C-reactive Protein and Risk of OSA in Four US Cohorts

Abstract

Background

Individuals with OSA have elevated levels of inflammatory markers, but no prospective study has examined the role of inflammation in the development of OSA.

Research Question

Is C-reactive protein (CRP) prospectively associated with risk of developing OSA?

Study Design and Methods

We followed 1,882 women from the Nurses’ Health Study (NHS) (2002-2012), 3,854 women from Nurses’ Health Study II (NHSII) (1995-2013), 3,075 men from the Health Professionals Follow-up Study (HPFS) (1996-2012), and 1,919 women and men from the Multi-Ethnic Study of Atherosclerosis (MESA) (2000-2012) who did not have diagnosed OSA at baseline and for whom CRP levels were available. In NHS/NHSII/HPFS, physician-diagnosed OSA was self-reported. In MESA, at-home polysomnography was performed and OSA was identified as an apnea-hypopnea index ≥ 30. Logistic regression was used to estimate the OR for OSA risk according to baseline CRP level, adjusted for multiple inflammation-related factors.

Results

After multivariable adjustment not including BMI, the pooled OR for OSA risk per doubling of baseline CRP level was 1.24 (95% CI, 1.18-1.30). Additional adjustment for BMI substantially attenuated the association (pooled OR, 1.07; 95% CI, 1.01-1.12). The fully adjusted association was consistently stronger in individuals < 55 vs ≥ 55 years of age (P interaction = .01), in individuals with BMI < 25 vs ≥ 25 kg/m² (P interaction = .02), and in pre- vs postmenopausal women (P interaction = .002). CRP was more strongly associated with risk of OSA associated with excessive daytime sleepiness, high airway collapsibility, and low arousal threshold (P heterogeneity < .05).

Interpretation

Higher CRP was prospectively associated with increased OSA risk, particularly among younger individuals, underweight/normal-weight individuals, or premenopausal women. The differential associations by OSA phenotype/endotype suggest possible mechanisms through which inflammation operates to modulate OSA risk. Given our reliance on a single CRP level measured a decade before OSA assessment, future studies with repeated CRP measurements are warranted to confirm these prospective associations.

Repeated arousals from sleep among individuals with OSA lead to increased sympathetic nervous system (SNS) activity and enhanced inflammatory response.¹ Cumulative evidence provides support of OSA as a low-grade chronic inflammatory disease. Cross-sectional studies have documented elevated inflammation in individuals with OSA compared with healthy control subjects,^{2, 3, 4} potentially contributing to cardiometabolic and neurodegenerative comorbidities.^{5, 6, 7, 8} These studies, however, only demonstrated unfavorable inflammatory profiles in individuals with OSA; it remains to be elucidated whether alterations in inflammatory profiles occur prior to or after OSA onset or both.

A growing body of evidence suggests that inflammation may promote development of OSA. Obesity, physical inactivity, and diet each are associated with OSA prevalence and severity as well as with systemic inflammation,^{9, 10, 11, 12} which may be a critical intermediate linking these risk factors to OSA. Experimental studies suggest that inflammation specifically may increase upper airway collapsibility during sleep and predispose to OSA by weakening upper airway muscles, including changing muscle denervation and reducing muscle contractility.^{13, 14, 15, 16, 17} Inflammatory changes in upper airway soft tissues may also increase their volume, reducing airway patency. Additionally, multiple genetic loci identified for OSA traits have been implicated in inflammation-related genes, suggesting a potential causal association between inflammation and OSA.^{18, 19, 20, 21, 22}

Emerging studies suggest that OSA is a highly heterogeneous disorder with variable contributions related to obesity and specific physiological mechanisms,^23,24 such as excessive daytime sleepiness (EDS), airway collapsibility, ventilatory control stability (eg, loop gain), and ventilatory drive (eg, arousal threshold). High airway collapsibility is a marker of reduced airway muscle function or excessive upper airway soft tissue, whereas high loop gain (ie, unstable ventilatory control) and low arousal threshold (ie, high ventilatory drive) are markers for abnormalities in ventilatory control and possibly increased SNS activity. EDS, a key treatment target, is variably present in OSA and may indicate a more severe phenotype. Because proinflammatory cytokines can disrupt sleep and contribute to EDS,^{25, 26, 27, 28} inflammation may be more strongly associated with risk of developing OSA in individuals with EDS. Examining the associations of inflammation with these OSA subtypes may provide novel insights into subgroup heterogeneity.

Despite plausible evidence, no population-based, prospective studies have examined inflammation in relation to future OSA risk. We conducted secondary data analyses in four US cohorts to evaluate the association between circulating levels of C-reactive protein (CRP), a sensitive but nonspecific marker of inflammation, and OSA risk. We posited that higher CRP levels at baseline would be associated with increased risk of developing OSA after a decade or more of follow-up. We further hypothesized that the observed association would be partly confounded by adiposity and particularly strong for OSA subtypes characterized by EDS, high airway collapsibility, high loop gain, or low arousal threshold.

Methods

Study Population

The Nurses’ Health Study (NHS) was established in 1976 recruiting 121,700 US female registered nurses 30 to 55 years of age. The Nurses’ Health Study II (NHSII) was a similar sister cohort initiated in 1989 of 116,429 nurses 25 to 42 years of age. The Health Professionals Follow-up Study (HPFS) included 51,529 US male health professionals 40 to 75 years of age at enrollment in 1986. All cohort participants completed a baseline questionnaire on their lifestyles and disease history, which was prospectively updated by biennial follow-up questionnaires. Between 2000 and 2002, 18,743 NHS women had their blood drawn into heparin tubes and shipped back with an ice pack by overnight courier to the Brigham and Women's Hospital/Harvard Cohorts Biorepository, where it was processed and stored in liquid nitrogen freezers as plasma and blood cell components. The same protocol was used to collect blood from 29,611 NHSII women between 1996 and 1999 and 18,159 HPFS men between 1993 and 1995 (except blood was collected in ethylenediaminetetraacetic acid tubes). The current analysis included 1,882 NHS, 3,854 NHSII, and 3,075 HPFS participants who had CRP assayed from archived blood samples and were free of self-reported clinical OSA diagnoses at blood collection. The average time interval between baseline CRP measurement and questionnaire-based OSA assessment was 11 years in NHS, 16 years in NHSII, and 18 years in HPFS. The study was approved by the institutional review boards of the Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health.

The Multi-Ethnic Study of Atherosclerosis (MESA) is a community-based study of subclinical cardiovascular disease, which recruited 6,814 men and women 45 to 84 years of age in 2000 through 2002 from six US field centers and included 38% whites, 28% blacks, 22% Hispanics, and 12% Chinese Americans.²⁹ Blood samples, medical history, lifestyle factors, and laboratory data were collected at baseline and follow-up examinations. Baseline plasma high-sensitivity CRP data (2000-2002) were available for 6,762 participants; of these, 2,043 completed the MESA Sleep Ancillary Study in 2010 through 2013. After excluding 124 participants who reported physician-diagnosed OSA at MESA examination 2 (2002-2004), 1,919 participants were included in the analysis, with an average interval of 10 years between baseline CRP measurement and sleep study. The institutional review board at each site approved the study, and all participants provided written informed consent.

Assessment of OSA

In 2012, NHS participants were asked about clinically diagnosed OSA and year of first diagnosis. OSA was assessed in a similar manner in NHSII in 2013 and in HPFS in 2012. A previous validation study among 108 NHS/NHSII women showed that for all participants with self-reported OSA, a diagnosis of OSA was confirmed against medical records based on at least one objective method, with 92% diagnosed by in-laboratory polysomnography and 89% meeting criteria for moderate-to-severe disease (apnea-hypopnea index [AHI] ≥ 15).³⁰ Furthermore, OSA prevalence estimates for BMI < 25 vs ≥ 40 kg/m² were 2.5% and 30.8% in NHS/NHSII women and 7.4% and 49.3% in HPFS men, which were highly similar to BMI-specific estimates for moderate-to-severe OSA in general US women and men based on polysomnagraphy.³¹ Taken together, self-reported OSA in NHS, NHSII, and HPFS reliably captured moderate-to-severe OSA cases.

Participants in the MESA Sleep Ancillary Study completed in-home polysomnography and a sleep questionnaire.³² Polysomnography data were generated from a 15-channel monitor (Compumedics Ltd) and scored at a central sleep reading center (Brigham and Women’s Hospital, Boston, MA). Apneas and hypopneas were scored using standard criteria,³² with high inter- and intrascorer reliability (intraclass correlation coefficients > 0.94). AHI was calculated as the average number of all apneas and hypopneas with a 4% oxygen desaturation per hour. To facilitate comparison with self-reported OSA in NHS/NHSII/HPFS, OSA was defined as AHI ≥ 30 in the primary analysis (which resulted in similar age- and BMI-adjusted prevalence estimates). Secondarily, we used AHI ≥ 15 as an alternative OSA definition. We also defined OSA by severity (mild: AHI 5-14, moderate: AHI 15-29, severe: AHI ≥ 30) and according to several physiological endotypes, including upper airway collapsibility, arousal threshold, and loop gain assessed in nonrapid eye movement sleep (see table footnote for derivation of these measures).²⁴

In all four cohorts, sleepiness was assessed by a single question regarding the number of days per week the participants felt overly sleepy during the day. Because of different response categories, EDS was defined as ≥ 4 d/wk in NHS/NHSII/HPFS and ≥ 3 d/wk in MESA. Participants were considered to have habitual snoring if they reported snoring every night or most nights in NHS/NHSII/HPFS or ≥ 3 d/wk in MESA.

Measurement of CRP

CRP assays in NHS/NHSII/HPFS have been described in detail elsewhere.³³ Briefly, plasma aliquots were measured for high-sensitivity CRP using an immunoturbidimetric assay (Denka Seiken) at the Clinical Chemistry Laboratory of Boston Children’s Hospital (Boston, MA). All assays included 10% blinded quality control samples, and consistently had interassay coefficients of variation (CVs) < 10%. To further correct for potential interassay variations, the average batch method was used to recalibrate CRP levels.³⁴ In MESA, high-sensitivity CRP from baseline blood samples was assayed in a central laboratory at University of Vermont (Burlington, VT) by immunonephelometry (BNII nephelometer; Dade Behring Inc). The intraassay CVs ranged from 2.3% to 4.4% and the interassay CVs ranged from 2.1% to 5.7%.²⁹

Statistical Analysis

CRP level was log2-transformed to reduce the right skewness, so that 1-unit increase corresponded to a doubling of the biomarker level. Potential outliers on the logarithmic scale were identified by the generalized extreme Studentized deviate many-outlier procedure and excluded (participants excluded were as follows: one in NHS, zero in NHSII, two in HPFS, and zero in MESA).³⁵

We used multivariable logistic regression to estimate ORs and 95% CIs for OSA risk according to baseline CRP level, adjusted for age, race/ethnicity, sex (MESA only), menopausal status (in women), smoking status, sleep duration, diet quality, physical activity, diabetes, and hypertension. BMI was additionally controlled for in a separate model. We evaluated the CRP level in quintiles and tested the linear trends by using CRP level as a continuous variable. Analyses were conducted separately in each cohort first, followed by a pooled analysis combining individual-level data with additional adjustment for cohort indicators. Overall, the multivariable models fit the data well (eg, P = .59 for the continuous CRP in the pooled fully adjusted model, based on the Hosmer-Lemeshow goodness-of-fit test). Between-cohort heterogeneity was evaluated by testing the cohort-specific estimates associated with the continuous CRP value based on a random-effects meta-analysis approach.³⁶ In MESA, we repeated the analysis using AHI ≥ 15 as an alternative outcome and performed multinomial logistic regression to characterize the associations with OSA severity (AHI < 5, 5-14, 15-29, and ≥ 30). We conducted a sensitivity analysis excluding participants with CRP > 10 mg/mL to evaluate the influence of high CRP levels (participants excluded were as follows: 110 in NHS, 25 in NHSII, 51 in HPFS, and 133 in MESA).

In the pooled sample, multinomial logistic regression was used to obtain separate association estimates for OSA with vs without EDS. To assess whether inflammation was differentially associated with OSA with vs without EDS, we performed a likelihood ratio test comparing the model that allowed the associations between inflammatory biomarkers and OSA to vary by EDS with a model that forced the associations to be the same by EDS.³⁷ We performed similar analyses in MESA to explore whether associations differed by several polysomnography-derived OSA endotypes (upper airway collapsibility, arousal threshold, and loop gain).

Finally, we conducted prespecified subgroup analyses by age, sex, menopausal status in women, and BMI, with statistical significance of subgroup heterogeneity evaluated by likelihood ratio tests of the multiplicative interaction terms. Analyses were performed in SAS 9.4 (SAS Institute).

Results

Consistent across the four cohorts, participants with higher baseline CRP levels tended to have higher BMI, lower physical activity, and worse diet quality and were more likely to report habitual snoring, EDS, and a history of diabetes and hypertension (Table 1). In MESA, participants with higher CRP levels were more likely to be black and have higher AHI.

Table 1.

Characteristics Across Cohort-Specific Quintiles of C-reactive Protein in Participants Without Diagnosed OSA at Baseline

Characteristic	NHS			NHSII			HPFS			MESA
	Q1	Q3	Q5	Q1	Q3	Q5	Q1	Q3	Q5	Q1	Q3	Q5
No. of participants	376	376	376	770	769	770	614	615	614	382	382	382
Age, y	68.8 ± 7.1	68.7 ± 6.7	68.9 ± 6.8	39.8 ± 4.1	41.2 ± 4.2	42.5 ± 4.3	59.4 ± 7.3	61.6 ± 7.4	62.5 ± 7.6	57.8 ± 9.3	59.9 ± 9.4	59.5 ± 9.1
Men	0	0	0	0	0	0	100	100	100	57	54	22
Postmenopausal	100	100	100	2	5	12	…	…	…	80	79	80
Race/ethnicity
White	94	96	95	96	97	96	94	96	95	41	35	33
Asian	1	0	0	1	0	0	0	0	0	23	11	2
Black	0	0	0	0	0	0	0	0	1	19	26	40
Hispanic	0	1	0	1	1	1	…	…	…	17	28	25
Other/unknown	5	2	4	2	1	2	6	4	4	0	0	0
BMI, kg/m2	24.2 ± 3.8	27.0 ± 4.7	29.1 ± 6.2	22.0 ± 2.7	24.3 ± 4.0	29.8 ± 7.0	24.6 ± 2.3	26.1 ± 3.0	27.6 ± 3.9	25.2 ± 3.7	28.1 ± 4.3	32.0 ± 6.2
Current smokers	6	5	5	7	7	12	4	4	9	10	11	11
Sleep duration, ha	7.2 ± 1.1	7.2 ± 1.2	7.3 ± 1.2	7.1 ± 0.9	7.1 ± 1.0	6.9 ± 1.1	7.2 ± 1.0	7.2 ± 1.0	7.3 ± 1.1	7.1 ± 1.4	7.0 ± 1.4	7.2 ± 1.5
Snoringb	24	32	41	16	23	39	31	35	44	54	58	61
EDSc	5	4	7	9	8	15	8	12	15	12	12	17
AHEId	54.8 ± 9.1	52.6 ± 8.4	51.4 ± 8.5	51.3 ± 10.4	49.2 ± 9.8	48.8 ± 9.9	55.5 ± 10.5	53.3 ± 10.4	51.8 ± 10.2	53.2 ± 9.9	50.8 ± 9.4	50.9 ± 9.9
Exercise, MET-hours/wk	20.9 ± 18.8	17.3 ± 14.9	15.4 ± 14.3	23.5 ± 25.9	18.9 ± 19.6	18.5 ± 21.5	37.1 ± 28.6	32.0 ± 26.8	28.5 ± 23.9	30.8 ± 45.0	23.7 ± 32.1	24.1 ± 41.8
Hypertension	45	57	67	4	6	16	20	31	40	32	39	49
Diabetes	7	8	10	0	0	1	3	6	9	8	12	11
AHI	…	…	…	…	…	…	…	…	…	12.1 ± 14.2	14.4 ± 17.2	14.5 ± 15.4

Values are mean ± SD, %, or as otherwise indicated. AHEI = alternate healthy eating index; AHI = apnea-hypopnea index; EDS = excessive daytime sleepiness; HPFS = Health Professionals Follow-up Study; MESA = Multi-Ethnic Study of Atherosclerosis; MET = metabolic equivalent task; NHS = Nurses’ Health Study; NHSII = Nurses’ Health Study II; Q = quartile.

^aBased on self-reports in NHS, NHSII and HPFS and 7-d average actigraphy measures in MESA.

^bDefined as self-reported snoring every night or most nights in NHS/NHSII/HPFS and ≥3 nights/wk in MESA.

^cDefined as self-reported daytime sleepiness ≥4 d/wk in NHS/NHSII/HPFS and ≥5 d/wk in MESA.

^dIncluded all obstructive apneas plus hypopneas associated with 4% oxygen desaturation.

Overall Associations Between CRP and OSA Risk

After 10 to 18 years of follow-up, 75 of 1,882 (4.0%) NHS women, 273 of 3,854 (7.1%) NHSII women, and 400 of 3,075 (13.0%) HPFS men reported a clinical diagnosis of OSA, and 264 of 1,919 (13.8%) MESA participants were found to have OSA as defined by an AHI ≥ 30. After adjusting for other inflammation-related factors except BMI, we observed significant positive associations between baseline CRP levels and OSA risk in all cohorts (Table 2). However, there was significant between-cohort heterogeneity, with the strongest positive association observed in NHSII (P heterogeneity by cohort < .0001). The OR for OSA risk per doubling of CRP was 1.19 (95% CI, 1.02-1.38) in NHS, 1.73 (95% CI, 1.56-1.93) in NHSII, 1.12 (95% CI, 1.05-1.21) in HPFS, and 1.14 (95% CI, 1.04-1.24) in MESA. Further adjustment of BMI substantially attenuated these associations, and only the association in NHSII remained significant. In the pooled analysis including 1,012 incident cases from 10,730 participants, every doubling of CRP was associated with 24% higher odds for OSA (95% CI, 1.18-1.30; P trend < .0001) before adjusting for BMI and 7% higher odds for OSA (95% CI, 1.01-1.12; P trend = .02) after adjusting for BMI.

Table 2.

Circulating Levels of CRP at Baseline and Risk of Developing OSA^a

Study Cohort	Quintiles of CRP					Per Doubling of CRP	P Trend
	Q1	Q2	Q3	Q4	Q5
NHS
Median (mg/L)	0.63	1.50	2.62	4.80	9.26	…	…
Cases/No. of participants	8/376	11/377	17/376	17/377	22/376	…	…
Model 1b	1.00 (ref)	1.42 (0.56, 3.59)	2.11 (0.89, 5.00)	1.96 (0.83, 4.65)	2.58 (1.11, 5.98)	1.19 (1.02, 1.38)	.03
Model 2c	1.00 (ref)	1.11 (0.43, 2.84)	1.43 (0.60, 3.46)	1.08 (0.44, 2.66)	1.18 (0.48, 2.90)	1.00 (0.85, 1.19)	.99
NHSII
Median (mg/L)	0.37	0.61	0.98	1.62	3.28	…	…
Cases/N	19/770	20/773	44/769	81/772	109/770	…	…
Model 1b	1.00 (ref)	1.02 (0.54, 1.92)	2.24 (1.29, 3.90)	4.17 (2.49, 7.00	5.42 (3.24, 9.07)	1.73 (1.56, 1.93)	< .0001
Model 2c	1.00 (ref)	0.88 (0.46, 1.66)	1.67 (0.95, 2.93)	2.46 (1.44, 4.19)	2.09 (1.19, 3.67)	1.32 (1.17, 1.50)	< .0001
HPFS
Median (mg/L)	0.26	0.56	0.93	1.66	3.73	…	…
Cases/N	64/614	71/616	73/615	86/616	106/614	…	…
Model 1b	1.00 (ref)	1.07 (0.75, 1.54)	1.12 (0.78, 1.62)	1.38 (0.96, 1.96)	1.64 (1.16, 2.33)	1.12 (1.05, 1.21)	.002
Model 2c	1.00 (ref)	0.97 (0.67, 1.39)	0.97 (0.67, 1.40)	1.09 (0.76, 1.58)	1.19 (0.83, 1.72)	1.06 (0.98, 1.14)	.16
MESA
Median (mg/L)	0.43	0.96	1.79	3.50	7.96	…	…
Cases/N	40/382	59/387	57/382	53/386	55/382	…	…
Model 1b	1.00 (ref)	1.47 (0.94, 2.29)	1.49 (0.95, 2.34)	1.52 (0.96, 2.42)	2.10 (1.30, 3.39)	1.14 (1.04, 1.24)	.007
Model 2c	1.00 (ref)	1.24 (0.79, 1.94)	1.16 (0.73, 1.84)	0.99 (0.61, 1.61)	1.08 (0.64, 1.82)	0.98 (0.89, 1.09)	.73
Pooled analysisd
Median (mg/L)	0.35	0.70	1.23	2.31	5.52	…	…
Cases/N	140/2,146	153/2,146	211/2,145	251/2,146	257/2,147	…	…
Model 1b	1.00 (ref)	1.16 (0.91, 1.47)	1.63 (1.30, 2.05)	2.13 (1.70, 2.67)	2.53 (2.00, 3.20)	1.24 (1.18, 1.30)	< .0001
Model 2c	1.00 (ref)	1.04 (0.81, 1.32)	1.26 (0.99, 1.59)	1.39 (1.10, 1.76)	1.30 (1.01, 1.68)	1.07 (1.01, 1.12)	.02

CRP = C-reactive protein; HPFS = Health Professionals Follow-up Study; MESA = Multi-Ethnic Study of Atherosclerosis; NHS = Nurses’ Health Study; NHSII = Nurses’ Health Study II; Q = quartile; ref = reference.

^aThe follow-up period was 2002 to 2012 for NHS, 1995 to 2013 for NHSII, 1996 to 2012 for HPFS, and 2000 to 2012 for MESA. The association estimates were ORs and the corresponding 95% CIs.

^bModel 1: adjusted for cohort (pooled analysis), age, race/ethnicity, sex (MESA and pooled analysis), menopausal status in women, smoking, sleep duration, diet quality, physical activity, diabetes, and hypertension.

^cModel 2: model 1 and continuous BMI.

^dBased on random-effects model, there was significant between-study heterogeneity (for the estimates associated with per doubling of CRP, P heterogeneity < .0001 for model 1 and P heterogeneity = .005 for model 2).

Sensitivity Analyses

In MESA, when we considered AHI ≥ 15 as the outcome, the associations were similar to those in the primary analysis (AHI ≥ 30) (e-Table 1). In the multinomial analysis categorizing AHI into four groups, compared with individuals with AHI < 5, the multivariable-adjusted OR associated with a doubling of baseline CRP ranged from 1.20 to 1.28 (e-Table 2). BMI adjustment completely attenuated the associations for moderate and severe OSA; however, a significant association remained for mild OSA (OR, 1.10; 95% CI, 1.01-1.19). A sensitivity analysis excluding participants with CRP levels > 10 mg/L resulted in similar positive association patterns between CRP and OSA risk (e-Table 3).

Associations by OSA Phenotype/Endotype

In the pooled analysis without BMI adjustment, a doubling of CRP was more strongly associated with risk of OSA with EDS (pooled OR, 1.38; 95% CI, 1.24-1.54) than OSA without EDS (pooled OR, 1.21; 95% CI, 1.15-1.27; P heterogeneity = .02) (Table 3). In MESA, the positive associations between CRP and OSA risk were stronger for OSA characterized by high airway collapsibility and low arousal threshold (both P heterogeneity = .04) but did not differ by loop gain (P heterogeneity = .62). Adjustment for BMI attenuated all subtype-specific associations.

Table 3.

Associations of OSA Risk With Every Doubling of Baseline CRP Level by OSA Subtype^a

OSA Endotypes	Casesb	Model 1c		Model 2d
		OR (95% CI)	P heterogeneity	OR (95% CI)	P heterogeneity
Sleepiness			.02		.15
OSA with EDS	174	1.38 (1.24-1.54)		1.15 (1.02-1.30)
OSA without EDS	838	1.21 (1.15-1.27)		1.05 (0.99-1.11)
Collapsibility in NREMe			.04		.06
High collapsibility	62	1.24 (1.04-1.48)		1.13 (0.93-1.37)
Low collapsibility	189	1.08 (0.95-1.23)		0.94 (0.84-1.06)
Loop gain in NREMf			.62		.36
High loop gain	62	1.10 (0.91-1.31)		0.93 (0.76-1.12)
Low loop gain	189	1.14 (1.03-1.27)		1.01 (0.90-1.14)
Arousal threshold in NREMg			.04		.32
Low arousal threshold	63	1.27 (1.06-1.51)		1.07 (0.88-1.30)
High arousal threshold	188	1.07 (0.97-1.19)		0.96 (0.85-1.08)

CRP = C-reactive protein; EDS = excessive daytime sleepiness; NREM = nonrapid eye movement.

^aBased on multinomial logistic regression.

^bThe number of cases included all four cohorts for the analysis of EDS and included the Multi-Ethnic Study of Atherosclerosis only for the analysis of OSA endotypes. Endotype data were not available for 13 OSA cases.

^cModel 1: adjusted for cohort, age, race/ethnicity, sex, menopausal status in women, smoking, sleep duration, diet quality, physical activity, diabetes, and hypertension.

^dModel 2: model 1 and continuous BMI.

^eCollapsibility was quantified by the median level of ventilation during normal ventilatory drive (Vpass). High vs low collapsibility was defined as < 25th percentile vs > 25th percentile of Vpass according to the distribution among OSA cases.

^fLoop gain measured the magnitude of the excess ventilatory drive increase in response to apnea or hypopnea events. High vs low loop gain was defined as > 75th percentile vs < 75th percentile of loop gain value according to the distribution among OSA cases.

^gArousal threshold was measured by the median ventilation level preceding arousals from sleep. Low vs high arousal threshold was defined as < 25th percentile vs > 25th percentile of the arousal threshold value according to the distribution among OSA cases.

Subgroup Associations

In the pooled samples (Table 4), we observed stronger associations among participants < 55 years of age compared with older ones (P interaction = .01) and among participants with BMI < 25 kg/m² compared with heavier ones (P interaction = .02). The association did not differ significantly by sex (P interaction = .18). However, in women, the association was stronger in premenopausal compared with postmenopausal women (P interaction = .002). Similar association patterns across these subgroups were observed within individual cohorts; however, the differences did not reach statistical significance because of smaller sample size (e-Table 4).

Table 4.

Pooled Subgroup Associations Between CRP and OSA Risk by Age, Sex, Menopausal Status, and BMI

Subgroups	Doubling of CRP, OR (95% CI)	P Interaction
Age, y		.01
< 55	1.15 (1.06-1.25)
≥ 55	1.02 (0.95-1.09)
Sex		.18
Men	1.04 (0.97-1.11)
Women	1.11 (1.02-1.21)
Menopausal status		.002
Premenopausal	1.28 (1.13-1.45)
Postmenopausal	0.98 (0.87-1.10)
BMI, kg/m2		.02
< 25	1.17 (1.06-1.30)
≥ 25	1.02 (0.96-1.08)

Adjusted for cohort, age, race/ethnicity, sex, menopausal status in women, smoking, sleep duration, diet quality, physical activity, diabetes, hypertension, and BMI. CRP = C-reactive protein.

Discussion

In four US cohorts, we observed that higher baseline CRP level was associated with increased OSA risk over more than a decade of follow-up. Although adjustment for BMI substantially attenuated the association, the positive relationship between CRP and OSA risk remained, and was significantly stronger in younger than older individuals, in underweight/normal-weight than overweight/obese individuals, and in premenopausal than in postmenopausal women. In addition, CRP was more strongly associated with risk of developing the OSA phenotype characterized by the presence of EDS and the OSA endotypes characterized by high airway collapsibility and low arousal threshold. This prospective evidence highlights the role of inflammation in the pathogenesis of OSA, expanding on prior evidence of increased OSA risk among individuals with inflammation-related conditions (eg, rheumatoid arthritis,³⁸ diabetes³⁹ asthma⁴⁰) or proinflammatory lifestyles (eg, physical inactivity, unhealthy diet^{10, 11, 12}).

As expected, obesity had a large impact on the association between CRP and OSA risk, as also reported in studies that considered elevated CRP as an outcome of OSA.^{2, 3, 4} Obesity may predispose to OSA through mechanical effects that result in upper airway narrowing from pharyngeal fat deposition and from reduction in lung volumes from excess abdominal fat.⁴¹ Because both visceral and subcutaneous fat contribute to chronic systemic inflammation,^42,43 obesity may also influence OSA risk through effects of inflammatory mediators on upper airway structures and/or ventilatory control. It is possible that the mechanical effects of obesity on OSA outweigh the impact by inflammation. This may be why we observed stronger associations with CRP among underweight/normal-weight individuals, for whom the mechanical effects of obesity played a less important role in OSA development, and may also explain why the attenuation in the association after BMI adjustment was less for mild OSA compared with more severe OSA.

In addition to BMI, the association between CRP and OSA risk differed by age and menopausal status in women. Considering that older age, higher BMI, and postmenopausal status are well-established OSA risk factors, our results suggest that inflammation may represent an important pathogenic mechanism among those with earlier OSA onset who did not have traditional OSA risk factors. Of note, although we observed similar differences across subgroups within individual cohorts, we cannot exclude the potential cohort effect that the observed subgroup differences may be driven by the strong association in NHSII, which included younger, leaner, and predominantly premenopausal women. Replication in other populations is warranted.

Etiologic heterogeneity of OSA is increasingly recognized through advances in identification and definition of key phenotypes/endotypes (eg, EDS, airway collapsibility, arousal threshold).²³ The sleep homeostat is partially regulated by inflammatory cytokines,⁴⁴ and EDS, which reflects heightened sleep propensity, may be induced by elevated inflammatory profiles.^{25, 26, 27, 28} Heterogeneity in the association by EDS suggests that inflammation may contribute to development of OSA with concurrent EDS, an OSA phenotype associated with adverse cardiovascular outcomes.⁴⁵ The stronger association of inflammation with the high collapsibility OSA endotype supports prior evidence that upper airway inflammation is associated with muscle denervation and pharyngeal muscle weakness.¹³ Specifically, tumor necrosis factor-alpha has been shown to act on respiratory muscles to narrow the upper airway and enhance its collapsibility during sleep.^{14, 15, 16, 17} Given that CRP is a nonspecific marker that integrates responses to multiple upstream cytokines, research is needed to elucidate specific inflammatory pathways in relation to OSA development. Furthermore, the differential associations by arousal threshold suggest that inflammation may sensitize the SNS, alter the neural circuit controlling sleep architecture, and reduce sleep continuity.^46,47 Interestingly, OSA with a low arousal threshold is more common in women than in men,²⁴ and may relate to insomnia, in which inflammatory processes are also involved.⁴⁴

Our findings are in line with prior studies linking genetic predisposition to inflammation with OSA susceptibility. A candidate gene analysis found that genetic polymorphism in the CRP gene was associated with AHI and almost doubled odds of moderate-to-severe OSA in European Americans after adjusting for BMI.¹⁸ Other genetic association studies identified additional inflammation-related loci for OSA traits, including single nucleotide polymorphisms in the TNFA gene and the PTGER3 gene (encoding the prostaglandin E2 receptor).^19,20 A genome-wide association study in Hispanics/Latinos linked AHI with a locus in the GPR83 gene, which is highly expressed in regulatory T cells to modulate inflammatory responses.²¹ A genome-wide association study of overnight oxyhemoglobin saturation implicated genetic variants in inflammatory pathways mediated by the nucleotide-binding domain and leucine-rich repeat protein-3 inflammasome protein complex form.²² Future research using polygenic risk score and Mendelian randomization approach may provide causal insights into the biomarker associations observed in the current study.

The study strengths included use of both subjectively reported diagnoses and objectively measured OSA, which are complementary (eg, larger sample size vs higher accuracy) and provided consistent associations between CRP and OSA risk. The large sample size from four US cohorts allowed evaluation of subgroup heterogeneity, whereas the endotype data derived from polysomnography shed light on the underlying mechanisms. We were able to control for a large number of inflammation-related factors in the analysis.

Some limitations should be noted. First, because identification of prevalent OSA at baseline was from self-reported diagnosis, the observed association may reflect the impact of data from individuals with undiagnosed OSA (ie, reverse causality). Use of clinically diagnosed OSA in NHS/NHSII/HPFS may lead to overestimation of the association if participants with health conditions causing elevated CRP were more likely to be referred for OSA evaluation. However, associations were similar in magnitude compared with those based on objective measurements in MESA. We used a single measure of CRP at baseline to assess inflammation, which may be inadequate to reflect chronic exposure over time (intraclass correlation coefficient = 0.66 over 4 years) and could bias the estimated associations toward null.⁴⁸ We had limited statistical power for the endotype analysis because the data were only available in MESA, and the observed differences by endotype would not be statistically significant if we accounted for multiple testing using Bonferroni correction (P heterogeneity < .05/4 = .0125). The cutoffs we used to dichotomize endotype were somewhat arbitrary and should be explored in further studies. Finally, findings from the homogeneous NHS/NHSII/HPFS populations (predominantly white health professionals) may have limited generalizability. However, similar patterns observed in the diverse MESA sample suggest that our findings may be extrapolated to broader populations.

In conclusion, higher CRP at baseline was prospectively associated with increased OSA risk more than a decade later, independent of BMI, particularly among younger individuals, underweight/normal-weight individuals, or premenopausal women. The differential associations by EDS, airway collapsibility, and arousal threshold suggest possible mechanisms through which inflammation operates to modulate OSA risk.

Take-home Points.

Study Question: Is elevated inflammation at baseline measured by C-reactive protein (CRP) associated with increased risk of developing OSA?

Results: In four cohorts of US men and women, higher CRP was prospectively associated with increased OSA risk, especially OSA characterized by excessive daytime sleepiness, high airway collapsibility, and low arousal threshold. The observed association was particularly strong among younger women and normal-weight participants.

Interpretation: Elevated inflammation may predispose to OSA development.