Sleep & CPAP Report — Night of 2026-08-20

Sources: health-report-2026-08-20-to-2026-08-21.json (Apple Watch sleep stages, exported 05:55 on 08-21) + OSCAR_Greg_by_Day_2026-08-20.csv (ResMed AirSense/AirFlow 11 via OSCAR). Session start was 23:08 on 08-20 — 40 minutes later than 08-19's 22:28, reverting to the 08-16/17/18 bedtime cluster and ending the one-night early-bedtime experiment. CPAP hours 6.27 vs. Watch TIB 6.10 (Δ ≈ 10.4 min, CPAP longer — machine ran briefly past Watch session close; tight alignment).

---

⚠️ Flags

• NATURAL EXPERIMENT RESOLVED — LEAK FIX WORKED, BUT AHI/OA DID NOT FOLLOW. The mechanical-therapy hypothesis is REJECTED. The 08-19 report set up an explicit test: "if fixing the leak also fixes the AHI, this whole cluster is a mechanical-therapy problem and not an underlying obstructive-load problem." The mask intervention ran overnight. Leak avg collapsed from 1.92 → 0.69 L/min (−64%) — the single biggest one-night drop in the current cycle, decisively reversing the 5-night monotonic run. Leak 90 fell 4.8 → 1.2 (−75%). Leak max fell 26.4 → 15.6. The mask fix worked. But AHI moved the wrong way: 2.20 → 2.55 (+16%), and OA count climbed 10 → 14 (+40%). So the leak-and-AHI covariance seen on 08-18/19 was not a causal linkage — it was two things elevated in parallel for different reasons. The residual AHI/OA elevation now needs its own explanation: this is a real underlying obstructive-load elevation, not a therapy-delivery artifact. Escalate to sleep clinician on next check-in.
• AHI CLUSTER SEVERELY ESCALATED — AHI 2.55 exceeds the 2.5 reopen threshold; OA count 14 is the highest of the entire post-07-11 pressure-instrumented window (55 matched nights) except for the outlier 06-26 (OA=23) which predates the current cluster. 3-night AHI trajectory: 1.63 → 2.20 → 2.55 (monotonic, three consecutive rising nights). 3-night OA trajectory: 7 → 10 → 14 (monotonic, +100% over three nights). OA/hr tonight = 2.23 — nearly 2× the 30-day baseline (1.14/h) and the highest single-night OA rate since at least 08-11 (12 OA / 6.62 h = 1.81/h). The 08-11 cluster that opened two weeks ago is now clearly not a two-week noise event; it is an escalating trend. AHI is +54% vs. 30-day (1.66), +53% vs. 7-day (1.67).
• DEEP-SLEEP COLLAPSE — 7.68% is the 2nd-lowest of 55 matched nights (only 08-01's 5.82% is lower, and that was a 66-min-awake anomaly with abnormal REM at 10.6%). Deep dropped from 16.50% → 7.68% (−8.82 pts) night-over-night — the largest single-night deep-sleep decline of the trailing 30 nights. Deep is −6.60 pts vs. 30-day (14.27%) and −6.00 pts vs. 7-day (13.68%). This inverts the 08-19 architecture recovery in a single night. Combined with tonight's AHI escalation, this is the first night in the current cycle where the architecture channel and the CPAP-mechanics channel are BOTH clearly deteriorating — the "decoupling" finding that held through the 08-18/19 window is under pressure. Two-of-two most-recent nights with deep <9% now (08-18: 8.86%, 08-20: 7.68%), with 08-19's 16.50% looking increasingly like the outlier in the middle rather than a return-to-baseline.
• PRESSURE STAYS ELEVATED DESPITE LEAK FIX — P_90 11.16 is +0.82 vs. 08-19 and rank 40/55. P_avg pulled back from 8.18 to 7.80 (still +0.36 vs. 30-day 7.44), but P_90 actually climbed to 11.16 — the 4th-highest P_90 of matched history. The 08-19 report predicted: "if a leak fix on 08-20 also drops P_avg back to ~7.5, that confirms the whole pressure elevation was a compensation for lost seal. If P_avg stays ≥8 even after a leak fix, the machine is seeing real sustained obstructive demand." P_avg landed at 7.80 (partial recovery) and P_90 is at 11.16 (no recovery on the tail). The auto-titrator is still working hard against real load, not phantom leak. Consistent with the leak-fix-didn't-fix-AHI finding — same explanation.
• TIB SHORTENED — 6.10 h is the shortest matched night in the last 11 nights. Later bedtime (23:08 vs. 22:28 = +40 min) combined with typical wake time cut TIB by ~34 min vs. 30-day (399.6 min). Sleep efficiency stayed normal (96.43% vs. 30-day 95.43%) and awake time was actually low (13.0 min, better than 30-day 18.8), so the shorter TIB translated directly into less total sleep (5.88 h TST vs. 30-day 6.36 h). Deep-sleep minutes (27.1 min) are ~half the recent typical (50–65 min), which is partly TIB compression and partly architecture — the pct drop makes clear this isn't just less-time-in-bed math.

---

1. Summary Table — Night of 2026-08-20

Metric	Value
Bedtime (session start)	23:08 on 08-20 (+40 min vs. 08-19's 22:28)
Time in bed	6.10 h (365.8 min) — shortest of last 11 nights; −33.8 min vs. 30-day (399.6)
Total sleep duration	5.88 h (352.7 min)
Sleep efficiency (TST/TIB)	96.43% — +1.00 pt vs. 30-day (95.43%); rank 33/55
Awake time	13.05 min — −31% vs. 30-day (18.83); rank 23/55 (better than average)
Core sleep	4.29 h (257.9 min) — 71.41% (elevated; typical band 55–65%)
Deep sleep	0.45 h (27.1 min) — 7.68% — rank 2/55 (2nd-lowest of matched history); −8.82 pts vs. 08-19 (16.50%); −6.60 pts vs. 30-day (14.27%)
REM sleep	1.23 h (73.8 min) — 20.91% — −3.20 pts vs. 08-19 (24.11%); −2.87 pts vs. 30-day (23.78%)
CPAP usage	6.27 h — −8.2% vs. 08-19 (6.83); −6.7% vs. 30-day (6.73)
AHI	2.55 — exceeds the 2.5 cluster-reopen threshold; +16% vs. 08-19 (2.20); +54% vs. 30-day (1.66); rank 38/55
RDI	2.55 (RERA = 0)
OA / H / CA / UA / RERA	14 / 2 / 0 / 0 / 0 — OA count is highest of the current cluster window; H, CA, UA, RERA all clean
OA per hour	2.23 /h — ~2× 30-day baseline (1.11/h); +53% vs. 08-19 (1.46/h)
Pressure avg	7.80 cmH₂O — −0.38 vs. 08-19 (8.18); +0.36 vs. 30-day (7.44); rank 25/55
Pressure 90th %ile	11.16 cmH₂O — +0.82 vs. 08-19 (10.34); +1.11 vs. 30-day (10.05); rank 40/55 (4th-highest of matched history)
Pressure max	12.46 cmH₂O
Leak avg	0.69 L/min — −64% vs. 08-19 (1.92); breaks the 5-night monotonic run; rank 28/55
Leak 90th %ile	1.2 L/min — −75% vs. 08-19 (4.8); back to baseline territory
Leak max	15.6 L/min — −41% vs. 08-19 (26.4)
Compliance flag	1 (compliant)

---

2. 11-Night Trend & Context

Date	AHI	OA	H	CA	RERA	Leak avg	Leak 90	P_avg	P_90	Hours	Deep %	REM %	Awake (min)	Eff
08-09	1.34	6	3	0	2	3.00	10.8	7.79	10.94	6.72	10.97	23.65	15.1	96.19%
08-10	0.58	3	0	0	0	0.17	0.0	6.89	9.46	5.20	15.30	20.31	20.1	94.73%
08-11	2.57	12	4	0	0	0.42	1.2	7.28	10.02	6.62	13.57	25.24	21.1	94.61%
08-12	2.68	13	3	2	1	0.28	1.2	7.99	10.00	6.72	20.78	19.11	28.1	92.80%
08-13	1.29	8	0	2	0	0.20	1.2	7.01	9.54	7.75	10.93	26.70	20.1	95.60%
08-14	1.63	11	1	0	1	0.22	0.0	7.94	10.86	7.38	9.73	29.20	33.6	92.19%
08-16	1.13	3	3	1	2	0.91	4.8	7.92	10.94	6.22	14.73	19.78	6.0	98.34%
08-17	1.11	5	2	0	0	1.14	1.2	7.57	9.54	6.28	14.21	25.35	7.5	97.95%
08-18	1.63	7	2	1	1	1.19	3.6	7.55	11.11	6.12	8.86	18.71	13.5	96.34%
08-19	2.20	10	5	0	0	1.92	4.8	8.18	10.34	6.83	16.50	24.11	7.0	98.25%
08-20	2.55	14	2	0	0	0.69	1.2	7.80	11.16	6.27	7.68	20.91	13.0	96.43%

(08-15 not in table: no matched Watch sleep-stage session — CPAP-only that night was AHI 1.53, OA 8, Leak 0.31, P_avg 8.35, Hours 5.88.)

3-night composite (08-18 → 08-20): AHI mean 2.13 (+30% vs. 30-day 1.66) • OA total 31 across 19.22 h use = 1.61 OA/h (+45% vs. 30-day 1.11/h) • Deep mean 11.01% (−3.26 pts vs. 30-day 14.27%) • REM mean 21.24% (−2.54 pts vs. 30-day 23.78%). All four channels — event load, event rate, deep, REM — worse than 30-day averages, in the same 3-night window.

Leak trajectory closure (6-night matched, 08-16 → 08-20): 0.91 → 1.14 → 1.19 → 1.92 → 0.69. The monotonic run ended tonight with a −64% drop after the mask intervention. Whatever was mechanically wrong — cushion seat, strap tension, pillow seal — appears to be fixed. Leak avg is now back into the mid-range bucket; 90th %ile back to 1.2 (baseline). This is the diagnostic win of the cycle. But it also isolates the AHI/OA problem: whatever is driving the event-load elevation is not the mask seal.

The 08-19 natural experiment — results table

Prediction from 08-19 report	Actual 08-20 outcome	Verdict
Leak avg ≤0.5 = "fix confirmed"	Leak avg 0.69	Fix partial-to-confirmed (0.69 is well below 08-19's 1.92 and below the ≥1.0 mechanical-issue threshold; not quite ≤0.5 but close)
Proportional AHI recovery with leak = "coupled system"	AHI rose (2.20 → 2.55)	Coupled-system hypothesis REJECTED
Independent AHI elevation = "real event-load"	AHI up 16%, OA up 40% despite leak fix	Independent-load hypothesis CONFIRMED
P_avg returns to ~7.5 = "titrator-compensation confirmed"	P_avg 7.80 (partial return); P_90 11.16 (no return)	Titrator still responding to real load, not phantom leak
Deep/REM % continued baseline-normal	Deep 7.68% (2nd-lowest of matched history)	Prediction FAILED — architecture crashed

Interpretation. The clean version of the story is: 08-18/19 had two things going on simultaneously — a real mechanical mask problem AND a real underlying obstructive-load elevation. The mask problem was fixable and got fixed. The obstructive load was not the mask; it is something else. Tonight isolates that "something else" cleanly.

Leak vs. sleep architecture — 55 matched nights (through 08-20)

Buckets on nightly leak average (added a mid-range bucket for cleaner separation):

• Low-leak (≤ 0.30 L/min): n = 16 — Mean deep: 16.23% | Mean awake: 17.47 min | Mean REM: 23.43%
• Mid-leak (0.30–1.0 L/min): n = 22 — Mean deep: 15.58% | Mean awake: 19.07 min | Mean REM: 23.33%
• Very-high leak (≥ 1.0 L/min): n = 17 — Mean deep: 13.00% | Mean awake: 15.38 min | Mean REM: 24.65%

the user correlations (n=55): leak_avg vs. deep_pct = −0.19 (weak negative, up from −0.05 at n=32 — tonight's low-leak/low-deep point actually pushes against this direction, but the broader dataset has a mild negative signal); leak_avg vs. awake_min = −0.09 (essentially zero); ahi vs. deep_pct = +0.18 (weak positive — high-AHI nights actually associate with slightly higher deep, oddly); pressure_90 vs. awake_min = −0.29 (moderate negative — higher pressure associates with less awake time). The main finding remains: no strong CPAP-mechanics-to-architecture linkage in the full dataset. Tonight's high-AHI/low-deep combination is not a systemic pattern; it's a single-night event that could be shared root cause (obstructive load driving both) or coincidence.

---

3. Comparison to Trailing Averages

Metric	Tonight (08-20)	7-day avg (through 08-19)	30-day avg (through 08-19)
AHI	2.55	1.67	1.66
OA count	14	8.14	7.50
OA per hour	2.23 /h	~1.19 /h	~1.11 /h
Leak avg (L/min)	0.69	0.84	0.85
Leak 90th (L/min)	1.2	2.40	2.56
Pressure avg (cmH₂O)	7.80	7.74	7.44
Pressure 90th (cmH₂O)	11.16	10.33	10.05
Usage hours	6.27	6.76	6.73
Deep sleep %	7.68%	13.68%	14.27%
REM sleep %	20.91%	23.28%	23.78%
Awake time (min)	13.05	16.56	18.83
Efficiency	96.43%	95.93%	95.43%
TIB (h)	6.10	6.62	6.66

Verdict: the pattern reverses from 08-19 — now the CPAP-mechanics picture is mixed (leak much better, event load much worse) and the architecture picture is clearly worse (deep at rank 2/55, REM down 3 pts). The reconcilable reading now: the mask fix succeeded on its own terms, but exposed a separate obstructive-load problem that had been co-occurring. Deep-sleep drop plausibly coupled to the event-load elevation (14 OA in 6.27 h use = 2.23/h means ~1 obstructive event every 27 minutes, which is disruptive to deep-sleep continuity). This is a different picture than 08-19: on 08-19 the CPAP was straining and the sleep was excellent; tonight the CPAP is still straining AND the sleep is bad.

Weekly AHI/Leak/Pressure/OA trend (CPAP-only, full week nights)

Week (Wed–Tue)	AHI avg	Leak avg	P_avg	P_90 avg	OA avg	Hours	n
07-16 to 07-21	1.62	0.46	8.18	10.78	6.7	6.38	6
07-22 to 07-28	1.68	0.87	7.41	10.22	8.6	6.76	7
07-29 to 08-04	1.74	0.70	7.03	9.27	7.3	7.12	7
08-05 to 08-11	1.83	1.16	7.51	10.13	7.6	6.37	7
08-12 to 08-18	1.57	0.61	7.76	10.34	7.9	6.62	7
08-19 to 08-20 (partial, n=2)	2.38	1.30	7.99	10.75	12.0	6.55	2

The new partial week (2 nights) is opening with the highest AHI (2.38), highest OA count (12.0/night), and highest P_90 (10.75) of any week in the trailing 5 weeks. Leak has already dropped from 1.92 to 0.69 within the partial week, so the leak channel is normalizing, but the AHI and OA channels are opening at cluster-worst levels. Extrapolated, this is on track to be the worst week for obstructive event load since the pressure-instrumented window began (07-11).

---

4. Outlook

PRIMARY FINDING: the event-load problem is real and not therapy-delivery. The 08-19 natural experiment ran cleanly. Leak got fixed; AHI/OA got worse. This confirms an underlying obstructive-load elevation independent of the mask seal issue. This warrants a note to the sleep clinician with the 3-night data (08-18/19/20: AHI 1.63/2.20/2.55, OA 7/10/14, all monotonic; leak now controlled at 0.69). Possible drivers to consider: (1) recent weight change, (2) positional changes (side vs. back sleeping), (3) nasal congestion or seasonal allergy load, (4) alcohol/sedative intake in the evening, (5) new medication, (6) upper-airway inflammation, (7) reflux, (8) simply a natural airway-tone shift that may resolve on its own. None of these are diagnosable from the CPAP data — they are hypotheses for the clinician to sort through.

MASK CHECK GETS A WIN. The 08-19 report's mechanical-check recommendation triggered an intervention that produced a clean −64% leak reduction and −75% Leak_90 reduction in one night. Whatever was done — cushion re-seat, strap adjustment, cushion swap, pillow re-fit — worked. Continue with the current setup and watch the leak channel over 08-21/22/23 for stability. If leak stays ≤ 0.5 for 3 consecutive nights, the mechanical issue can be considered fully resolved and closed. If leak drifts back up to ≥1.0, the fix was temporary and a cushion replacement is likely warranted.

PRESSURE CHANNEL IS THE UNDERAPPRECIATED SIGNAL. P_90 climbed to 11.16 despite the leak fix and despite P_avg pulling back slightly. This is the auto-titrator's read of "there is real obstructive demand at the pressure tail." Combined with tonight's OA=14 and rising 3-night trajectory, P_90 is corroborating the event-load story. Watch P_90 alongside AHI — if P_90 stays ≥ 10.5 for the next few nights, the machine's own algorithm is confirming the event-load elevation is not a one-night blip.

DEEP-SLEEP DROP NEEDS ONE MORE NIGHT TO INTERPRET. Tonight's 7.68% is the 2nd-worst of matched history and follows an 08-18 dip (8.86%) — 2-of-3 recent nights are anomalously low. But the middle night (08-19: 16.50%) was fine. Two competing readings: (a) tonight is a coincidence — noisy architecture channel that happened to drop on a night of elevated events; or (b) the elevated event load is finally starting to disrupt deep-sleep consolidation, breaking the previously robust decoupling. 08-21 will disambiguate. If Deep ≥12% on 08-21, tonight was noise. If Deep <10% again, the coupling story becomes plausible and the event-load problem is starting to affect sleep quality — much more urgent.

What to watch on 08-21. (1) AHI/OA — if 08-19/20/21 trajectory continues (2.20 → 2.55 → higher), this is a hardening cluster and warrants clinical escalation this week rather than "on the next check-in"; if AHI drops back below 2.0 with a similar OA count, the 08-20 spike was the local peak. (2) Leak avg — needs to stay ≤ 0.5 to confirm the mask fix held. (3) P_90 — should track AHI: if AHI cools and P_90 cools with it, coupled; if AHI cools but P_90 stays elevated, the machine is still seeing something the event counter is not. (4) Deep % — the disambiguating metric; ≥12% closes the concern, <10% opens it further. (5) TIB — tonight was short (6.10 h); if the pattern of later bedtimes continues, deep-minutes will keep suffering just from time compression, independent of any physiology.

---

5. Caveats

• The rank comparisons in this report use n=55 matched nights (all Watch-tracked sleep sessions ≥3 h paired with CPAP nights that have valid Pressure_Avg data, from mid-June through 08-20). Prior reports used n=32, which was a narrower window that excluded the June JSONs. The 30-day rolling averages are effectively identical between the two conventions (AHI 30-day: 1.66 in both) because both are dominated by the same recent nights, but rank/extreme statements will be more conservative under n=55 than under n=32. Tonight's deep-sleep rank (2/55) is robust to either denominator — it would be rank 2 or 3 in n=32 as well.
• CPAP/Watch alignment tonight is close. CPAP hours 6.27 exceed Watch TIB 6.10 by 10.4 min — machine ran briefly past Watch session close, consistent with 08-19. Per-hour rates and architecture comparisons remain directly interpretable.
• The natural-experiment interpretation depends on assuming the mask fix was the primary intervention overnight and nothing else changed. If the user also, e.g., took a different medication, had a different alcohol intake, slept in a different position, or had a congestion change, those are confounds. But given the report explicitly recommended a mask check and the leak channel responded exactly as predicted for a mask fix (large single-night drop, all leak-related metrics down together), the mask-fix explanation for the leak recovery is very likely correct even under confound risk.
• Deep-sleep interpretation caveat. Apple Watch sleep-stage detection has known limitations — deep-sleep classifications are noisier than total-sleep-duration classifications. A single-night deep-sleep read of 7.68% could be Watch measurement noise as much as physiological reality. This is why the 08-21 read matters — a repeat low reading would be much harder to attribute to measurement noise than a single anomaly.
• Core sleep at 71.4% is elevated vs. the typical 55–65% band. This is the flip side of the low deep pct — if deep and REM both come in low, the residual by construction lands in core. Not independently informative.
• The 08-18 → 08-20 3-night composite (AHI 2.13, OA/h 1.61) is now the worst 3-night rolling composite in the matched history since at least the 08-06/07/08 window (which was influenced by outlier nights). Reference: 30-day rolling AHI is 1.66, so tonight's 3-night is +28% above baseline.
• P_90 = 11.16 is the highest of the last 11 nights and 4th-highest in matched history (n=55). Historical high: 12.94 on an earlier night. The pressure channel has been slowly ratcheting upward across the 08-12 → 08-20 window (P_90 avg over these 8 matched nights = 10.42 vs. 30-day 10.05).
• No SpO2 data in tonight's CPAP CSV (columns present but blank) — consistent with every export in the current format.
• Session-start convention. The date field of the sleep session (2026-08-20T23:07:50) marks the start calendar date, which lines up with the OSCAR "Period" for 2026-08-20. This convention continues to hold consistently.
• CPAP-only nights (no Watch match) in the last 6 weeks remain: 07-17, 08-07, 08-15. No new gaps this cycle.