The Four-Hour Rule: How a 1993 Cut Point Still Grades PAP Therapy Innovation

Four hours a night. Seventy percent of nights. Any thirty consecutive days in the first ninety.

Every compliance dashboard in American homecare reports against those three numbers. Ask a respiratory therapist where they came from and you usually get a shrug and the word Medicare. Ask what they mean physiologically and the answer thins out fast. That's worth pausing on, because those three numbers are the scoreboard that PAP therapy innovation gets graded against — every new pressure algorithm, mask frame, humidifier, and adherence app is ultimately judged by whether it walks patients across a line that was first drawn, for entirely different reasons, in a study of 35 people.

The line was drawn to describe patients, not to treat them

In 1993, Kribbs and colleagues published a paper in the American Review of Respiratory Disease on objective CPAP use. Patients knew their machines logged something; they did not know the logger distinguished time at pressure from time merely switched on. The headline finding was that self-report was generous — people overestimated their use by roughly an hour a night.

But the authors needed a way to split a messy distribution into groups they could talk about. So they called someone a "regular user" if they used the machine at least four hours a night on at least 70 percent of nights. Fewer than half the cohort qualified.

That's it. That's the source. It was a descriptive cut point — a place to put a knife through a histogram — and nothing in the paper claimed four hours was the dose at which apnea stops mattering. The belief that grew out of it is considerably more confident than the thing it grew from.

How many hours of CPAP count as compliant?

For Medicare, and for the commercial payers that follow its lead, compliance means using the device at least four hours per night on 70 percent of nights during a 30-day window in the first three months of therapy. It's written into CMS's national coverage determination for CPAP in obstructive sleep apnea (NCD 240.4), effective March 2008, and it governs whether a DME provider keeps getting paid past the trial period.

It is a billing definition. It sets the threshold at which a payer agrees the equipment isn't sitting in a closet. It has never been a statement about how much treatment a given airway needs, and CMS didn't claim otherwise.

The slippage happened downstream, in the language. "Compliant" is a word with a moral shape to it. Once a patient is compliant, the chart is quiet, the resupply cycle starts, and the clinical question closes.

What the dose-response data actually shows

The best-known attempt to find a real threshold is Weaver et al. (2007) in Sleep, which followed 149 patients on CPAP and asked how many hours of nightly use it took to normalize three different outcomes. The answer was that there is no single threshold — the relationship is continuous, and where you draw the line depends entirely on what you're trying to fix.

Treat that table as a shape, not a set of constants. It's one cohort, the confidence intervals around breakpoints in dose-response curves are wide, and individual variation swamps the averages. What survives replication is the shape itself: more hours, more benefit, no cliff.

Which makes the four-hour bar an interesting choice. It sits at the low end — the point where people stop reporting that they feel sleepy, which is also, conveniently, the least demanding of the three endpoints.

The cardiovascular literature sharpens this. The SAVE trial (McEvoy et al., 2016, New England Journal of Medicine) randomized 2,717 patients with OSA and established cardiovascular disease and found no reduction in cardiovascular events. Mean adherence was 3.3 hours a night. A post hoc propensity-matched analysis of patients using at least four hours suggested a stroke benefit — but that's post hoc, non-randomized, and exactly the kind of finding that should be labeled plausible but thin rather than repeated as though the trial had shown it.

The hours you skip are the expensive ones

Here's the part that makes four hours worse than merely arbitrary, in the order it happens in the body.

A patient falls asleep. The first cycles are NREM-heavy, with slow-wave sleep concentrated in the first third of the night. REM appears about 90 minutes in, briefly. Across the night, REM periods lengthen — the last one can run half an hour or more, and REM density climbs toward morning.

In REM, skeletal muscle atonia includes the upper-airway dilators. Genioglossus tone drops. The pharynx becomes more collapsible than at any other point in the night, so obstructive events run longer and desaturate deeper. Add the drift toward supine posture in the back half of the night, and add REM rebound in the first weeks of therapy in patients who arrived chronically REM-deprived.

So a patient who pulls the mask off at 4 a.m., after a clean four hours, has treated their least apneic sleep and left the worst of it bare. The download reads compliant. The physiology reads otherwise.

What a payment metric does to a product roadmap

If you build or distribute PAP equipment, the return on innovation is concentrated wherever the money is. The money is at the threshold and inside the 90-day window. A feature that carries a 3.4-hour patient to 4.2 hours is financially legible to a DME. A feature that carries a 5-hour patient to 7 hours produces no billable event at all.

This explains a lot about the last two decades of PAP innovation without requiring anyone to be cynical. Expiratory pressure relief, ramp, auto-titration, quieter blowers, lighter headgear, cloud adherence dashboards that page a therapist on day 60 — these are genuinely good engineering, and they cluster hard around getting patients over an early hurdle. Very little clusters around month nine.

The same gap shows up outside PAP. The STAR trial of hypoglossal nerve stimulation (Strollo et al., 2014, NEJM, 126 patients) reported AHI reduction and device usage hours, and the field still lacks an agreed usage endpoint for non-mask therapies. We standardized a threshold for the modality we were reimbursing, not for the disease.

And there's a boundary worth stating plainly: no device fixes what isn't a device problem. Comorbid insomnia, nasal obstruction, claustrophobia, a partner who leaves at hour four — a better blower does not fix a nose.¹

An honest rule of thumb

Tonight, when you open a compliance report, stop reading it as pass/fail and start reading it as a distribution. Look at the mask-off clock time, not just the nightly average. If a patient sits at 4.1 hours with a flat removal time around 4 a.m., that's a failure with paperwork attached — check leak in the final two hours, ask what wakes them, and set the next target at six hours, not four.

What this doesn't settle

I can't tell you whether raising the payer threshold would actually raise usage or simply strip coverage from partial responders who were getting real benefit at three hours. That's an empirical question nobody has run as a trial, and the equity consequences fall on exactly the patients least able to absorb them.

The places to look next: phenotype-specific dose-response, since a REM-predominant patient and a positional patient plainly need different hours; residual on-therapy AHI in the final sleep cycle, which almost nobody reports; and validation work on whether device-reported "usage" tracks mask-on sleep rather than mask-on wake. Then watch CMS, and watch whether the AASM's push to revisit the criterion produces anything more than a position paper.

Four hours was never a finding. It was a place someone put a line so they could count what fell on either side of it — and an industry has been billing against it, and designing against it, for thirty years.