Sleep Architecture as Health Signal
Module 3: Comparative Sleep Analysis
P1 AVG SLEEP
Below target6.0hrs
Target: 7-9 hrs
P1 EFFICIENCY
Watch78.9%
12% nights below 75%
P1 DEEP SLEEP
In range15.1%
Norm: 13-23%
P2 AVG SLEEP
Below target6.7hrs
Target: 7-9 hrs
P2 EFFICIENCY
Watch82.8%
10% nights below 75%
P2 DEEP SLEEP
In range18.8%
Norm: 13-23%
P3 AVG SLEEP
Below target6.2hrs
Target: 7-9 hrs
P3 EFFICIENCY
Watch84.0%
0% nights below 75%
P3 DEEP SLEEP
Watch12.0%
Norm: 13-23%
ARCHITECTURE
Sleep Architecture Over Time
EFFICIENCY
Sleep Efficiency Trends
TIMING
Sleep Timing Analysis
BENCHMARKS
Benchmark Comparison
| Metric | General Norms | Patient 1 (post-HSCT) | Patient 2 (post-Stroke) | Patient 3 (Healthy Control) |
|---|---|---|---|---|
| Deep Sleep % | 13 - 23 | 15.1 (z=-0.6) | 18.8 (z=+0.2) | 12.0 (z=-1.2) |
| REM Sleep % | 20 - 25 | 11.9 (z=-4.3) | 17.2 (z=-2.1) | 12.9 (z=-3.8) |
| Efficiency % | 85 - 100 | 78.9 (z=-1.8) | 82.8 (z=-1.3) | 84.0 (z=-1.1) |
| Total Hours | 7 - 9 | 6.0 (z=-2.0) | 6.7 (z=-1.3) | 6.2 (z=+0.0) |
RECOVERY
Recovery Trajectory
STATS
Statistical Comparison
| Metric | Patient 1 | Patient 2 | p-value | Cohen's d | Cliff's Δ | 95% CI (median diff) |
|---|---|---|---|---|---|---|
| Deep Sleep % | 15.1 | 18.8 | p<0.001 *** | -0.65 (medium) | -0.37 (medium) | [-4.9, -2.1] |
| REM Sleep % | 11.9 | 17.2 | p<0.001 *** | -1.29 (large) | -0.67 (large) | [-7.2, -4.5] |
| Light Sleep % | 52.0 | 46.8 | p<0.001 *** | +0.80 (large) | +0.43 (medium) | [+2.4, +6.7] |
| Awake % | 21.1 | 17.2 | p<0.001 *** | +0.62 (medium) | +0.50 (large) | [+4.0, +5.9] |
| Efficiency | 78.9 | 82.8 | p<0.001 *** | -0.62 (medium) | -0.50 (large) | [-6.0, -4.0] |
| Total Hours | 6.0 | 6.7 | p<0.001 *** | -0.64 (medium) | -0.38 (medium) | [-1.1, -0.5] |
| Bedtime Hour | 24.7 | 23.9 | p<0.001 *** | +0.57 (medium) | +0.37 (medium) | [+0.7, +1.4] |
* p<0.05, ** p<0.01, *** p<0.001 · CI = bootstrap 95% confidence interval for median difference (Patient 1 - Patient 2)
INTERPRETATION
Clinical Interpretation
- Patient 1 (post-HSCT): REM sleep (11.9%) is below population norms (20.0-25.0%), suggesting possible autonomic interference with dream-stage cycling.
- Patient 1 (post-HSCT): Sleep efficiency (79%) is below the recommended 85% threshold.
- Patient 1 (post-HSCT): Average daily sleep debt of 59 minutes below the 7-hour target, accumulating chronic sleep restriction.
- Patient 1 (post-HSCT): High bedtime variability (SD=116 min) suggests inconsistent sleep schedule, which may impair circadian entrainment.
- Patient 2 (post-Stroke): REM sleep (17.2%) is below population norms (20.0-25.0%), suggesting possible autonomic interference with dream-stage cycling.
- Patient 2 (post-Stroke): Sleep efficiency (83%) is below the recommended 85% threshold.
- Patient 2 (post-Stroke): Efficiency trend is statistically declining (-0.06%/week, p=0.000).
- Patient 2 (post-Stroke): High bedtime variability (SD=77 min) suggests inconsistent sleep schedule, which may impair circadian entrainment.
- Patient 3 (Healthy Control): Deep sleep (12.0%) is below the general population norm (13.0-23.0%), consistent with post-treatment sleep disruption.
- Patient 3 (Healthy Control): REM sleep (12.9%) is below population norms (20.0-25.0%), suggesting possible autonomic interference with dream-stage cycling.
- Patient 3 (Healthy Control): Sleep efficiency (84%) is below the recommended 85% threshold.
- Patient 3 (Healthy Control): Average daily sleep debt of 49 minutes below the 7-hour target, accumulating chronic sleep restriction.
- Patient 3 (Healthy Control): High bedtime variability (SD=65 min) suggests inconsistent sleep schedule, which may impair circadian entrainment.
- Deep sleep percentage differs significantly between patients (p=0.000, Cohen's d=-0.65, medium effect).
- Sleep efficiency differs significantly (p=0.000, Cohen's d=-0.62, medium effect).
METHODOLOGY
Methodology & Limitations
Data Source
Sleep architecture data from Oura Ring wearable sensors (oura_sleep_periods table, type='long_sleep'). Durations are recorded in seconds by the Oura API and converted to hours/percentages for analysis.
Architecture Percentages
Computed as stage_duration / (total_sleep_duration + awake_time) × 100,
ensuring all stages sum to 100% of time in bed.
Statistical Tests
- Mann-Whitney U: Non-parametric test for distribution differences (does not assume normality)
- Cohen's d: Standardized mean difference (pooled SD); |d| < 0.2 = negligible, < 0.5 = small, < 0.8 = medium, else large
- Cliff's delta: Non-parametric effect size based on rank ordering
- Bootstrap CI: 5,000-iteration bootstrap for median difference confidence interval
- Linear regression: Ordinary least squares for efficiency trend
- Spearman rank correlation: Monotonic trend detection for recovery trajectory
Bedtime Handling
Bedtime hours past midnight are encoded as 24+ (e.g., 00:30 = 24.5) to avoid discontinuities in variability and midpoint calculations.
Population Norms
- General (age 30-39): Deep 13-23%, REM 20-25%, Efficiency ≥85%, Total 7-9h (Ohayon et al. 2004, Hirshkowitz et al. 2015)
- Post-HSCT: Deep 8-15%, REM 10-18%, Efficiency 70-82% (Jim et al. 2014, Rischer et al. 2020)
- Post-stroke: Deep 10-18%, REM 12-20%, Efficiency 72-85% (Leppavuori et al. 2002, Duss et al. 2018)
Limitations
- N=2 case study: findings are descriptive, not generalizable
- Oura Ring is a consumer wearable, not a polysomnograph; sleep staging has known accuracy limitations
- Different observation windows and data density between patients
- No control for confounders (medications, environment, activity levels)
- Population norms are age-adjusted approximations, not individual-level standards