Atlas/Phenotypes

Cross-domain OSA phenotypes

Three near-independent axes. Eight octant phenotypes.

Physiologic severity, symptom burden, and comorbidity burden describe complementary dimensions of OSA. Median-splitting each cohort score produces eight near-balanced groups and classifies all 70,880 people in the shared phenotype cohort.

Axis 1

Physiologic severity

Higher hypoxemia burden

Sleep-study physiology including oxygen desaturation and saturation burden.

score_c1 cut 1.002

Axis 2

Symptom burden

Higher symptom burden

Sleepiness, insomnia, functional impairment, mood screening, and related symptoms.

score_c2 cut 1.004

Axis 3

Comorbidity burden

Higher coded burden

A broad coded-disease gradient that also reflects healthcare contact and record completeness.

score_c3 cut 0.996

Cohort-derived medians are descriptive, not clinical thresholds, and do not transfer unchanged to another sample.

Octant explorer

What characterizes each phenotype?

Select an octant, then search or filter 26 aggregate measures to inspect its distribution and coverage.

□□□

Selected phenotype

Mild All

Below the cohort median on physiologic severity, symptoms, and comorbidity burden.

Physiologic severityBelow median

Symptom burdenBelow median

Comorbidity burdenBelow median

Phenotype cohort: 9,249 (13.0%)
Median AHI: 13.9 events/h
Median PheCodes: 16

Aggregate phenotype profile26 of 26 measures

Find a measure

physiologic

ODI 4% (events/h)

14.1 events/h

-0.42 SDfrom the containing cohort

Mean ± SD: 14.1 ± 12.1
Median [Q1–Q3]: 10.4 [5.90–18.3]
Cohort mean: 23.7

physiologic

minimum SpO2 (%)

84.4 %

+0.60 SDfrom the containing cohort

Mean ± SD: 84.4 ± 5.03
Median [Q1–Q3]: 85.0 [82.0–88.0]
Cohort mean: 78.9

physiologic

mean SpO2 (%)

94.3 %

+0.56 SDfrom the containing cohort

Mean ± SD: 94.3 ± 1.07
Median [Q1–Q3]: 94.0 [94.0–95.0]
Cohort mean: 92.9

physiologic

T90 (fraction)

0.7%

-0.50 SDfrom the containing cohort

Mean ± SD: 0.7% ± 1.1%
Median [Q1–Q3]: 0.3% [0.0%–1.0%]
Cohort mean: 8.7%

symptom

Epworth total

5.59 points

-0.51 SDfrom the containing cohort

Mean ± SD: 5.59 ± 3.53
Median [Q1–Q3]: 5.00 [3.00–8.00]
Cohort mean: 8.20

symptom

Insomnia Severity Index

8.20 points

-0.65 SDfrom the containing cohort

Mean ± SD: 8.20 ± 4.25
Median [Q1–Q3]: 8.00 [5.00–11.0]
Cohort mean: 12.2

symptom

FOSQ impairment

4.59 points

-0.58 SDfrom the containing cohort

Mean ± SD: 4.59 ± 3.78
Median [Q1–Q3]: 4.00 [1.00–7.00]
Cohort mean: 8.39

symptom

PHQ-2

0.63 points

-0.46 SDfrom the containing cohort

Mean ± SD: 0.63 ± 0.99
Median [Q1–Q3]: 0.00 [0.00–1.00]
Cohort mean: 1.38

symptom

STOP score

1.76 points

-0.28 SDfrom the containing cohort

Mean ± SD: 1.76 ± 0.98
Median [Q1–Q3]: 2.00 [1.00–3.00]
Cohort mean: 2.02

comorbidity

distinct phecodes

17 count

-0.76 SDfrom the containing cohort

Mean ± SD: 17 ± 10
Median [Q1–Q3]: 16 [9–25]
Cohort mean: 42

comorbidity

obesity

46.6%

-0.33 SDfrom the containing cohort

Count / denominator: 4,308 / 9,249
Cohort proportion: 62.6%

comorbidity

hyperlipidemia

41.8%

-0.29 SDfrom the containing cohort

Count / denominator: 3,868 / 9,249
Cohort proportion: 56.3%

comorbidity

hypertension

27.9%

-0.31 SDfrom the containing cohort

Count / denominator: 2,581 / 9,249
Cohort proportion: 43.3%

comorbidity

impaired fasting glucose

31.7%

-0.35 SDfrom the containing cohort

Count / denominator: 2,932 / 9,249
Cohort proportion: 49.3%

comorbidity

anxiety disorders

22.3%

-0.39 SDfrom the containing cohort

Count / denominator: 2,060 / 9,249
Cohort proportion: 41.7%

comorbidity

GERD

15.9%

-0.39 SDfrom the containing cohort

Count / denominator: 1,471 / 9,249
Cohort proportion: 34.6%

external

index AHI (events/h)

18.4 events/h

-0.37 SDfrom the containing cohort

Mean ± SD: 18.4 ± 14.2
Median [Q1–Q3]: 13.9 [8.50–23.6]
Cohort mean: 27.2

demographic

age

45.6 years

-0.37 SDfrom the containing cohort

Mean ± SD: 45.6 ± 12.4
Median [Q1–Q3]: 44.0 [36.0–54.3]
Cohort mean: 50.9

demographic

BMI

33.5 kg/m2

-0.16 SDfrom the containing cohort

Mean ± SD: 33.5 ± 7.66
Median [Q1–Q3]: 32.0 [28.0–37.7]
Cohort mean: 34.8

demographic

female

29.8%

-0.29 SDfrom the containing cohort

Count / denominator: 2,077 / 6,980
Cohort proportion: 44.2%

external

AHI < 5 (no OSA)

0.5%

+0.01 SDfrom the containing cohort

Count / denominator: 47 / 9,249
Cohort proportion: 0.5%

external

AHI 5-15 (mild)

53.3%

+0.25 SDfrom the containing cohort

Count / denominator: 4,931 / 9,249
Cohort proportion: 41.0%

external

AHI 15-30 (moderate)

30.1%

+0.07 SDfrom the containing cohort

Count / denominator: 2,783 / 9,249
Cohort proportion: 27.2%

external

AHI 30+ (severe)

16.1%

-0.33 SDfrom the containing cohort

Count / denominator: 1,488 / 9,249
Cohort proportion: 31.3%

follow up

incwas follow-up

1.32 years

-0.08 SDfrom the containing cohort

Mean ± SD: 1.32 ± 1.08
Median [Q1–Q3]: 1.07 [0.50–1.91]
Cohort mean: 1.42

utilisation

baseline encounter rate

6.00 encounters/year

-0.30 SDfrom the containing cohort

Mean ± SD: 6.00 ± 6.34
Median [Q1–Q3]: 4.04 [2.14–7.44]
Cohort mean: 8.20

CPAP treatment documentation

Documented PAP setup by phenotype

This section summarizes documented setup signals in the retained CPAP extract. It does not estimate true treatment initiation, uptake, adherence, or effectiveness.

documented new setup after index through 90 daysPhenotype members whose existing diagnosis, insurance, and death censor reaches index plus 90 days.The 90-day view is an interface default, not a protocol-designated primary horizon.

□□□

Selected phenotype

Mild All

90 days view, 2,599 observable through this window

Documented PAP setup signal37.7%979 of 2,599

Retained CPAP record45.9%1,194 of 2,599

Adherence data available94.0%920 of 979

Usage measure available94.0%920 of 979

Outcome not released

Adherent percentage unavailable

No adherence outcome estimate is released because the source field's exact algorithm and numeric coding are not authoritatively confirmed.

Distribution not released

Usage distribution unavailable

The released aggregate contains no mean or distribution; its averaging denominator, missing-night handling, and setup-based first-90-day window are unconfirmed.

All-phenotype comparison for 90 days. Availability percentages use documented setups as their denominator.
Phenotype	Observable	Documented setup	Retained record	Adherence data	Usage data
	2,599	37.7%979 / 2,599	45.9%1,194 / 2,599	94.0%920 / 979	94.0%920 / 979
	2,442	47.4%1,158 / 2,442	57.0%1,393 / 2,442	94.7%1,097 / 1,158	94.7%1,097 / 1,158
	2,393	44.7%1,070 / 2,393	53.7%1,284 / 2,393	95.0%1,016 / 1,070	95.0%1,016 / 1,070
	3,043	37.9%1,152 / 3,043	48.8%1,486 / 3,043	92.7%1,068 / 1,152	92.7%1,068 / 1,152
	2,578	54.0%1,391 / 2,578	64.1%1,653 / 2,578	95.0%1,321 / 1,391	95.0%1,321 / 1,391
	3,122	45.8%1,430 / 3,122	58.2%1,816 / 3,122	92.4%1,322 / 1,430	92.4%1,322 / 1,430
	3,145	44.4%1,396 / 3,145	54.6%1,717 / 3,145	93.6%1,306 / 1,396	93.6%1,306 / 1,396
	3,049	52.7%1,608 / 3,049	65.0%1,981 / 3,049	93.3%1,500 / 1,608	93.3%1,500 / 1,608
ALLAll phenotypes	22,371	45.5%10,184 / 22,371	56.0%12,524 / 22,371	93.8%9,550 / 10,184	93.8%9,550 / 10,184

Definitions and interpretation limits

Documented setup is the earliest parseable setup value in the retained extract, not a confirmed lifetime first PAP initiation.

A retained CPAP record contains at least one documented setup or usage signal. An absent row may mean no signal or no capture; it is not a no-CPAP classification.

Fixed-window denominators require observation through the full window using an existing diagnosis, insurance, and death censor; that boundary is not confirmed as PAP data-capture end.

Adherence and usage coverage describe whether a source field is present among documented setups, not adherence or treatment effectiveness.

Protected adherence- and usage-availability cells are null and shown as Suppressed (<11), never as zero.

Cluster comparison

Compare all eight phenotypes in native units

Blue cells are below and orange cells above the containing cohort. Select any column to update the profile.

− Lower than cohortHigher than cohort +

Aggregate cluster measures. Cell text is in native units; color is the standardized difference from the containing cohort.
Measure
ODI 4% (events/h)events/h
minimum SpO2 (%)%
mean SpO2 (%)%
T90 (fraction)proportion
Epworth totalpoints
Insomnia Severity Indexpoints
FOSQ impairmentpoints
PHQ-2points
STOP scorepoints
distinct phecodescount
obesityproportion
hyperlipidemiaproportion
hypertensionproportion
impaired fasting glucoseproportion
anxiety disordersproportion
GERDproportion
index AHI (events/h)events/h
ageyears
BMIkg/m2
femaleproportion
AHI < 5 (no OSA)proportion
AHI 5-15 (mild)proportion
AHI 15-30 (moderate)proportion
AHI 30+ (severe)proportion
incwas follow-upyears
baseline encounter rateencounters/year

Standardized difference means group estimate minus cohort estimate, divided by the cohort SD. Each octant is part of that cohort; this is not a two-independent-group standardized mean difference. Coverage is shown on every profile card.

Score dependence check

Pairwise Spearman correlations

Values near zero indicate that the three cohort scores capture largely distinct dimensions.

Pairwise Spearman correlation matrix for the three phenotype scores.
Axis	Physiologic severity	Symptom burden	Comorbidity burden
Physiologic severity	1.00	0.02	0.04
Symptom burden	0.02	1.00	-0.01
Comorbidity burden	0.04	-0.01	1.00

Supplemental phenotype-construction figures

The data-driven comparison above is the primary cluster view. These source figures remain available for publication context and full-resolution download.

Octant construction from median splits of three nearly independent phenotype scores — The three score planes are nearly uncorrelated, yielding eight groups of similar size and complete cohort coverage. Open full resolution ↗

Heatmap of the aggregate clinical signature for each octant phenotype — The source clinical-signature figure is retained as a supplement; the interactive table above includes the expanded 26-measure export. Open full resolution ↗

Interpretation guardrails

A research taxonomy—not a clinical diagnostic rule

Continuous gradients

The physiology, symptom, and comorbidity octants are regions of continuous gradients, not discovered biological subpopulations.

Coverage varies

Measures use their available aggregate denominators. Follow-up and encounter metrics cover the 25,380-person IncWAS spine; age, BMI, and sex cover 53,012 people; several symptom scales are incomplete.

Adjusted associations

Octants are strongly structured by age and sex. M4 models adjust for both, but unadjusted cluster and curve comparisons are neither causal nor clinical treatment effects.

CPAP documentation limits

Released setup and source-field coverage describe the retained extract, not true treatment initiation or adherence. An absent row cannot distinguish no signal from no capture; adherence outcomes and usage distributions remain unavailable.

Race/ethnicity withheld

Race/ethnicity is included in M4 adjustment, but no octant breakdown is published because one cell falls below the disclosure threshold and the breakdown is not approved for release.

Sparse-data and PH checks

Eighty of 168 panels are flagged EPV < 10. Proportional-hazards diagnostics were not computed, so unflagged panels should not be labeled stable.

Phenotype outcomes

Continue to the incident-outcome panels

Search all 168 octant–outcome models and open the interactive cumulative-incidence curves on their dedicated page.

Open outcome panels →