Time below 90% oxygen: lower with CPAP
23 min with60 min withouttypical night of 35 with CPAP usage, 124 without
Guides · A diary drawn from a real record, rebuilt every night. Figures as of Nov 14, 2026.
Year two of this diaryA ring for each year, since Oct 2025.A diary · CPAP · nights since Oct 18, 2025
Hollis is a real person who keeps their health record in Selfsense. Hollis is not their name. Hollis has had a CPAP machine for years and uses it on and off. Its nightly record here starts on Oct 18, 2025, which is as far back as the record reaches, not when the machine started. This page rebuilds itself from that record every night.
Of 392 nights the machine reported, it ran at least 4 hours on 78 and not at all on 301. The other 13 were shorter, and are left out of the comparison below: they are neither.
Each row is one measure from the watch: the middle half of the nights with CPAP usage and the middle half of the nights without, on one scale, with the typical (median) night marked on each. Where the two bars sit apart the nights differ; where one lies over the other they do not. Pick a card and its measure stays lit while the rest step back; pick it again and the drawing is plain.
23 min with60 min withouttypical night of 35 with CPAP usage, 124 without
95.0% with94.0% withouttypical night of 73 with CPAP usage, 294 without
3.4/hr with3.8/hr withouttypical night of 35 with CPAP usage, 124 without
68 bpm with68 bpm withouttypical night of 77 with CPAP usage, 301 without
Also drawn, without a card of its own: Heart rate variability (Garmin): no real difference (24 ms with, 25 ms without); Deep sleep: no real difference (68 min with, 74 min without); REM sleep: no real difference (85 min with, 91 min without).
Every measure comes from one device, the watch, which reads oxygen at the wrist: it is not a sleep study. A difference inside the measure’s normal night-to-night wobble is called no real difference. Time below 90% oxygen and oxygen dips per hour (ODI 3%) start later, in June 2026, when the watch began recording them, so their comparison rests on fewer nights.
On the 78 nights it ran at least 4 hours, the machine ran a typical 7.6 hours and counted a typical 0.3 breathing pauses an hour (its AHI). Under 5 an hour is the usual line for sleep apnea under control on the machine. It only counts while the mask is on, so it says nothing about the nights without.
The nights with the machine were not picked at random. Hollis used it in stretches, 10 of them, 3 to 12 nights long, the longest Oct 20, 2026 – Nov 2, 2026, with weeks and months without in between (the columns above). So nights with CPAP against nights without is also those weeks against the rest: season, weight, other changes, whatever made it a good stretch to wear the mask.
What it does show is which measures move with the mask and which do not: time below 90% oxygen differs; blood oxygen, oxygen dips per hour (ODI 3%), resting heart rate, heart rate variability (Garmin), deep sleep and REM sleep barely do.
Every night Selfsense rebuilds a snapshot of this diary from Hollis’s record: how many nights each week the machine ran, the middle half and the typical night of each measure with and without it, and the machine’s own typical figures. The page reads only that snapshot, and the record itself is never public; no single night’s reading is published. Every sentence about the numbers is worked out by code from the numbers themselves; no AI reads the record to write them.
This is what Selfsense does with your own record, privately: your machine’s nights next to your watch’s, on one page. Start yours.
One person’s record, not medical advice, and not a recommendation of any treatment. One person’s numbers say nothing about what will happen to you. Sleep apnea is diagnosed with a sleep study and treated by a clinic.