A Review of Genetic Association Studies of Obstructive Sleep Apnea: Field Synopsis and Meta-Analysis

Abstract

Study Objectives:

Obstructive sleep apnea (OSA) is a multifactorial disorder with a heritable component. We performed a field synopsis of genetic association studies of OSA to synthesize the available evidence.

Design:

Systematic literature review and meta-analysis.

Setting:

Genetic association studies.

Patients or Participants:

We searched multiple databases to identify studies of non-HLA genetic associations in OSA. We assessed the power of the primary studies to identify odds ratios (OR) in a plausible range and performed random effects meta-analyses for genetic associations investigated by at least 3 studies. We explored the consistency of the findings between population- and family-based studies.

Interventions:

None

Measurements and Results:

We identified a total of 31 population-based case-control studies reporting allele-frequency data on 48 polymorphism-OSA associations. Sample sizes were generally small (median number of cases = 102, 25^th-75^th percentile = 55-151; median number of controls = 79, 25^th-75^th percentile = 58-137), and genetic effects were moderate in magnitude (median OR = 1.15, 25^th-75^th percentile = 0.89-1.40). Studies were severely underpowered to detect ORs as high as 2; only eight comparisons (in 6 studies) had more than 90% power to identify a genetic effect of that magnitude at its current sample size. Four genetic associations had been investigated in ≥ 3 studies: TNFA (-308 A/G) rs1800629, ACE I/D, APOE ε2, and APOE ε4. Only TNFA rs1800629 was significantly associated with OSA under an allele frequency model (3 studies, odds ratio [OR] = 1.82, 95% confidence interval [CI] 1.26-2.61). These results were robust to alternative genetic models; findings for APOE variants were consistent with those from family-based studies.

Conclusions:

The developing field of OSA genetics is currently dominated by small and underpowered investigations. Promising findings regarding TNFA rs1800629 need to be replicated in larger studies using more comprehensive genotyping methods.

Citation:

Varvarigou V; Dahabreh IJ; Malhotra A; Kales SN. A review of genetic association studies of obstructive sleep apnea: field synopsis and meta-analysis. SLEEP 2011;34(11):1461-1468.

INTRODUCTION

Obstructive sleep apnea (OSA) is a complex chronic disorder characterized by repetitive episodes of pharyngeal collapse leading to apneas and hypopneas during sleep. These episodes result in oxygen desaturation, nocturnal catecholamine surges, intrathoracic pressure swings, sleep fragmentation, and excessive daytime sleepiness.¹ OSA is associated with a wide range of adverse clinical outcomes, (cognitive impairment, depression, cardiovascular disease) and has a substantial economic impact on society.² OSA diagnosis and severity assessment is based on the frequency of apnea or hypopnea episodes per hour, i.e., the apnea-hypopnea index (AHI). Because AHI cutoffs in epidemiological investigations conducted to date have been variable, it is hard to directly compare disease prevalence statistics across studies.^3–6 However, in the United States, symptomatic OSA may affect up to 1 in 4 men and 1 in 9 women, for an estimated total of 23 million affected individuals (30-60 years old), approximately half of whom have moderate or severe disease.⁷

There is growing evidence that genetic factors, and their interaction with environmental exposures, influence the development of OSA. Well-established ethnic differences in OSA prevalence and clinical correlates, evidence from family-based and twin studies, as well as the association of specific hereditary craniofacial disorders and the Prader-Willi syndrome with OSA, indicate that the disease may have a genetic basis.^8–11 However, most cases of OSA do not exhibit classical Mendelian patterns of inheritance, suggesting a multi-gene pathogenesis, where many common variants, with small or moderate genetic effects, determine disease heritability. Phenotypes associated with OSA, such as central obesity and “obstructive” craniofacial morphometry, also have a complex genetic basis.¹²

Multiple epidemiologic studies have investigated genetic associations with OSA; however, most studies have been relatively small and possibly underpowered to detect even moderate effect sizes. We performed a systematic review of genetic association studies in OSA, to provide an overview of the field and to assess the design characteristics of published studies.¹³ When possible, we conducted meta-analyses to synthesize the results of studies that had investigated the same polymorphisms to increase precision in the estimation of genetic effects and assess between study heterogeneity.^14,15

METHODS

Search Strategy, Inclusion and Exclusion Criteria

We searched the MEDLINE database (through November 25, 2010), the Human Genome Epidemiology Network (HuGE Net) Literature Finder¹⁶ (through November 14, 2010) and the National Institutes of Health Genetic Association Database¹⁷ (through November 25, 2010) to identify studies reporting on the association of non-HLA genetic variants with obstructive sleep apnea (OSA). We also performed a search for genome-wide association studies using the National Human Genome Research Institute database (through December 19, 2010).¹⁸ For all searches we used combinations of search terms, such as “sleep apnea,” “sleep disordered breathing,” “airway resistance,” “snoring” and “polymorph,*” “genetics,” “variant,” in appropriate combinations for each database; the full search strategy is available from the authors upon request. To increase the yield of our search, we also reviewed the reference lists of all eligible studies and relevant review articles. Studies were considered eligible if they were population-based, used validated genotyping methods to measure genotype or allele frequencies in OSA affected (case) and unaffected (control) individuals, and had an appropriate analytical design (case-control, cohort, or nested case control). The first 300 abstracts were screened jointly by 2 reviewers (VV and IJD) to ensure that uniform criteria were being used; non-overlapping sets of abstracts were screened thereafter.

We accepted the disease definition employed in each article. Studies of hereditary apnea syndromes were excluded, as were studies that did not report genotyping information or those that assessed conditions or diseases other than OSA, such as excessive daytime somnolence, nonspecific sleep disturbances or sleep apnea of non-obstructive etiology, unless data were reported separately for OSA patients. We also excluded case-only designs or genotype prevalence studies (for example, studies of healthy individuals only or individuals drawn randomly from the general population) because they do not inform directly on the relative genotypes (or allele) frequency in OSA affected and unaffected individuals (i.e., they do not provide information for the genetic odds ratio [OR]). We excluded letters, editorials, narrative reviews, and other manuscripts not reporting primary research results. We only considered studies published in English¹⁹ and in full-text form; studies published only in abstract form were excluded since they are often not peer reviewed and generally provide insufficient detail to evaluate critically.

Data Extraction

One author (VV) extracted data from all studies, and another author (IJD) verified all extractions. Discrepancies were resolved by consensus involving a third reviewer (SK). The following information was extracted from each study: author, year, and journal of publication; genes and variants genotyped; ethnicity and age group (adults versus children) of participants; study design (case-control versus cohort); and the genotype and allele frequencies for each variant investigated in cases and controls. For studies included in quantitative analyses we also extracted the following additional information: genotyping method, case definition, case and control selection (for case-control studies) and cohort selection (for cohort studies), mean age, participant sex, body mass index (BMI), AHI, Epworth sleepiness score, systolic and diastolic blood pressure, and neck circumference. This information was extracted separately for affected and unaffected individuals when relevant data were available.

Evidence Synthesis

For each study, we calculated ORs with their variance using an allele frequency contrast when appropriate data were available. For polymorphisms that had been investigated by ≥ 3 studies, we performed meta-analyses using a random effects inverse-variance model (DerSimonian and Laird).²⁰ For meta-analyses of the APOE ε2 and ε4 variants we performed analyses by carrier status (i.e., a dominant model), both for population- and family-based studies. This choice was based on a previous meta-analysis of APOE ε4 variants and OSA,²¹ other systematic reviews of APOE genetics,²² and the availability of data from the primary studies (which typically presented data assuming a dominant mode of inheritance). For all other variants, because the “true” underlying genetic mode of inheritance is not known, we used allele-frequency comparisons as our primary analysis. Alternative analyses under dominant and recessive genetic models were also performed.

Between-study heterogeneity was assessed with Cochran's Q statistic and inconsistency was quantified with the I² index.^23,24 Because of the low power of the Q statistic, we used p_Q < 0.1 as the threshold for statistical significance. I² takes values between 0 and 100% and quantifies the amount of heterogeneity that cannot be explained by sampling variability; higher values indicate greater inconsistency.

Sensitivity Analysis and Assessment of Bias

For polymorphisms that had been investigated by ≥ 3 studies and were considered in quantitative analyses, we performed sensitivity analyses by including studies that identified patients with sleep apnea that did not specifically state that patients had obstructive disease. In addition, we repeated all analyses by excluding the first published study on each association.²⁵ To assess whether the genotype frequencies in control groups were in Hardy-Weinberg equilibrium (HWE) we used an exact test.²⁶ We did not conduct additional exploratory regression analyses, particularly for patient-level factors such as age or sex, to avoid ecological bias and to limit type I error.²⁷ Comparisons between subgroups of studies (for example, between studies with control groups in HWE versus those in violation) were performed using univariate meta-regression with the predictors of interest coded as binary covariates. These analyses are tests of interaction of the genetic effect by study-level characteristics.^28,29

For meta-analyses other than those for polymorphisms in the APOE gene, we used the Harbord modification of the Egger regression-based test to assess whether less precise (smaller) and more precise (larger) studies produced different results.³⁰ For studies of polymorphisms in the APOE gene, we used the original Egger test because they often directly reported ORs with their standard errors (instead of the actual allele frequencies).³¹ These tests are commonly referred to as publication bias tests; however, significant differences between large and small studies could result from true underlying heterogeneity, chance or other systematic differences between studies.^32,33

Power Calculations for Future Studies

To explore whether studies had adequate power to detect low to moderate effect sizes, for each study and each variant for which a 2×2 allele-based table could be constructed, we calculated the number of participants that would be required for 90% power to detect ORs of 1.1, 1.2, 1.5, and 2.0 at alpha = 0.05. Sample size calculations were based on the minor allele frequency for each variant in the controls and the case/control ratio of each study. We used the sample size estimates to calculate how many times larger the “ideal” sample size was compared with the observed sample size of each study.³⁴ This approach avoids the theoretical limitations of retrospective power calculations,³⁵ while providing an indication of the sample requirements to reliably identify plausible genetic effects of moderate magnitude.

Family-Based Studies

For polymorphisms assessed in ≥ 3 population-based genetic association studies, we performed targeted searches to identify family-based studies investigating the same polymorphisms in non-syndromic OSA. Family-based studies were considered separately from population-based studies because of different design and analysis considerations.³⁶ We extracted and tabulated descriptive characteristics of the study design and populations enrolled in these studies. We also qualitatively assessed whether their results were in agreement with those of population-based studies.

Software

All analyses were conducted using Stata version 11.1/SE (Stata Corp., College Station, TX). Statistical significance was defined as a 2-sided P-value < 0.05, for all tests except those for heterogeneity. We did not perform corrections of meta-analytic P-values for multiple comparisons because our analyses were exploratory in nature.³⁷

RESULTS

Population-Based Genetic Association Studies

Our searches identified a total of 1482 citations. Of these, we considered 125 as potentially eligible and reviewed them in full text. We excluded 92 studies and considered 31 population-based studies to be eligible for our main analysis and 2 additional studies in sensitivity analyses (see below). Figure 1 presents the flow of the search strategy along with reasons for exclusion of studies that were reviewed in full text. A list of citations to studies considered in this review is provided in the Supplementary References. In total, the 31 eligible studies provided data on 50 polymorphism-OSA associations (48 for the primary analysis and 2 for the sensitivity analysis). Studies had been published between 1999 and 2010, with a substantial increase in the number of studies occurring in the last few years: studies on 32 of the 50 associations (64%) had been published between 2008 and 2010. All population-based studies had a case-control design and adopted a candidate gene approach; we did not identify any published genome-wide association study. Eligible studies used established genotyping techniques (such as restriction fragment length polymorphism or allele-specific methods) that are expected to have adequate analytic accuracy.

Figure 1.

Search strategy flow. GAS, genetic association study; OSA, obstructive sleep apnea.

Sample sizes were generally small, with a median number of 102 (25^th-75^th percentile = 55-151) affected and 80 (25^th-75^th percentile = 58-137) unaffected individuals per study. The magnitude of the genetic effects was moderate with the median OR in allele frequency contrasts being 1.15 (25^th-75^th percentile = 0.89-1.40). Thirteen of the 48 (27%) associations were statistically significant at the 0.05 level.

Meta-Analyses for Associations Investigated by at least 3 Population Studies

Four polymorphisms had been investigated by ≥ 3 studies: TNFA (-308 A/G) rs1800629 (3 studies, 309 cases/370 controls),^38–40 ACE I/D (6 studies, 342 cases/318 controls),^41–46 APOE ε2 (3 studies, 506 cases/941 controls),^47–49 and APOE ε4 (3 studies, 560 cases/1048 controls).^47,49,50 Details about study design, participant selection methods, and population characteristics for studies included in quantitative analyses are summarized in Supplementary Tables S1 and S2. We found a significant association between TNFA rs1800629 and OSA under an allele frequency comparison, OR = 1.82 (95% CI, 1.26-2.61; P = 0.001) with little evidence of between-study heterogeneity (I²= 28% and p_Q = 0.25). This finding was consistent under different genetic models. All other meta-analyses produced nonsignificant results: OR = 0.83 (95% CI, 0.57-1.20; P = 0.31) for ACE I/D under an allele frequency contrast and OR = 1.39 (95% CI, 0.67-2.90; P = 0.38) for APOE ε2 and OR = 1.73 (95% CI, 0.75-3.96; P = 0.20) for APOE ε4, both under a dominant genetic model. There was some evidence that estimates for smaller studies differed from those of larger studies for ACE I/D (Harbord test, P = 0.08). There was no evidence for such a difference for any of the remaining polymorphisms assessed by meta-analysis (P > 0.3 in all comparisons). In all studies included in meta-analyses of TNFA rs1800629 or ACE I/D the control groups were not in violation of HWE (HWE could not be assessed for most studies included in APOE analyses). There was no indication that the first study on any of the associations we considered produced different results compared to subsequent studies.

Meta-analysis results for population-based studies are summarized in Table 1; Figure 2 presents forest plots for the 4 polymorphisms for which we performed meta-analyses. Results from analyses using alternative genetic models produced qualitatively similar results with allele-frequency comparisons and are summarized in Supplementary Table S3.

Table 1.

Meta-analysis results for gene-disease associations investigated in at least 3 population-based studies

Gene	Variant	Genetic contrast	Studies (cases/controls)	OR (95% CI)	Heterogeneity (pQ; I2)	P-value
TNFA	rs1800629	Allele frequency	3 (309/370)	1.82 (1.26-2.61)	0.25; 28%	0.001
ACE	I/D	Allele frequency	6 (342/318)	0.83 (0.57-1.20)	0.02; 62%	0.31
APOE	ε2	Dominant	3 (506/941)	0.59 (0.21-1.68)	0.004; 82%	0.32
APOE	ε4	Dominant	3 (560/1048)	1.73 (0.75-3.96)	0.008; 80%	0.20

CI, confidence interval; OR, odds ratio.

Figure 2.

Forest plots of meta-analyses for the 4 polymorphism-OSA associations investigated in ≥ 3 population-based studies. Each study is shown by the point estimate of the OR (square proportional to the weight of each study) and 95% CI for the OR (extending lines); the summary OR and 95% CIs by random effects calculations are depicted as a diamonds. Point estimate values > 1 indicate that the minor allele is associated with increased OSA risk. Calculations are based on allele-frequency comparisons for TNFA rs1800629 and ACE I/D and a dominant genetic model for the ε2 and ε4 APOE alleles. CI, confidence interval; OR, odds ratio; OSA, obstructive sleep apnea.

Sensitivity Analysis

Two studies reported on the association of the APOE ε4 allele and sleep disordered breathing (Kadotani et al.⁵¹ and Foley et al.⁵²). Although these studies did not fulfill our primary inclusion criteria, they had been included in a prior meta-analysis of APOE ε4 polymorphisms,²¹ and we considered them in sensitivity analysis.^51,52 Their inclusion did not substantially affect our conclusions: the ε4 allele was not associated with OSA, OR = 1.40 (95% CI, 0.86-2.27; P = 0.18) and there was substantial between-study heterogeneity (p_Q = 0.001; I² = 78%). One study of the ACE polymorphism used controls of the same ethnicity as cases but sampled from a different study base.⁴² Excluding this study did not materially affect the meta-analysis results OR = 0.91 (95% CI, 0.59-1.39).

Power Analyses for Future Studies

We calculated the necessary sample size to attain 90% power to detect genetic effects of plausible magnitude,⁵³ ranging from OR = 1.1 (a small effect size akin to those typically identified in genome-wide association studies) to OR = 2 (a fairly large effect size), based on the case/control ratio and minor allele frequency observed in the primary studies. Figure 3 demonstrates how many times larger each study should be in order to detect plausible ORs with 90% power under different scenarios for the underlying “true” genetic effect. Studies were severely underpowered, particularly for smaller effect sizes. Even for a relatively strong genetic effect (OR = 2) only eight comparisons (in 6 studies) had adequate sample size to attain power of 90%.

Figure 3.

Sample size analyses for population-based OSA candidate gene studies. For each genetic association in our database, we calculated the sample size necessary for 90% power to detect a true genetic effect size of 1.1, 1.2, 1.5, and 2.0 based on the observed minor allele frequency and case/control ratio. The histograms show how many times larger an ideal study (i.e., a study with 90% power to detect the hypothetical true genetic effect) would have to be compared to the observed sample size for a given genetic effect. In each panel, the vertical dashed line indicates the point where the current sample size of a study would be adequate to attain 90% power for a give OR value. Studies to the right of the dashed line would need to have larger sample sizes to attain 90% power. The number of studies that would need to be larger by a given amount of times to attain 90% power is depicted on the y-axis. For example, assuming a true OR of 2, only eight comparisons (in 6 studies) had sample sizes that were adequate to attain 90% power. OSA, obstructive sleep apnea; OR, odds ratio.

Family-Based Studies

Our targeted searches identified 3 family-based studies investigating the polymorphisms assessed by population-based studies included in meta-analyses: 2 studies for APOE ε4, 1 study for APOE ε2, and 1 study for ACE I/D.^54–56 The study on ACE I/D did not provide extractable data comparing individuals affected and unaffected by OSA.⁵⁶ For APOE variants, family- and population-based studies provided consistent results and did not support an association with OSA risk. Specifically, for APOE ε4, the summary OR of the studies by Larkin et al.⁵⁵ and Gottlieb et al.⁵⁴ under a dominant genetic model was 1.03 (95% CI, 0.55-1.91). Similarly, for APOE ε2, the OR was 1.46 (95% CI, 1.02-2.08) in the study by Larkin.⁵⁵

DISCUSSION

We comprehensively reviewed published population genetic association studies in OSA: the literature is growing at an increasing pace, with more than 60% of the associations we reviewed having been published in that last 3 years. Using meta-analysis of allele frequency contrasts we identified a statistically significant association for the TNFA rs1800629 variant, a finding that remained robust under alternative genetic models.

Several lines of evidence suggest that non-syndromic OSA has a substantial hereditary component.¹² A study of OSA susceptibility loci in 66 white pedigrees identified suggestive evidence of linkage with multiple genetic loci. Two important phenotypes associated with OSA, increased AHI and BMI, were found to have multiple genetic determinants, only some of which were common between the two traits.⁵⁷ These results were subsequently confirmed in a linkage study of African American families.⁵⁸ Moreover, a twin study of snoring and daytime sleepiness demonstrated that genetic factors account for 40% of the observed variance in daytime sleepiness, an effect that was in part independent from that of genetic influences on obesity.⁵⁹

We identified only 4 polymorphism-disease associations that have been investigated by at least 3 population studies: TNFA rs1800629, ACE I/D, and the APOE ε2 and ε4 alleles. Only TNFA rs1800629 was significantly associated with OSA. TNFA variants have been associated with disease phenotypes such as ischemic heart disease, heart failure, and chronic obstructive pulmonary disease; results, however, have been inconsistent across studies.^34,60–62 The association of TNFA rs1800629 with OSA suggests the potential existence of common genetic pathways between OSA and these disorders; other variants that have been shown to be associated with these phenotypes could be prioritized for investigation in OSA. However, because our finding on TNFA rs1800629 is based on only 3 studies, it should be viewed with caution pending further validation.

Field synopses using approaches similar to ours have been conducted in Alzheimer disease,⁶³ schizophrenia,⁶⁴ chronic obstructive pulmonary disease,³⁴ as well as cardiovascular diseases, including stroke⁶⁵ and coronary artery disease,⁶⁶ and have been encouraged by the Human Genome Epidemiology Network.¹³ To the best of our knowledge, there has only been one previously published meta-analysis of genetic association studies in OSA.²¹ This work was limited to the APOE ε4 allele and, similar to our analysis, assumed a dominant genetic model for this contrast. Differences in selection criteria and analytical approaches resulted in the inclusion of a somewhat different set of studies in our analysis; however, both systematic reviews produced convergent results regarding this variant.

We did not identify any published genome-wide association study in OSA; all studies considered herein were candidate gene studies which are known to be susceptible to having limited replication validity.²⁵ We note however that primary genotype data and AHI information from a subset of the Sleep Heart Health Study (using the Affymetrix 100K SNP GeneChip) have been deposited on the Database of Genotypes and Phenotypes (dbGAP),⁶⁷ and that additional genome-wide investigations of OSA have been conducted but remain unpublished (Sanjay R. Patel, personal communication). Genome-wide association population-based studies have been conducted in multiple phenotypes that often coexist with OSA, such as hypertension, diabetes, and obesity, and have identified genomic regions strongly and consistently associated with each of these phenotypes.^68–72 Because of the frequent coexistence of these phenotypes and OSA, it may be worthwhile investigating the genetic variants identified by genome-wide association studies of these phenotypes in OSA populations; this strategy can be regarded as a potentially “high-yield” approach to prioritize further research, until genome-wide investigations for OSA become available.⁷³

Our review of OSA genetics, demonstrated that effect sizes are generally small or moderate, with typical ORs ranging between 0.87 and 1.3. Sample sizes were also small, with the majority of studies including fewer than 100 cases and fewer than 60 control individuals. Small sample sizes, combined with genetic effects of low to moderate magnitude, can be expected to limit study power severely. According to our sample size calculations, future studies would have to be orders of magnitude larger to reliably detect effect sizes within the plausible range of ORs (lower than 2).⁵³ Similar observations have been made for genetic associations in other complex diseases⁷⁴ and have spurred the emergence of investigator networks for collaborative research in genetic epidemiology. Such an approach may be a promising way forward for OSA genetics.

Some limitations need to be considered when interpreting our findings. First, because this review was based on published data, information on covariates such as sex, age, disease severity, or details of phenotype ascertainment was often not available. Although this limited our ability to perform subgroup analyses, due to Mendelian randomization, confounding is not a major concern for our study.⁷⁵ Although minor allele frequencies vary between populations of different ethnic descent, genetic effects (i.e., relative allele frequencies between affected and unaffected individuals) are often similar. As such, genetic effect size heterogeneity is rarely induced by racial or ethnic heterogeneity.⁷⁶ Similarly, empirical evidence suggests that few sex-specific effects are consistently replicated in the literature.⁷⁷ We also attempted to assess the sensitivity of our findings to alternative phenotype definitions. Although our results were robust to these sensitivity analyses, study-level differences in phenotype definitions could potentially explain some of the observed between-study heterogeneity. Furthermore, gene-environment interactions were not explored in the primary studies, consequently limiting our ability to explore such effects. Finally, for most associations where we performed meta-analysis, the number of available studies was relatively small: significant findings for TNFA rs1800629 need to be validated in additional population studies, and nonsignificant findings for other variants cannot definitively exclude the presence of small genetic effects.

In summary, the OSA genetic association studies published to date have had small sample sizes that rendered them underpowered to detect small or moderate genetic effects that are the norm in genetic associations of complex diseases.⁵³ Our work provides a comprehensive review of OSA population genetic association studies and, given the growth of the literature that we observed, can serve as a baseline “snapshot” of this rapidly expanding field.

DISCLOSURE STATEMENT

This was not an industry supported study. Dr. Kales has received research funds from Respironics and has consulted for Novartis Pharmaceuticals. Dr. Malhotra has received research and/or consulting income from Philips, Pfizer, Merck, SHC, SGS, Apnex, ApniCure, Ethicon, Medtronic, Cephalon, and Sepracor. The other authors have indicated no financial conflicts of interest.

Table S1.

Characteristics of the studies included in meta-analysis

Author, year (country)	Gene, polymorphism	Study design	Genotype method	Ethnicity	Case Definition [study specific term]	Case Selection	Control Selection (Cohort Selection, for cohort studies)
Khalyfa, 2010 (USA)	TNF-a, G308A (rs1800629)	Case-control	TaqMan	Mixed ethnicity	OSA was defined as absence of airflow with continued chest wall and abdominal movement for duration of at least 2 breaths. Dx based on PSG evidence of AHI≥2/h TST in the presence of snoring during the night and a nadir O2 sat<%. [OSA]	Children with OSA based on PSG criteria.	Age, sex & ethnicity matched with cases children, attending same schools, no h/o snoring, normal PSG.
Riha, 2005 (UK)	TNF-a, G308A (rs1800629)	Case-control	TaqMan	Caucasian	Based on scores derived from the AHI as measured by overnight or home PSG studies and sleepiness as measured by the ESS. [OSAHS]	Patients attending the Scottish Sleep Center with AHI>15.	The cases' siblings and additional randomly selected anonymous UK blood donors.
Bhushan, 2009 (India)	TNF-a, G308A (rs1800629)	Case-control	PCR-RFLP	Mixed ethnicity	Dx of OSA made on the basis of international classification of sleep disorders. From the PSG studies participants with AHI≥10 were diagnosed as having OSA. ESS was used to measure EDS. [OSA]	Obese subjects from Medicine outpatient department of a tertiary care referral hospital of North India.	Obese subjects without OSA matched with cases for age, BMI, and % body fat.
Popko, 2008 (Poland)	TNF-a, G308A (rs1800629)	Case-control	PCR-RFLP	Caucasian	Based on PSG evidence of AHI≥5. [OSAS]	Overweight (BMI>25) or obese (BMI> 30) patients aged 21-77 with newly OSA dx (AHI≥5) referred to the Lung Diseases Clinic in Warsaw.	Non-apneic controls (AHI<5) age and BMI matched with cases from the same clinic.
Zhang, 2000 (China)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	East Asian	Based on clinical exam and PSG findings; OSA was defined as the absence or reduction of airflow in the presence of rib cage and abdominal excursions. AHI from 5-20=mild OSA; 21-50=moderate OSA; AHI>50= severe OSA. [OSAHS]	Hypertensive patients (SBP≥140 mm Hg, DBP≥90 mm Hg) from the Beijing area with AHI≥5 based on PSG findings classified as Mild OSA cases & Moderate-Severe OSA	Hypertensive patients (SBP≥140 mm Hg, DBP≥90 mm Hg) from the same area as cases with AHI<5 based on PSG findings.
Xiao, 1999 (China)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	East Asian	Based on clinical (PE& sleep questionnaire) and laboratory findings (PSG). [OSAS]	OSA patients	Normotensive patients without OSA, IHD or DM.
Yakut, 2010 (Turkey)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	Caucasian	Based on PSG evidence of AHI≥5. [OSAS]	Patients with AHI ≥5 determined by standard PSG in the Thoracic Diseases Department at Uludag University Med. School.	Subjects with AHI <5 determined by standard PSG at the same center.
Ogus, 2010 (Turkey)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	Caucasian	Based on PSG evidence, the AHI was used to determine the severity of OSA (5-15 mild; 15-30 moderate; >30 severe) & PE. [OSAS]	Unrelated Turkish patients with OSA dx determined by PSG at the Department of Chest Diseases of Alkdeniz University Med. School.	Healthy controls age-matched to the cases.
Barcelo,2001 (Spain)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	Caucasian	Based on PSG evidence of AHI>20 plus daytime somnolence. [OSAS]	Male subjects, <65 years old, with OSA dx, based on clinical and laboratory findings from the Sleep Unit of Hospital Universitari Son Dureta.	Healthy, non-smoking, non-obese volunteers, matched for sex and age with no FH of CVD or DM, not on any medication, with OSA diagnosis excluded clinically; (no snoring, or EDS).
Benjamin, 2008 (UK)	ACE, I/D I (490bp) D (190bp)	Case-control	AS-PCR	Caucasian	Based on clinical findings (EDS, h/o snoring and/or nocturnal apneas) and positive laboratory findings (on a limited channel home sleep study; ≥10 events/h of 4% O2 dip rate). [OSAHS]	Patients 18-80 years old with symptoms suggestive of OSAHS from a SDB clinic in the UK, who agreed to have a home sleep study, excluding those with sleepiness from other causes.	Healthy, non-sleepy controls selected from hospital staff from the same geographical population, ethnicity matched with cases.
Kalra, 2008 (USA)	ApoE, ε2, rs7412	Case-control	OLA-PCR	Caucasian	Based on PSG evidence of >1 obstructive apnea (OA) or obstructive hypopnea (OH) episodes per hour of sleep. [OSA]	Children with OSA based on PSG evidence, 2-21 years of age, with no genetic syndromes, current or past tracheostomy, h/o airway reconstruction, and/or neuromuscular disorders.	Race and gender matched control group from a population-based cohort of children enrolled in the PSDS.
Saarelainen, 1998 (Finland)	ApoE ε2, rs7412	Case-control	Isoelectric focusing, cysteamine treatment and immunoblotting	Caucasian	Based on PSG evidence of AHI≥5 events/h. [OSA]	Patients with established OSA dx, without neurological or cerebrovascular diseases.	Random population sample
Cosentino, 2008 (Italy)	ApoE ε2, rs7412 ApoE ε4	Case-control	PCR-RFLP	Caucasian	Based on clinical findings and PSG evidence of AHI ≥15 events/h. [OSA]	OSA patients diagnosed following standard clinical/laboratory criteria with no h/o stroke, traumatic brain injury, or other neurological condition from the Sleep Center of Troina, Italy	Spouses or other significant relatives of the spouses of patients such as siblings, without any symptoms of OSA (such as EDS, snoring, craniofacial malformation, HTN)
Gozal, 2007 (USA)	ApoE ε4	Case-control	PCR	Mixed ethnicity	Based on PE, a validated questionnaire completed by the children's parents, and PSG evidence. OSA: absence of airflow with continued chest wall & abdominal movement for duration of ≥2 breaths. Obstructive apnea index (AI): number of apneas per hour of TST. Diagnostic criteria for OSA: an obstructive apnea index greater than 1/h TST and/or an obstructive AHI > 2/h TST with a nadir O2 saturation value of at least <92%. [OSA]	Children with an obstructive apnea index > 1/h TST and/or an obstructive AHI greater than 2/h TST with a nadir O2 saturation value of at least <92%. Children fulfilling overweight or obesity criteria were excluded.	Non-snoring children with an obstructive AHI≤1/h TST. Children fulfilling overweight or obesity criteria were excluded.
Kadotani, 2001 (USA)	ApoE, ε4	Cohort	PCR-RFLP	Caucasian	SDB; based on PSG findings of AHI≥ 15. [SDB]	NA	Participants were selected from an ongoing longitudinal cohort study of sleep disorders the SHHS that began in 1989.
Foley, 2001 (USA)	ApoE, ε4	Cohort	NR	East Asian	SDB; based on PSG findings of AHI≥ 15. [SDB]	NA	Japanese-American participants aged 79-97 years, evaluated for SDB as part of a recent follow-up examination of participants in the Honolulu-Asia Aging Study of dementia that began in 1991.
Gottlieb, 2004 (USA)	ApoE, ε4	Cohort	NR	88% Caucasian	OSAH of AHI≥15. [OSAHS]	NA	1775 participants of the SHHS
Larkin, 2006 (USA)	ApoE, ε2, ε4	Cohort	PCR-RFLP	Caucasian & African American	Laboratory evidence of OSA diagnosis (AHI>20). [OSA]	NA	Participants of the CFSC, a longitudinal genetic epidemiology study of OSA in families.

AHI, apnea hypopnea index; AS-PCR, allele specific-PCR; CFSC, Cleveland Family Study Cohort; CVD, cardiovascular disease; DBP, diastolic blood pressure; DM, diabetes mellitus; dx, diagnosis; EDS, excessive daytime sleepiness; FH, family history; h/o, history of; HTN, hypertension; IHD, ischemic heart disease; NA, not applicable; OLA-PCR, oligonucleotide ligation-PCR; OH, obstructive hypopnea; OSA, obstructive sleep apnea; OSAHS, obstructive sleep apnea/hypopnea syndrome; OSAS, obstructive sleep apnea syndrome; PE, physical exam; PCR, polymerase chain reaction; PSG, polysomnography; PSDS, Princeton school district study; RFLP, restriction fragment length polymorphism; SBP, systolic blood pressure; SDB, sleep disordered breathing; SHHS, sleep heart health study; SpO₂, arterial oxygen saturation; TST, total sleep time.

Table S2A.

Characteristics of case-control studies included in primary and sensitivity analyses

Author, Year (Country)	CASES
	Age mean [±SD]	Gender %M	BMI mean [SD]	AHI mean [±SD]	ESS mean [±SD]	SBP/DBP mean [±SD]	NC (cm) mean [SD]
TNF-a, G308A (rs1800629)
Khalyfa, 2010 (USA)	7.2 [0.2]	50	NR	8.9 [2.7]	6.5 [1.2]	NR	NR
Riha, 2005 (UK)	52 [9]	81	30 [6]	NR	NR	138 [18]/84 [12]	41 [4]
Bhushan, 2009 (India)	46 [18]	81	31.5 [4]	48 [25]	14.7 [5.2]	135 [16]/90 [11]	39.6 [4]
Popko, 2008 (Poland)	NR	72,5	NR	NR	NR	NR	NR
ACE, I/D
Zhang, 2000 (China)	MO 53 [12]	MO 85	MO 27 [3]	MO 11 [9]	NR	MO 147 [22]/92 [14]	NR
	MSO 55 [11]	MSO 85	MSO 30 [3]	MSO 48 [22]	NR	MSO 153 [24]/94 [14]	NR
Xiao, 1999 (China)	51 [11]	90	NR	NR	NR	NR	NR
Yakut, 2010 (Turkey)	50 [11]	83	31 [4]	NR	NR	NR	NR
Ogus, 2010 (Turkey)	51 [10]	91	31 [6]	24 [18]	11 [5]	NR	42 [3]
Barcelo, 2001 (Spain)	50 [1]	100	33 [0.6]	55 [3]	NR	140 [2]/87 [1]	NR
Benjamin, 2008 (UK)	48 [11]	81	38 [8]	42 [26]	15 [6]	140 [24]/83 [12]	NR
ApoE, ε2
Kalra, 2008 (USA)	13 [5]	62	30 [11]	11 [4]	NR	NR	NR
Saarelainen, 1998 (Finland)	53 [NR]	91	NR	38 [NR]	NR	NR	NR
Cosentino, 2008 (Italy)	59 [9]	67	36 [7]	45.5 [27]	13 [6]	NR	NR
ApoE, ε4
Gozal, 2007 (USA)	6 [0.3]	54	17 [0.4]	9 [2]	NR	NR	NR
Cosentino, 2008 (Italy)	59 [9]	67	36 [7]	45.5 [27]	13 [6]	NR	NR
Author, Year (Country)	CONTROLS
	Age mean [±SD]	Gender %M	BMI mean [SD]	AHI mean [±SD]	ESS mean [±SD]	SBP/DBP mean [±SD]	NC (cm) mean [SD]
TNF-a, G308A (rs1800629)
Khalyfa, 2010 (USA)	7.2 [0.3]	50	NR	0.5 [0.2]	4.5 [0.5]	NR	NR
Riha, 2005 (UK)	51 [10]	50	27 [5]	NR	NR	130 [17]/82 [12]	37 [4]
Bhushan, 2009 (India)	44 [10]	63	31 [4]	2.8 [1.7]	8 [3]	133 [18]/86 [11]	38 [3]
Popko, 2008 (Poland)	NR	51	NR	NR	NR	NR	NR
ACE, I/D
Zhang, 2000 (China)	52 [12.5]	58	27 [5]	1.2 [1.5]	NR	123 [11]/80 [9]	NR
Xiao, 1999 (China)	31 [7]	60	NR	NR	NR	NR	NR
Yakut, 2010 (Turkey)	50 [10]	70	29 [5]	NR	NR	NR	NR
Ogus, 2010 (Turkey)	60 [10]	46	NR	NR	NR	NR	NR
Barcelo, 2001 (Spain)	49 [1]	100	26 [0.6]	NA	NR	117 [2]/72 [1]	NR
Benjamin, 2008 (UK)	SC: 48 [7]	SC: 62	SC: 32 [5]	SC: 5 [3]	SC: 13 [5]	SC: 130 [12]/82 [11]	NR
	HC: 40 [12]	HC: 54	HC: NR	HC: NR	HC: 3 [2]	HC: NR
ApoE, ε2
Kalra, 2008 (USA)	15 [2]	58	27 [3]	NR	NR	NR	NR
Saarelainen, 1998 (Finland)	54 [NR]	78	NR	NR	NR	NR	NR
Cosentino, 2008 (Italy)	58 [10]	64.5	30 [5]	NR	NR	NR	NR
ApoE, ε4
Gozal, 2007 (USA)	6 [0.3]	55	17 [1]	0.8 [0.3]	NR	NR	NR
Cosentino, 2008 (Italy)	58 [10]	64.5	30 [5]	NR	NR	NR	NR

AA, African American; AHI, apnea-hypopnea index; BMI, body mass index; DBP, diastolic blood pressure; ESS, Epworth sleepiness scale; HC, healthy controls; MO, mild OSA; MSO, moderate-severe OSA; NC, neck circumference; NA, not applicable; NR, not reported; OSA, obstructive sleep apnea; SC, sleepy controls; SBP, systolic blood pressure; SD, standard deviation; %M, percentage male.

Table S2B.

Characteristics of cohort studies included in primary and sensitivity analyses

Author, year (Country)	At Least One Variant Allele
	Age mean [±SD]	Gender %M	BMI mean [SD]	AHI mean [±SD]	ESS mean [±SD]	SBP/DBP mean [±SD]	NC (cm) mean [SD]
ApoE, ε4
Kadotani, 2001 (USA)	49 [9]	64	30 [6]	6.5 [0.6]	NR	NR	NR
Foley, 2001 (USA)	NR	NR	NR	NR	NR	NR	NR
Gottlieb, 2004 (USA)	72 [10]	45	27 [5]	NR	NR	NR	37 [4]
ApoE, ε2, ε4
Larkin, 2006 (USA)	NR	NR	NR	NR	NR	NR	NR
Author, year (Country)	No Variant Allele
	Age mean [±SD]	Gender %M	BMI mean [SD]	AHI mean [±SD]	ESS mean [±SD]	SBP/DBP mean [±SD]	NC (cm) mean [SD]
ApoE, ε4
Kadotani, 2001 (USA)	49 [8]	56	30 [6]	4.8 [0.3]	NR	NR	NR
Foley, 2001 (USA)	NR	NR	NR	NR	NR	NR	NR
Gottlieb, 2004 (USA)	71 [11]	45	28 [5]	NR	NR	NR	38 [4]
ApoE, ε2, ε4
Larkin, 2006 (USA)	NR	NR	NR	NR	NR	NR	NR

AA, African American; AHI, apnea-hypopnea index; BMI, body mass index; DBP, diastolic blood pressure; ESS, Epworth sleepiness scale; HC, healthy controls; MO, mild OSA; MSO, moderate-severe OSA; NC, neck circumference; NA, not applicable; NR, not reported; OSA, obstructive sleep apnea; SC, sleepy controls; SBP, systolic blood pressure; %M, percentage male.

Table S3.

Meta-analysis results for gene-disease associations investigated in at least 3 studies

Gene	Variant	Genetic contrast	Studies (cases/controls)	OR (95% CI)	P-value	Heterogeneity (pQ; I2)
TNFA	rs1800629	Dominant	4 (369/450)	1.76 (1.18-2.62)	0.006	0.21; 33%
TNFA	rs1800629	Recessive	3 (309/370)	2.01 (0.88-4.60)	0.098	0.55; 0%
ACE	I/D	Dominant	6 (342/363)	0.76 (0.49-1.19)	0.23	0.21; 31%
ACE	I/D	Recessive	6 (342/363)	0.81 (0.41-1.61)	0.55	0.01; 67%

CI, confidence interval; OR, odds ratio.