Sleep & CPAP Report — Night of 2026-08-14

Sources: health-report-2026-08-14-to-2026-08-15.json (Apple Watch sleep stages, exported 08:17 on 08-15) + OSCAR_Greg_by_Day_2026-08-14.csv (ResMed AirSense/AirFlow 11 via OSCAR). Session start was 23:46 on 08-14, so the sleep-stage session's calendar date lines up directly with the CPAP Date. CPAP usage (7.38 h) actually exceeds Apple Watch TIB (7.18 h) tonight by ~12 min — mask went on before the Watch scored "in bed"; noted as a minor data-alignment caveat.

---

⚠️ Flags

• The obstructive cluster the last report expected to resolve has reasserted. OA = 11 tonight — 4 of the last 5 nights have run OA ≥ 8, and 3 of the last 4 nights are at OA ≥ 11 (12, 13, 8, 11). OA is now elevated vs. the 30-day mean of 7.93 on 6 of the last 7 nights. The 08-13 walk-back (OA 8, AHI 1.29) did not carry forward — instead auto-titration pushed pressure back up (7.01 → 7.94 cmH₂O, +0.93) and the machine also drove 90th-percentile pressure from 9.54 → 10.86 (+1.32). This is the strongest evidence yet of a genuine multi-night obstructive shift rather than 2-night noise. Yesterday's "watch condition" (OA stays ≤ 10 for 3 consecutive nights to clear the concern) is now clearly broken.
• Awake time = 33.6 min — rank 3 of 50 matched nights (98th percentile). That's +69% vs. the 7-day mean (20.0 min) and +80% vs. the 30-day mean (18.6 min). Sleep efficiency dropped to 92.19%, well below the 30-day mean of 95.5% and the lowest efficiency of the last 10 nights. This is a fragmentation signal — and unusually, it is not accompanied by high leak (leak is bottom-quintile clean). The likely mechanistic driver is the returned obstructive load (OA=11, RERA=1 after zero last night, pressure and pressure-tail both stepped up).
• Deep sleep = 9.73% — 2nd consecutive bottom-decile night (rank 45/50; last night rank 45/50 also at 10.93%). Two consecutive sub-11% deep-sleep nights is unusual — the 30-day mean is 14.79% and the median 14.55%. REM stayed elevated at 29.20% (rank 5/50), matching yesterday's high-REM/low-deep pattern. Unlike yesterday, tonight the low deep % now has plausible respiratory-fragmentation support (elevated OA, elevated awake time, elevated pressure) rather than just being a duration-artifact of a long sleep window.
• Leak stays fully resolved — 0.22 L/min, 5th consecutive very-clean night (0.17 → 0.42 → 0.28 → 0.20 → 0.22 across 08-10 → 08-14). This is now the cleanest 5-night leak stretch in the entire matched-night dataset. Cleanly rules out mask leak as the cause of the resurgent OA. The obstructive signal is not a fit problem.
• Second consecutive long-sleep night: TIB 7.18 h + CPAP 7.38 h. Combined with yesterday's 7.60 h TIB / 7.75 h CPAP, the last two nights are the longest 2-night stretch in the current dataset. High-usage nights mechanically raise absolute event counts; note that the per-hour AHI (1.63) is still modest despite OA=11, and stays below the 30-day AHI mean (1.78).

---

1. Summary Table — Night of 2026-08-14

Metric	Value
Bedtime (session start)	23:46 on 08-14
Time in bed	7.18 h (430.5 min) — rank 5/50
Total sleep duration	6.61 h (396.9 min)
Sleep efficiency (TST/TIB)	92.19% — well below 7-day (95.0%) and 30-day (95.5%); lowest of last 10 nights
Awake time	33.6 min — rank 3/50 (98th percentile); +69% vs. 7-day (20.0), +80% vs. 30-day (18.6)
Core sleep	4.04 h (242.4 min) — 61.06%
Deep sleep	0.64 h (38.6 min) — 9.73% — rank 45/50; 2nd consecutive bottom-decile night
REM sleep	1.93 h (115.9 min) — 29.20% — rank 5/50 (top decile)
CPAP usage	7.38 h — rank 4/53; +11% vs. 7-day (6.62), +10% vs. 30-day (6.70)
AHI	1.63 — near baseline; -9% vs. 7-day (1.78), -8% vs. 30-day (1.78)
RDI	1.76 — slightly above AHI (1 RERA scored)
OA / H / CA / UA / RERA	11 / 1 / 0 / 0 / 1 — OA elevated; CA cleared; 1 RERA back after zero last night
Pressure avg	7.94 cmH₂O — +0.93 vs. last night's 7.01; equals dataset median
Pressure 90th %ile	10.86 cmH₂O — +1.32 vs. last night's 9.54; highest of last 5 nights
Leak avg	0.22 L/min — 5th consecutive clean night
Leak 90th %ile	0.0 L/min — clean
Compliance flag	1 (compliant)

---

2. 10-Night Trend & Leak Context

Date	AHI	OA	H	CA	RERA	Leak avg	Hours	Deep %	REM %	Awake (min)	Efficiency
08-05	1.49	7	2	1	0	1.90	6.72	11.33	23.31	10.5	97.3%
08-06	2.83	11	4	2	1	0.56	6.00	15.91	19.42	9.5	97.3%
08-07	1.69	8	2	1	1	1.20	6.50	—	—	—	—
08-08	2.34	6	10	0	0	0.87	6.83	16.17	26.60	15.6	96.1%
08-09	1.34	6	3	0	2	3.00	6.72	10.97	23.65	15.1	96.2%
08-10	0.58	3	0	0	0	0.17	5.20	15.30	20.31	20.1	94.7%
08-11	2.57	12	4	0	0	0.42	6.62	13.57	25.24	21.1	94.6%
08-12	2.68	13	3	2	1	0.28	6.72	20.78	19.11	28.1	92.8%
08-13	1.29	8	0	2	0	0.20	7.75	10.93	26.70	20.1	95.6%
08-14	1.63	11	1	0	1	0.22	7.38	9.73	29.20	33.6	92.2%

Four-night OA composite (08-11 → 08-14): 44 OA + 8 H + 4 CA + 2 RERA across 28.47 h of use, with leak effectively zero every night. Per-hour OA rate: 1.55 events/h — clearly above the trailing 30-day OA-rate (~1.19/h). The cluster is real and is not being driven by mask fit.

Leak vs. sleep architecture — 50 matched nights (full history through 08-14)

Buckets on nightly leak average:

• Low-leak (≤ 0.30 L/min): n = 16 — Mean deep: 16.23% | Mean awake: 17.5 min
• Higher-leak (> 0.30 L/min): n = 34 — Mean deep: 14.76% | Mean awake: 18.6 min
• Very-high-leak (≥ 1.0 L/min): n = 14 — Mean deep: 12.96% | Mean awake: 16.7 min

the user r (leak avg vs. deep %): −0.21 (n=50). the user r (leak 90th vs. deep %): −0.25. the user r (leak avg vs. awake min): −0.08. the user r (AHI vs. deep %): +0.18. the user r (AHI vs. awake min): −0.01.

Interpretation: Tonight both the leak/deep model and the AHI/awake model misfire directionally — a low-leak night with modest AHI is not what predicts a 33.6-min awake time and a bottom-decile deep %. What the correlations don't capture is the sub-threshold obstructive/RERA/pressure-response signal: the machine had to work harder tonight (pressure avg +0.93, 90th %ile +1.32, RERA present, OA elevated) while the leak channel showed nothing — and the sleep architecture (top-3 awake, bottom-decile deep, low efficiency) responded to that hidden respiratory work rather than to leak. This is the second consecutive night with a low-deep/high-REM profile, but tonight it comes with clear fragmentation markers (awake +80% vs. baseline, efficiency −3.3 pp) that weren't present last night. Yesterday's "duration artifact" read is weaker tonight; a real respiratory-fragmentation contribution is now plausible.

---

3. Comparison to Trailing Averages

Metric	Tonight (08-14)	7-day avg (08-06–08-13)	30-day avg (through 08-13)
AHI	1.63	1.78	1.78
Leak avg (L/min)	0.22	0.88	0.75
Leak 90th (L/min)	0.0	3.09	2.24
Pressure avg (cmH₂O)	7.94	7.43	7.55
Usage hours	7.38	6.62	6.70
OA count	11	8.0	7.9
Deep sleep %	9.73%	14.62%	14.79%
REM sleep %	29.20%	23.60%	24.09%
Awake time (min)	33.6	20.0	18.6
Efficiency	92.2%	95.0%	95.5%

Split verdict: CPAP leak/pressure numbers still read clean vs. baseline, but the event side (OA, RERA, pressure step-up) and the sleep side (awake +80%, efficiency −3.3 pp, deep bottom-decile) are all pointing the same direction — a respiratory-fragmentation night despite low leak. AHI (events/hour) understates the story because a longer window normalized the count.

Weekly AHI/Leak/Pressure trend

Week	AHI avg	Leak avg	Pressure avg	OA avg	Deep %	Awake (min)	n
07-15 to 07-21	1.82	0.43	8.27	7.6	16.15%	14.5	7
07-22 to 07-28	1.68	0.87	7.41	8.6	16.30%	17.4	7
07-29 to 08-04	1.74	0.70	7.03	7.3	12.60%	24.7	7
08-05 to 08-11	1.83	1.16	7.51	7.6	13.87%	15.3	7
08-12 to 08-18 (partial, n=3)	1.87	0.23	7.65	10.7	13.81%	27.3	3

Two things to notice in the partial week: (1) OA has averaged 10.7 so far — the highest weekly OA-average in the current dataset by a clear margin, driven by the 12/13/11 pattern with only 08-13's 8 pulling it down. (2) Awake time is averaging 27.3 min — the highest weekly awake average in the dataset, roughly 50% above every prior week. Leak has completely reset (0.23 avg — the lowest weekly leak average anywhere in the dataset), which strengthens the read that the elevated OA/awake profile isn't a fit issue.

The 4-week AHI creep (1.82 → 1.68 → 1.74 → 1.83 → 1.87) has continued marginally upward for a 5th consecutive week, but AHI itself remains modest. The more informative weekly-level signal now is OA count and awake time, not AHI.

---

4. Outlook

The 08-11/12 obstructive cluster is now a 4-of-last-5-nights pattern, not a 2-night blip. OA has held ≥ 8 on 08-11, 08-12, 08-13, and 08-14 (with only 08-10 at OA=3 in that window), and the auto-titrator's response tonight — pushing average pressure up nearly a full cmH₂O and 90th-percentile pressure up 1.3 cmH₂O — is the machine confirming the same read. The concern the last report deferred ("if OA stays ≤ 10 for 3 consecutive nights, the persistent-obstructive concern fully clears") did not materialize. It's worth watching whether tomorrow's session sustains the elevated pressure or the machine releases it again.

The sleep-side effect is real, not just an architecture reshuffle. Last night's low-deep/high-REM read plausibly could have been a long-window artifact. Tonight can't be explained away that way: awake time is top-3 in the dataset, efficiency dropped 3+ points below baseline, and the pattern coincides with the machine working harder on the respiratory side. Two consecutive sub-11% deep-sleep nights is the first back-to-back bottom-decile deep-sleep pair in the recent record.

What to watch next. (1) OA on 08-15 — if it stays ≥ 8, this becomes a 5-of-6-night pattern and warrants raising minimum pressure or a positional check. (2) Auto-titration pressure — whether it holds the elevated setting or drops back. Sustained higher 90th-%ile pressure would confirm the machine is chasing genuine obstructive load, not noise. (3) Awake time and efficiency — whether tonight's 33.6 min / 92.2% is one bad night or the start of a fragmentation trend; if efficiency stays under 94% for 3 consecutive nights that's a meaningful shift. (4) Leak is a solved problem for now — 5 consecutive clean nights, cleanest 5-night stretch on record — no action needed there.

---

5. Caveats

• CPAP usage (7.38 h) vs. Apple Watch TIB (7.18 h): tonight CPAP exceeds TIB by ~12 minutes, meaning the mask was on before the Watch scored "in bed" (or the Watch missed the start of the sleep window). Event counts include ~12 min the Watch didn't see as sleep, so absolute counts (OA=11) may be very slightly inflated relative to a strict in-sleep window. Per-hour rate (AHI 1.63) is unaffected.
• Matched-night count computed on this run: n=50 with sleep (53 CPAP nights minus 3 unmatched: 07-03, 07-17, 08-07). Historical matched-nights drift slightly between runs as the file set changes and matching is recomputed each run; figures in this report are self-consistent within the current dataset but may differ marginally from prior reports.
• Unmatched CPAP-only nights in the current file set: 07-03, 07-17, 08-07 (no matching Apple Watch sleep session).
• Trailing averages are computed from the 7 and 30 most-recent matched nights before tonight. The 7-day window here spans 08-06 through 08-13 (08-07 has CPAP but no sleep stages, so it contributes to leak/pressure/AHI trailing means but not to deep/awake means). The 30-day window uses the full available matched depth (which is currently 30 nights back to ~07-15).
• Weekly OA rise (partial week n=3) should be read cautiously — only 3 nights so far, and one long-sleep night (08-13) has an OA count elevated purely by the longer window. The 4-of-5-nights OA≥8 pattern is a more robust framing than the weekly average.
• No SpO2 data in tonight's CPAP CSV export (columns present but blank) — consistent with every export in the current format.
• Leak 90th/95th column naming: current-format exports (07-16 onward) report a 90th-percentile leak column; earlier exports used a 95th-percentile column. Both treated as the same "tail leak" metric for trend purposes.
• Correlation coefficients and weekly averages are recomputed from the full historical file set on each run; small drifts vs. figures cited in earlier reports can occur — not data errors.
• Apple Watch sleep-stage classification is noisy at the single-night level; back-to-back low-deep readings should be interpreted with awareness that measurement variance is non-trivial. The fact that tonight's low-deep reading is accompanied by objectively-measured high awake time and reduced efficiency strengthens the read, but doesn't eliminate the caveat.