Sleep & CPAP Report — 2026-10-05

🚩 Flags

• 🔴 AHI crossed the clinical threshold of 5 for the first time in the 26-night matched window. AHI 5.20 (rank 1/26; 30d z = +4.64, +166.4% vs 30d mean of 1.95) and RDI 5.85 (rank 1/26, z = +5.44). This is a sharp, one-night outlier — nothing in the recent trajectory predicted it. The 30d trailing mean AHI is 1.95; the previous recent-window maximum was 3.23 (2026-09-19). Clinical note: AHI < 5 is the standard treatment target on CPAP; a single night just crossing the threshold is not a treatment failure, but it's the first time it's happened in the user's recent data.
• 🔴 Obstructive-event counts are unprecedented in the recent window. OA count 21 (rank 1/26; prev max 16; z = +3.45, +138%), H count 11 (rank 1/26; prev max 7; z = +3.50, +272%). These are frank, discrete respiratory events — not subthreshold resistance. Combined 32 events in 6.15 h of use.
• 🔴 Pressure 90th jumped to 12.36 cm H₂O — highest in the 26-night window (z = +4.36, +23.3% vs 30d mean of 10.02). The auto-titrator pushed harder than it has on any recent night, and still couldn't prevent the apneas. Pressure_Max 13.98 (vs 30d max avg ~11.6).
• 🟠 But sleep architecture was actually fine. Deep sleep 15.9% / 56.7 min (rank 6/26, z = +0.95 — above 30d avg of 12.7%). REM 25.2% / 89.8 min (near 30d avg). Awake only 8.0 min (below 30d avg of 10.0 min). Sleep efficiency 97.8%. The respiratory events did not translate into arousals or stage-architecture collapse. This is a striking disconnect: a night with 32 frank apnea/hypopnea events but preserved slow-wave sleep and no fragmentation.
• 🟠 Pattern reversed from last night. Yesterday (2026-10-04) was "high flow-limitation / low frank events" — FL_Avg rank 1/9, AHI 1.18, deep sleep collapsed to 7.9%. Tonight is the mirror image: FL metrics dropped back toward baseline (FL_Avg -41%, Frac_NonZero -54%, Min_Above_010 -39% vs yesterday), but frank apnea events exploded. Min_Above_020 is the one FL metric that spiked tonight (8.2 min vs 30d avg 3.6, rank 1/9, z = +2.62) — i.e., fewer total flow-limitation minutes but when they occurred they were more severe, consistent with the shift from chronic subthreshold resistance to episodic full occlusion.
• 🟡 Leak modestly elevated but not alarming. Leak_90 7.2 (rank 4/26, z = +1.18), Leak_Avg 1.76 (z = +0.96), Leak_Max 22.8 (near 30d avg). Not in mask-reseat territory, but worth watching if it rises again tomorrow.
• This single night erases the week-over-week event-count improvement. Prior 7d AHI avg 1.37 → including tonight the trailing 7d AHI jumps to ~1.92 (+40%). The ~5-week improving trend in event counts is temporarily broken; whether it's a one-off or a trend break depends on the next 2–3 nights.

Summary Table — Night of 2026-10-05

Metric	Value
CPAP usage hours	6.15
Session count	1
AHI	5.20
RDI	5.85
Obstructive apneas (OA)	21
Hypopneas (H)	11
Central apneas (CA)	0
Unclassified apneas (UA)	0
RERA	4
Pressure_Avg	7.93
Pressure_Min	4.00
Pressure_Max	13.98
Pressure_90th	12.36
Leak_Avg	1.76
Leak_Max	22.80
Leak_90th	7.20
FL_Avg	0.0141
FL_P95	0.09
FL_Max	0.58
Min_Above_005 (min)	25.2
Min_Above_010 (min)	15.7
Min_Above_020 (min)	8.2
Frac_NonZero (FL samples)	16.39%
Apple Watch time in bed	364.30 min
Apple Watch total sleep	356.27 min
Apple Watch awake	8.03 min
Apple Watch deep sleep	56.70 min (15.92%)
Apple Watch REM sleep	89.82 min (25.21%)
Apple Watch core sleep	209.75 min (58.87%)
Sleep efficiency	97.80%
Sleep session start	2026-10-05T23:18:09-05:00

Comparison to Trailing Averages

Trailing windows exclude tonight by construction (they cover the 7 and 30 CPAP nights ending 2026-10-04). Matched n = 7 and 25 for the stage-pair rows.

Metric	Tonight	7d avg	30d avg	Δ vs 7d	Δ vs 30d	30d z
AHI	5.20	1.37	1.95	+279.2%	+166.4%	+4.64
RDI	5.85	1.55	2.08	+277.4%	+181.7%	+5.44
OA count	21	6.43	8.84	+226.7%	+137.6%	+3.45
H count	11	2.14	2.96	+413.3%	+271.6%	+3.50
CA count	0	0.14	0.64	-100.0%	-100.0%	-0.62
RERA count	4	1.14	0.80	+250.0%	+400.0%	+2.86
Pressure_Avg	7.93	7.62	7.49	+4.1%	+5.9%	+1.05
Pressure_90th	12.36	10.09	10.02	+22.4%	+23.3%	+4.36
Leak_Avg	1.76	1.27	1.01	+38.1%	+74.1%	+0.96
Leak_Max	22.80	23.31	20.35	-2.2%	+12.0%	+0.29
Leak_90th	7.20	4.29	3.17	+68.0%	+127.3%	+1.18
Usage hours	6.15	6.46	6.41	-4.8%	-4.0%	-0.62
FL_Avg	0.0141	0.0156	0.0155	-10.6%	-9.1%	-0.33
FL_P95	0.09	0.094	0.095	-4.5%	-5.3%	-0.24
FL_Max	0.58	0.46	0.46	+24.9%	+25.7%	+1.23
Min_Above_005	25.20	39.60	39.00	-36.4%	-35.4%	-1.10
Min_Above_010	15.70	16.54	16.88	-5.1%	-7.0%	-0.21
Min_Above_020	8.20	3.39	3.60	+142.2%	+127.8%	+2.62
Frac_NonZero	0.1639	0.259	0.251	-36.8%	-34.6%	-1.43

Sleep-Stage Comparison

Metric	Tonight	7d avg	30d avg	Δ vs 7d	Δ vs 30d
Time in bed (min)	364.30	380.07	380.29	-4.1%	-4.2%
Total sleep (min)	356.27	367.66	370.31	-3.1%	-3.8%
Awake (min)	8.03	12.40	9.97	-35.3%	-19.5%
Deep (min)	56.70	51.47	46.67	+10.2%	+21.5%
REM (min)	89.82	88.89	88.27	+1.1%	+1.8%
Core (min)	209.75	227.31	235.38	-7.7%	-10.9%
Deep %	15.92	14.08	12.67	+1.84 pp	+3.25 pp
REM %	25.21	24.00	23.73	+1.21 pp	+1.48 pp
Core %	58.87	61.92	63.60	-3.05 pp	-4.72 pp
Sleep efficiency	97.80	96.68	97.38	+1.12 pp	+0.42 pp

Read: A short-ish, continuous, highly efficient night (6.07 h in bed, only 8 min awake). Despite the respiratory-event barrage, slow-wave sleep was actually above the 30d average by ~3 percentage points. REM was normal. The architecture suggests the apneas, while numerous, were brief or occurred during stages where the arousals didn't cost deep sleep — or the auto-titrator's aggressive pressure response (90th hit 12.36) terminated them fast enough that stage transitions didn't cascade.

Week-over-Week Deltas (recent 7d 2026-09-28–2026-10-04 vs prior 7d 2026-09-21–2026-09-27)

Metric	Recent 7d	Prior 7d	Δ	Δ %
AHI	1.371	2.409	-1.038	-43.1%
OA	6.429	9.143	-2.714	-29.7%
H	2.143	5.714	-3.571	-62.5%
CA	0.143	0.286	-0.143	-50.0%
RERA	1.143	0.714	+0.429	+60.0%
Pressure_Avg	7.617	7.703	-0.086	-1.1%
Pressure_90th	10.094	10.243	-0.149	-1.5%
Leak_Avg	1.274	1.431	-0.157	-11.0%
Leak_90	4.286	4.800	-0.514	-10.7%
Hours	6.461	6.277	+0.184	+2.9%
FL_Avg	0.0156	0.0136	+0.0021	+15.1%
FL_P95	0.094	0.100	-0.006	-5.7%
Min_Above_010	16.54	19.20	-2.66	-13.8%
Frac_NonZero	0.259	0.189	+0.070	+37.0%

The WoW comparison is from the pre-tonight 7d window. Tonight's event spike is NOT in these numbers. Adding tonight into a running 7d window would reset AHI from 1.37 → ~1.92 (-20% WoW instead of -43%), still improving but the direction depends on the next few nights.

Recent 11-Night Context

Date	AHI	OA	H	CA	RERA	PAvg	P90	LAvg	L90	FL_P95	MinAbv010	MinAbv020	FracNZ	Hours	Deep%
2026-09-23	1.94	9	3	0	2	7.78	10.06	2.87	12.00	—	—	—	—	6.18	18.83
2026-09-24	1.65	3	6	1	0	7.36	10.18	0.59	1.20	—	—	—	—	6.05	9.69
2026-09-26	2.72	13	6	0	1	7.66	10.50	2.37	9.60	0.10	19.2	2.4	0.189	6.98	12.85
2026-09-27	1.41	6	3	0	1	7.74	10.08	1.05	4.80	0.10	18.2	2.9	0.254	6.38	11.66
2026-09-28	1.23	7	1	0	1	7.54	10.36	1.04	3.60	0.11	22.2	3.6	0.237	6.48	13.79
2026-09-29	1.45	8	1	0	0	8.10	11.34	0.44	0.00	0.07	10.6	2.4	0.220	6.22	13.30
2026-09-30	1.44	7	2	0	0	8.01	10.40	1.18	3.60	0.10	16.9	3.5	0.326	6.25	16.10
2026-10-01	0.56	4	0	0	0	6.82	9.00	0.53	2.40	0.07	10.9	2.1	0.235	7.12	16.96
2026-10-03	2.33	7	6	1	2	7.21	9.30	2.77	10.80	0.08	11.3	2.6	0.188	6.00	18.86
2026-10-04	1.18	6	2	0	4	7.90	10.18	1.91	4.80	0.13	25.7	7.1	0.355	6.78	7.88
2026-10-05	5.20	21	11	0	4	7.93	12.36	1.76	7.20	0.09	15.7	8.2	0.164	6.15	15.92

Historical CPAP → Sleep-Stage Correlations

Based on 26 matched (CPAP + Apple Watch sleep-stage) nights in the recent window (CPAP dates 2026-09-06 to 2026-10-05, including tonight). FL correlations use the 9 nights with FL data (FL CSVs are a newer artifact in the pipeline).

Pair	the user r	n	Interpretation
AHI vs deep_pct	+0.001	26	essentially zero
AHI vs awake_min	-0.071	26	negligible
OA vs deep_pct	-0.053	26	negligible
OA vs awake_min	+0.098	26	negligible
H vs deep_pct	+0.027	26	negligible
H vs awake_min	-0.171	26	weak negative
L_avg vs deep_pct	+0.288	26	weak positive
L_90 vs deep_pct	+0.378	26	weak-moderate positive (counterintuitive)
L_max vs awake_min	-0.178	26	weak negative
RERA vs awake_min	-0.451	26	moderate negative
RERA vs deep_pct	+0.327	26	weak-moderate positive (counterintuitive)
FL_P95 vs deep_pct	-0.715	9	strong negative (small n)
FL_Avg vs deep_pct	-0.733	9	strong negative (small n)
Min_Above_010 vs deep_pct	-0.754	9	strong negative (small n)
FL_P95 vs awake_min	-0.569	9	moderate negative (small n)
P_90th vs awake_min	+0.022	26	essentially zero

Leak_90 median split (26 nights, median = 2.40): - Low-leak nights (n=14): mean deep% = 12.65, mean awake = 10.00 min - High-leak nights (n=12): mean deep% = 12.95, mean awake = 9.78 min

Important: tonight is the first night in the matched series where frank apnea events (AHI/OA/H) spiked without a corresponding FL spike or awake-time increase. It is a strongly influential outlier for the "AHI vs deep_pct" and "OA vs deep_pct" correlations — both went to essentially zero (prior-26 values were -0.176 and -0.196 respectively), which is because tonight pairs high AHI with high (not low) deep sleep, counteracting any trend. The FL↔deep_pct correlations (n=9) remain strongly negative — tonight's FL metrics are not extreme and continue to support the earlier reading that flow-limitation severity matters more for sleep architecture than frank event count. The clinically interesting observation here is that in the user's data, flow-limitation / effort metrics track with sleep-stage disruption, while frank apneas (which the machine terminates with a pressure response) do not.

Interpretation

This is the first clinically-notable AHI night in the matched 26-night series. AHI 5.20 just crosses the standard CPAP treatment target of < 5. By z-score it's +4.64 SD above the user's 30-day baseline — well beyond what random night-to-night variation would produce. OA count 21 and H count 11 are both the highest in the recent window by a wide margin. The auto-titrator responded aggressively: Pressure_Avg climbed to 7.93 (vs 7.49 baseline), and Pressure_90th to 12.36 — the highest in the entire recent window, 23% above the 30d mean. Despite that pressure response, 32 discrete respiratory events still broke through.

The architecture-vs-events disconnect is striking. On a night with this many respiratory events, you'd expect to see arousal, fragmentation, reduced deep sleep, maybe reduced REM. Instead: deep sleep 15.9% (actually above average), REM 25.2% (normal), awake 8 min (below average), efficiency 97.8%. The apneas were there, the pressure response was there, but the arousals and stage-architecture costs weren't. Two readings fit this: 1. The apneas clustered in a short period or in a stage (e.g., REM) where the arousals didn't interrupt slow-wave consolidation. Possible if the apnea burst happened early-morning or during a REM period. 2. The auto-titrator's aggressive pressure response terminated each apnea fast enough that the arousal threshold wasn't crossed. The pressure jumped to 13.98 cm H₂O max, well above typical.

The FL pattern reversed from last night. Yesterday was "high chronic resistance, no frank events, deep sleep collapsed". Tonight is "low chronic resistance, severe episodic events, deep sleep preserved". The one FL metric that did spike tonight — Min_Above_020 (minutes with FL above the severe threshold, 8.2 min, +128% vs 30d) — is directionally consistent with the episodic-occlusion pattern: fewer minutes with any flow limitation at all, but when limitation occurred it was severe enough to count as a full event.

Candidate causes to consider (none visible in the data, worth watching): - Supine sleep (back vs side) — common cause of position-dependent OSA breakthrough - Nasal congestion or early-season allergen exposure - Alcohol or sedative within ~4 hours of bed - Late or heavy meal - Different pillow / headboard setup for one night (travel, guest) - Mask cushion aging or displaced partway through the night (Leak_90 at 7.2 is modestly elevated — not alarming, but non-trivial)

Is this actionable? Not yet, from a single night. AHI 5.20 on CPAP is "mild" by clinical standards, and one night at the threshold after months of excellent control is almost certainly a one-off. The right response is: - Watch the next 2–3 nights. If AHI stays elevated (above 3) or OA/H counts stay in double digits on even one more night, that moves from "random outlier" to "pattern to investigate". - If tomorrow is normal, log tonight as a one-off and move on. The 5+ week improving trend in event counts was strong; one night does not break it. - If a cause can be identified from recall (position, congestion, alcohol, late meal, mask reseat issue), noting it will help interpret future spikes. - No immediate pressure or mask change is warranted — the auto-titrator responded correctly, and the sleep architecture tells us the airway events didn't harm the restorative value of the night.

Caveats

• Trailing 7d/30d CPAP windows exclude tonight by design (they cover the 7 and 30 nights ending 2026-10-04). The "Δ vs 7d/30d" and "30d z" columns compare tonight to those pre-tonight baselines.
• Sleep-stage trailing averages are computed over the matched Apple Watch sessions corresponding to those same CPAP windows. 7d matched: n = 7. 30d matched: n = 25. The 2026-09-25 CPAP night has no paired Apple Watch session, so it drops out of the matched window and the 30d matched count sits one short of the full 30d CPAP count.
• FL (flow-limitation) metrics are only available for the last 9 matched nights in the window; FL correlations are n = 9 and should be treated as directional rather than estimates. The pattern (FL effort → reduced deep sleep) has been consistent across prior reports and continues to hold.
• The 2026-10-05 Apple Watch session started at 23:18:09 on 2026-10-05, before midnight — the pairing convention places this with CPAP date 2026-10-05, consistent with how the 2026-10-04 session (start 23:42:16) was paired in yesterday's report. CPAP usage hours (6.15) and Apple Watch time-in-bed (6.07) line up to within 5 minutes, so the pairing is sound.
• All numbers in this report are from the OSCAR daily CSV (OSCAR_Greg_by_Day_2026-10-05.csv), the OSCAR flow-limitation CSV (OSCAR_Greg_FlowLimit_by_Day_2026-10-05.csv), and the Apple Health export (health-report-2026-10-05-to-2026-10-06.json).
• AHI 5.20 is just across the clinical threshold of "mild OSA on CPAP" (AHI < 5 is the standard target). This is not an urgent medical issue from a single night but is the kind of thing worth noting to a sleep clinician if it recurs.