happened in the one before it.

Nothing in a scheduling template enforces that. The template holds slots. It does not know that a patient who completed a study on the eleventh needs a follow up booked before a date the payer picked, and it has no way to tell you the second leg never got booked at all.

Meanwhile the phone line that would handle the sequencing is buried. Device and supply reorder calls, mask fit complaints, and replacement requests land on the same number as new patient requests, and they are the easiest calls to answer, so they get answered first. The hard, revenue bearing work of booking the next stage waits.

The second leg is the one that silently does not happen

Multi stage workflows fail the same way everywhere: one leg gets booked, the other does not, and nothing surfaces the gap until a human notices during a chart review weeks later. An analysis of primary care referrals to specialists in a large health system found the documented rate of closing the referral loop was under 35%. Sleep is more exposed than most, because the sequence has three or four handoffs instead of one.

The reason is structural. A patient calls to schedule a study, and that call has a clear job. The next call, the one that books the results visit and the titration, has no trigger. It depends on somebody watching a queue.

An AI layer treats the sequence as the unit of work rather than the appointment. When the study is complete, it calls the patient to book the next stage, offers the appointment type the practice actually wants used, and confirms it landed. When the patient does not answer, it tries again on a schedule instead of leaving a voicemail and closing the task.

What it does not do is decide what the patient needs next. Which study, which titration, and whether a device is appropriate are decided by a clinician and written into the order. The automation books what the order says and reports what it could not book.

The coverage window is an administrative deadline with a date on it

Device coverage is not open ended, and it is not a single event. Medicare pays a supplier to rent a CPAP machine for 13 months of continuous use before the patient owns it, and after the Part B deductible the patient pays 20% of the approved amount for the rental and the supplies. Continued coverage past the initial period also depends on a documented re evaluation by the treating practitioner inside a defined early window.

Read that as an operations problem and what the practice has to do gets obvious. There is a visit that must be scheduled and completed before a date, and if it does not happen the coverage argument collapses. There is also a patient who will call about a bill, repeatedly, for thirteen months.

The practice controls exactly one half of this. Whether a patient uses the device is not something an office can make happen. Whether the re evaluation gets booked, confirmed, and kept before the window closes is entirely within its control. Most missed windows are scheduling failures rather than patient failures.

It is a clean automation target because the deadline is knowable the day the device ships. An AI layer works backward from it: book the re evaluation early, confirm it, call again when the patient cancels, and escalate to staff when the window is inside two weeks with nothing on the schedule. It also absorbs the coinsurance questions that otherwise land on a scheduler. Reviewing the adherence data and deciding what it means belongs to the practitioner.

Supply calls are eating the line that should be booking studies

Mask replacements, tubing, filters, cushions, and the perennial question of what the insurance will cover this year. In a sleep practice this is high frequency, low complexity, and it arrives on the main line all day.

The damage is not the time each call takes. It is what those calls displace. When practice leaders were asked where to focus patient access effort, the answers split across no shows at 27%, online scheduling at 24%, phone access at 22%, and wait times at 21%. Those four are the same problem viewed from different chairs, and a phone line saturated by supply reorders produces all of them at once.

Supply and logistics calls are the most automatable traffic a sleep practice has. Confirming what is due, checking what has already shipped, capturing a reorder request, routing a device fit complaint to the right staff queue, and telling the patient a real date. None of it needs a clinician and none of it needs a scheduler.

Pulling that traffic off the main line is not primarily a cost story. It is what makes the outbound work possible. The staff time that was answering the fifteenth mask question of the day is the same time that books the titration.

Recall lists rot faster than anyone plans for

Every sleep practice has one: the list of patients who completed a study and never came back, or who were set up on a device eighteen months ago and have not been seen since. It is usually a spreadsheet, it is usually out of date, and working it is the first thing dropped in a busy week.

Staffing is why. In MGMA’s November 2023 Stat poll, only 24% of medical group leaders said patient access had improved that year, while 38% said it stayed the same and another 38% said it got worse – and MGMA’s December 2025 polling shows access pinch points (no-shows, phones, scheduling, wait times) still splitting leaders’ attention four ways. Recall outreach is the work that exists entirely in the margin, and the margin is where the shortage lands first.

The list also decays in ways that make the calls worse. Phone numbers change, insurance changes, and a share of the list has already been seen elsewhere. Half of a recall campaign is finding out which entries are still real, and that discovery work is exactly what a person should not be spending an afternoon on.

Automated recall outreach handles the top of that funnel. It calls, confirms the patient is still with the practice, offers a real appointment against the correct athena appointment type and provider group, books it, and marks the entries that are no longer valid so the list gets smaller instead of longer. Anything that turns into a question about symptoms or the device itself gets routed to clinical staff, with the call history attached.

Two appointment types that look identical and are not

Home sleep testing and an in lab study are different products with different prep, different durations, different authorization rules, and often different locations. To a caller they are the same request: I need a sleep study. To the schedule they are not interchangeable at all.

The generic slot problem makes it worse. Template slots configured as a generic block will be returned when you search for a specific type, and a generic slot may or may not be eligible for the in lab study a particular provider performs at a particular location. Which specific types a generic slot can actually hold, per provider and per department, is the whole mapping problem, and most practices carry it in one person’s head.

Getting that mapping written down is the prerequisite for automating anything. It is also the part of an implementation that takes the longest, because nobody has ever had to state the rules out loud before.

Once it exists, the automation can be strict where a human would guess. It books only the appointment type the order specifies, refuses to put an in lab study into a slot that cannot hold it, and hands the call to a scheduler when the order and the available slots genuinely conflict. That handoff is the point, not a failure of it.

Key takeaways

Treat the study, results visit, titration, and follow up as one sequence with one owner, not four independent bookings that each need a trigger.
Work backward from the coverage re evaluation window the day the device ships. The visit is administrative and controllable even when device use is not.
Move supply and reorder calls off the main line first. That traffic is what displaces the outbound calls that finish the episode.
Automate the top of recall outreach so the list shrinks. Marking dead entries is as valuable as booking live ones.
Write down which athena appointment types a generic template slot can actually hold, per provider and per department, before automating any of it.
Report the sequence completion rate, not the number of studies performed. Studies booked without a follow up are the leak.

A sleep practice does not usually lose patients at the front door. It loses them in the gaps between four appointments that nobody was assigned to connect. The study happened, the report came back, and then a queue nobody had time to work decided the rest.

The fix is unglamorous. Book the next stage while the patient is still engaged, know the coverage deadline before it is close, and stop letting mask questions occupy the line that should be making those calls. If your practice runs on athenahealth, the appointment types, provider groups, and order queue are already sitting there. Somebody just has to work them on a Tuesday evening.

Sources: