Obstructive sleep apnea and insulin resistance in children with obesity

Free AccessScientific InvestigationsObstructive sleep apnea and insulin resistance in children with obesity Rasintra Siriwat, MD, Lu Wang, MS, Vaishal Shah, MD, MPH, Reena Mehra, MD, MS, FCCP, FAASM, Sally Ibrahim, MD, FAAP, FAASM Rasintra Siriwat, MD Phramongkutklao Hospital, Bangkok, Thailand; Search for more papers by this author , Lu Wang, MS Sleep Disorders Center, Neurological Institute, Cleveland, Ohio; Search for more papers by this author , Vaishal Shah, MD, MPH Sleep Disorders Center, Neurological Institute, Cleveland, Ohio; Search for more papers by this author , Reena Mehra, MD, MS, FCCP, FAASM Sleep Disorders Center, Neurological Institute, Cleveland, Ohio; Search for more papers by this author , Sally Ibrahim, MD, FAAP, FAASM Division of Pulmonary, Allergy and Sleep Medicine, Rainbow Babies and Children's Hospital of University Hospitals, Cleveland, Ohio Search for more papers by this author Published Online:July 15, 2020https://doi.org/10.5664/jcsm.8414Cited by:8SectionsAbstractPDF ShareShare onFacebookTwitterLinkedInRedditEmail ToolsAdd to favoritesDownload CitationsTrack Citations AboutABSTRACTStudy Objectives:Because existing data investigating obstructive sleep apnea (OSA) and insulin resistance (IR) are inconsistent, we examine OSA and IR in a pediatric obesity clinic.Methods:Children (2–18 years) in the obesity clinic (2013–2017) undergoing polysomnography (PSG), anthropometric measurements, and fasting laboratory tests were included. Linear regression assessed OSA defined by the obstructive apnea-hypopnea index (oAHI) with the homeostatic model assessment of insulin resistance (HOMA-IR). Secondary aims assessed oxygen desaturation index (ODI) and age interactions with HOMA-IR. Logistic regression models and receiver operating characteristic analysis were performed to investigate optimal oAHI and ODI cutoffs relative to HOMA-IR ≥ 3.Results:Eighty children were included (mean age, 11.4 ± 4.0 years; 56% female; 46% Caucasian; median body mass index [BMI], 34.6 kg/m2 [interquartile ratio, 29.9–40.1], median BMI z-score, 2.5 [interquartile ratio, 2.3–2.8); 46% with oAHI ≥ 5 events/h. HOMA-IR was higher in the OSA group (oAHI ≥ 5 events/h): 5 vs 3.8 (P = .034). After adjustment for sex, race, and BMI z-score, oAHI ≥ 5 events/h retained significance with HOMA-IR (P = .041). HOMA-IR increased in older children (age ≥ 12 years) when adjusting for waist circumference z-score and waist-height ratio (statistical interaction, P = .020 and .034, respectively). Receiver operating characteristic showed optimal cut points of oAHI and ODI for predicting significant IR 4.9 (area under the curve, 0.70; 95% confidence interval, 0.57–0.83; sensitivity, 0.76; specificity, 0.66) and 4.6 (area under the curve, 0.68; 95% confidence interval, 0.55–0.80; sensitivity, 0.70; specificity, 0.67), respectively.Conclusions:In a clinic-based pediatric cohort with obesity, OSA is associated with increased IR even after adjusting for confounders including obesity defined by the BMI z-score. Age ≥ 12 years was associated with AHI relative to IR after adjustment for waist circumference z-score and waist-height ratio. Significant IR could be discriminated by oAHI ≥ 4.9 with moderate sensitivity/specificity. Future studies are needed to verify these findings.Citation:Siriwat R, Wang L, Shah V, Mehra R, Ibrahim S. Obstructive sleep apnea and insulin resistance in children with obesity. J Clin Sleep Med. 2020;16(7):1081–1090.BRIEF SUMMARYCurrent Knowledge/Study Rationale: Children with obesity are at risk for obstructive sleep apnea (OSA) and insulin resistance (IR). This study was performed to understand the relationship between OSA and IR while examining adiposity exclusively in children with obesity.Study Impact: In a clinic-based study of children with obesity, OSA severity, as measured by the obstructive apnea-hypopnea index, is a significant predictor of IR independent of body mass index z-score. Additionally, this study uniquely evaluates waist circumference z-score as a measure of adiposity in this population and identifies that the older age group age had a stronger association of OSA and IR. Furthermore, using a cutoff of obstructive apnea-hypopnea index ≥ 4.9 events/h may offer clinicians a way to predict pediatric patients with obesity at risk of IR with modest sensitivity and specificity.INTRODUCTIONPediatric obesity is a major public health problem, with approximately 17% of children and adolescents 2–19 years of age with obesity (ie a body mass index [BMI] of ≥95th percentile).1,2 The consequences of childhood obesity include negative influences on cardiovascular health such as hypertension and metabolic derangements such as the metabolic syndrome and insulin resistance (IR)/type 2 diabetes in early life and in adulthood.3,4 Identification of modifiable risk factors such as obstructive sleep apnea (OSA), a disorder characterized by repetitive collapse of the upper airway associated with intermittent bouts of hypoxia, sleep fragmentation, autonomic nervous system fluctuations, and intrathoracic pressure alterations, is imperative to minimize immediate and downstream health detriment.The prevalence of OSA is 2–3% in children,5 and some studies demonstrate the prevalence as high as 60% in children and adolescents with obesity.6,7 Similar to obesity, OSA is associated with increased cardiovascular risk and increased mortality if left untreated.8–11 Obesity and OSA share common cardiometabolic pathophysiologic effects on health through mechanisms such as increase in blood glucose and IR,12,13 as well as alterations in inflammatory cytokines and adipokines.14,15 This reflects the likely bidirectional nature of the obesity-OSA association characterized not only by obesity risk for OSA but also OSA affecting weight gain and metabolic derangements, leading to obesity.Adults with OSA are at greater risk of having type 2 diabetes, IR, and metabolic syndrome.16–18 In children, emerging data suggest insufficient sleep caused by OSA may lead to metabolic alterations including IR.19,20 For example, significantly higher levels of fasting insulin, homeostasis model assessment of insulin resistance (HOMA-IR) levels, and low-density lipoprotein cholesterol were identified in adolescents with OSA.21 These adolescents also had an increased systolic and diastolic blood pressure after adjusting for BMI percentile.21 In another study, OSA severity was associated with increased fasting insulin, blood glucose, and HOMA-IR even after controlling for age and BMI z-score.22Although many studies showed an association between OSA and IR, some have not. Treatment for OSA with either adenotonsillectomy in children or continuous positive airway pressure in small randomized controlled trials of primarily children without obesity have shown inconsistent improvements in IR.23,24 A large cohort study in children who snore found no significant correlation between sleep parameters and serum insulin, serum glucose, insulin/glucose ratio, or HOMA-IR for either children with or without obesity.25 Another study in children demonstrated no difference in whole-body insulin sensitivity in those with and without OSA.26Given inconsistencies in the literature, we chose to investigate the association of OSA and IR, leveraging prospective data collection in a high-risk pediatric obesity clinic. We hypothesized that increased severity of OSA, measured by the obstructive apnea-hypopnea index (oAHI) and oxygen desaturation index (ODI), will be predictive of measures of IR. Our secondary aims were to evaluate the interaction of age on the relationship between oAHI and ODI and IR. We also assessed oAHI and ODI cutoff values that rendered higher risk of having IR. To our knowledge, there are limited data in this high-risk group of children with obesity.METHODSRecruitmentThis is a retrospective examination of data collected from a tertiary care pediatric obesity management clinic at Cleveland Clinic Children’s Hospital. Consecutive patients 2–18 years of age with normal development who underwent overnight polysomnography (PSG) between January 1, 2013 and March 31, 2017 and who also had anthropometric measurements and laboratory testing for IR were included. Exclusion criteria were genetic or craniofacial abnormalities, and those children treated for IR with metformin before laboratory studies. Demographics including age, sex, and race were collected. Comorbidities associated with obesity (ie hypertension, prediabetes, dyslipidemia) and medical history (such as attention deficit hyperactivity disorder) were collected. The study was conducted with approval from the Institutional Review Board of the Cleveland Clinic Foundation.Data CollectionComorbidities were collected by review of the medical history in the electronic medical record. Additional comorbidity data were collected by clinical evaluation in the obesity clinic for obesity-related complications, such as diabetes and hypertension. Nonalcoholic fatty liver disease was defined by upper quadrant abdominal ultrasound showing >5% fatty deposition.Laboratory data on IR consisted of fasting morning glucose and insulin levels. Laboratory data within 6 months of the PSG were selected, with the median time frame of 2 months. Serum fasting glucose was measured using glucose hexokinase (Roche Cobas 8000 702 Platform, Roche, Basel, Switzerland), with an intra-assay coefficient of variation ≤ 5%. Fasting insulin was measured by enzymatic methods under Centers for Disease Control and Prevention guidelines27 using the chemiluminescence immunoassay (CLIA) (ADVIA Centaur XP Immunoassay system, an automated in vitro diagnostic analyzer, catalog 078-A011-03, Siemens Healthcare Diagnostics, Tarrytown, NY), with a coefficient of variation ≤ 8.0%.HOMA-IR, our primary outcome measure, was calculated as a product of fasting insulin and glucose using a standard equation (fasting insulin [μIU/mL] × fasting blood glucose [mg/dL]/405).28,29 Based on this score and taking into consideration previously published studies, clinically significant insulin resistance was defined using the HOMA-IR cut-point ≥ 3 in children.30,31Anthropometric measurements comprised of height, weight, waist-height ratio (WHR), waist circumference (WC), and neck circumference were also collected.32,33 Height and weight were recorded for each patient for calculation of BMI (kg/m2) and BMI z-score. BMI z-score was calculated as the number of standard deviations the child’s BMI differs from the median according to normative values. Compared with BMI, the BMI z-score is preferred by the US Preventative Services Task Force, because it is the only widely available measure that could be used to compare relative degree of excess weight across age groups.2 WC was measured after palpating the iliac crest in the midaxillary lines while participants/parents placed their hands on the opposite shoulders.34 The WC z-score was based on tables using the National Health and Nutrition Examination Survey III. We used an online link to calculate the WC z-score that uses data from these tables: https://apps.cpeggcep.net/WCz_cpeg. We included the WC because it is more strongly associated with cardiometabolic risks than the BMI z-score, likely because of the relationship with central adiposity.35PolysomnographyAll patients included in the analytic sample underwent polysomnography (PSG) using a commercially available PSG system (Nihon Kohden’s Polysmith, Tokyo, Japan) at the Cleveland Clinic sleep laboratory. Physiologic PSG parameters were collected as follows: electroencephalographic activity in wakefulness and in sleep staging, electro-oculograms, chin and lower limb electromyography, electrocardiogram, air flow signals using nasal pressure transducer and air flow thermistor, respiratory effort channels using chest wall and abdominal plethysmography, end-tidal carbon dioxide, and pulse oximetry to measure oxygen saturation. PSG variables considered for analyses included the obstructive apnea-hypopnea index (oAHI), oxygen desaturation index (ODI), arousal index, sleep stages, and total sleep time.PSG data were scored by a single scorer who was blinded to the metabolic variables and rereviewed to ensure consistency. The respiratory events (apnea and hypopnea) were identified and scored according to the American Academy of Sleep Medicine pediatric criteria as defined in the American Academy of Sleep Medicine Manual for Scoring of Sleep and Associated Events.36 Hypopnea were scored if the event was associated with a 30% reduction in amplitude of the nasal pressure transducer, lasting for at least 2 breaths, and was associated with an arousal/awakening or 3% desaturation. Apneas were defined as ≥90% reduction in airflow lasting at least 2 breaths in duration The oAHI was defined as the total number of obstructive and mixed apnea and hypopnea per hour of sleep. Clinically significant OSA was classified as oAHI ≥ 5 events/h (primary predictor) because of associations of this cutoff with metabolic syndrome in adolescents,21 elevated C-reactive protein levels, and adverse clinical outcomes in young children/adolescents.37 The ODI was defined as the number of times per sleep hour with oxygen desaturation of 3% or more.38Other PSG and sleep variables examined included percentage (%) of time in slow-wave sleep, sleep efficiency (percentage of the sleep period spent asleep), total sleep time, arousal index, and reported habitual sleep duration during both weekday and weekend.Statistical methodsLinear regression was used to assess the relationship of HOMA-IR with OSA predictors of interest. Models were unadjusted (model 1); adjusted for sex, race, and BMI z-score (model 2); adjusted for sex, race, and WC z-score (model 3); or adjusted for sex, race, and WHR (model 4). Age was not considered as a covariate given the inherent consideration of age in the z-score calculations. Linear models were performed with oAHI as a categorical variable (<5 vs ≥5 events/h) defined as the primary predictor or oAHI alternatively considered as a continuous variable (ie, interpreted as change in HOMA-IR per 5-unit increase in oAHI). In secondary analyses, we considered ODI as a predictor of HOMA-IR using the same methods as described for oAHI. In our secondary analyses, we also examined WHR, as prior data have suggested that this may be superior to BMI to discriminate obesity-related cardiometabolic risk in adults.39 The interaction between oAHI and age group (dichotomized at 12 years based on the sample median) was tested to assess the difference in the relationship between oAHI and HOMA-IR in patients <12 years vs 12–19 years old. No multicollinearity was identified by variance inflation factor before modeling. Beta estimates and 95% confidence intervals are presented. Outcome HOMA-IR was log-transformed to satisfy normal distribution and transformed back for presentation of estimates and for ease of interpretation.Receiver operating characteristic (ROC) analysis was used to investigate the optimal cutoff of oAHI and ODI on outcome HOMA-IR ≥ 3. The ROC curve is a plot of the true positive rate (sensitivity) against the false-positive rate (specificity) for the different possible cutoff values of a diagnostic test based on a logistic regression model. The area under the curve (AUC) demonstrated the overall discriminatory power of a diagnostic test over the whole range of testing values. Sensitivity and specificity have been calculated at all possible cutoff points to find the optimal cutoff value. The optimal cutoff value was selected based on the distance to (0, 1), Youden index (sensitivity + specificity − 1), absolute value of difference between sensitivity and specificity, and correct classification rate (weighted average of sensitivity and specificity).All analyses were performed in SAS software (version 9.4; SAS, Inc, Cary, NC), and a significance level of 0.05 was assumed for all tests.RESULTSA total of 80 patients with a mean age of 11.4 ± 4.0 years met inclusion criteria and were analyzed. Forty-five (56.3%) were female, and thirty-seven (46.3%) were Caucasian. All patients were obese (BMI z-score> 95th percentile), with a median BMI of 34.6 kg/m2 (interquartile range, 29.9–40.1) and a median BMI z-score of 2.5 (interquartile range, 2.3–2.8). Forty-three patients (54%) had an oAHI < 5 events/h (non-OSA group), and 37 patients (46%) had an oAHI ≥ 5 events/h (OSA group). Overall, comorbidities were similar in both groups: 19 (23.8%) with hypertension, 15 (18.8%) with asthma, 19 (23.8%) with attention deficit hyperactivity disorder, and 17 (21.3%) with nonalcoholic fatty liver disease. There were no group differences in sex, race, or habitual sleep duration on weekdays and weekends. The OSA group had a higher age, BMI, WHR, waist and neck circumference, arousal index, fasting glucose, and HOMA-IR than those in the non-OSA group. Contrary to the differences in BMI distribution across OSA and non-OSA groups, there were no statistically significant differences of BMI z-score and WC z-score between OSA and non-OSA groups (Table 1).Table 1 Descriptive subject characteristics.FactorOverall (n = 80)AHI < 5 Events/h (n = 43)AHI ≥ 5 Events/h (n = 37)PDemographics Age (yr)11.4 ± 4.010.4 ± 3.512.5 ± 4.3.017a Sex (% male)35 (43.8)16 (37.2)19 (51.4).20b Race/ethnicity (%).40c White46.316 (37.2)21 (56.8) Black36.317 (39.5)12 (32.4) Hispanic11.36 (14.0)3 (8.1) Asian1.31 (2.3)0 (0.0) Other5.03 (7.0)1 (2.7) BMI z-score2.5 [2.3, 2.8]2.5 [2.3, 2.7]2.7 [2.4, 3.1].055d Waist circumference z-score2.2 [2.0, 2.5]2.1 [2.0, 2.4]2.3 [2.0, 2.6].34d Neck circumference (cm)36.3 [34.0, 40.0]35.0 [32.0, 38.0]38.5 [35.5, 41.9].002d Waist-height ratio0.69 [0.60–0.78]0.66 [0.59–0.73]0.72 [0.62–0.82].004aComorbidity (%) Hypertension19 (23.8)8 (18.6)11 (29.7).24b Asthma15 (18.8)9 (20.9)6 (16.2).59b ADHD19 (23.8)10 (23.3)9 (24.3).91b Depression11 (13.8)4 (9.3)7 (18.9).21b Anxiety13 (16.3)6 (14.0)7 (18.9).55b Allergic rhinitis9 (11.3)6 (14.0)3 (8.1).49c GERD7 (8.8)5 (11.6)2 (5.4).44c NAFLD17 (21.3)9 (20.9)8 (21.6).94bPSG Variables Total sleep time (min)391.0 [361.5, 448.0]415.0 [371.0, 454.0]381.0 [353.0, 416.0].060d Sleep efficiency (%)87.6 [75.0, 92.7]88.7 [75.7, 92.4]83.8 [72.4, 94.1].67d Sleep latency (min)26.5 [12.8, 55.3]27.5 [14.5, 47.0]24.0 [8.0, 65.0].73d AHI (events/h)4.0 [1.9, 9.4]2.0 [1.00, 2.9]10.3 [6.9, 21.9]<.001d ODI (O2 desaturations/h)3.9 [2.1, 9.2]2.4 [1.05, 3.2]9.2 [6.2, 18.7]<.001d Mean SpO2 (%)96.0 [95.0, 97.0]97.0 [96.0, 98.0]96.0 [94.0, 96.0]<.001d Arousal index (events/h)13.1 [10.3, 18.3]11.6 [8.5, 15.1]16.3 [12.6, 24.6]<.001dHabitual sleep duration Weekday sleep duration (h)8.5 [8.0, 10.0]9.0 [8.0, 10.0]8.0 [8.0, 9.5].10d Weekend sleep duration (h)10.0 [8.0, 11.0]10.0 [8.0, 11.0]10.0 [8.5, 10.0].56dLabs/metabolic indices Time since labs (mo)2.0 [1.00, 7.0]3.0 [1.00, 9.0]2.0 [1.00, 5.0].24d Fasting glucose (mg/dL)82.0 [76.0, 88.0]79.0 [76.0, 85.0]85.0 [80.0, 90.0].012d Fasting insulin (mU/mL)24.0 [17.4, 35.6]22.0 [15.2, 29.0]29.0 [18.1, 43.9].065d HbA1c (%)5.6 [5.3, 5.7]5.6 [5.3, 5.7]5.6 [5.3, 5.8].56d HOMA-IR4.2 [3.0, 6.4]3.8 [2.7, 5.3]5.0 [3.3, 8.5].034dData are presented as mean ± SD, median [interquartile range], or n (%) as appropriate. P values: at test, bPearson's χ2 test, cFisher's exact test, and dKruskal-Wallis test. ADHD = attention deficit hyperactivity disorder, AHI = apnea-hypopnea index, BMI = body mass index, GERD = gastroesophageal reflux disease, HOMA-IR = homeostatic model assessment-insulin resistance, NAFLD = nonalcoholic fatty liver disease, ODI = oxygen desaturation index, OSA = obstructive sleep apnea, SpO2 = peripheral capillary oxygen saturation.In unadjusted analyses, linear regression models showed that OSA (defined as oAHI ≥ 5 events/h) was associated with 40% greater HOMA-IR levels compared with the non-OSA group (oAHI < 5 events/h; β coefficient = 0.40; 95% CI, 0.06, 0.86). After adjustment of sex, race, and BMI z-score, the association between HOMA-IR and oAHI ≥ 5 events/h remained significant (β coefficient = 0.38; 95% CI, 0.01, 0.87), but this association was mitigated and no longer significant in the WC z-score and WHR models (Table 2).Table 2 Linear model of obstructive sleep apnea defined by the obstructive apnea-hypopnea index relative to homeostatic model assessment-insulin resistance.PredictorCoefficient (95% CI)PModel 1 (unadjusted)oAHI ≥ 5 events/h0.40 (0.06, 0.86).020Model 2oAHI ≥ 5 events/h0.38 (0.01, 0.87).021Sex−0.12 (−0.35, 0.19).41Race0.02 (−0.25, 0.38).89BMI z-score0.14 (−0.16, 0.54).40Model 3oAHI ≥ 5 events/h0.23 (−0.11, 0.70).20Sex−0.10 (−0.35, 0.25).52Race−0.03 (−0.31, 0.34).83WC z-score0.08 (−0.21, 0.48).63Model 4oAHI ≥ 5 events/h0.08 (−0.22, 0.50).64Sex−0.12 (−0.36, 0.20).40Race−0.02 (−0.28, 0.34).91WHR (per 0.1 increase)0.21 (0.01, 0.44).039BMI = body mass index, CI = confidence interval, oAHI = obstructive apnea-hypopnea index, WC = waist circumference, WHR = waist-height ratio.Table 3 shows unadjusted and adjusted models of the association of HOMA-IR with the primary predictors, continuous oAHI and ODI, including the interaction of age group. For every 5-unit increase in oAHI, there was a 7% increase in HOMA-IR (β coefficient = 0.07, 95% CI, 0.01, 0.13). Similarly, when patients had a 5-unit increase in ODI, HOMA-IR would increase by 7% (β coefficient = 0.07; 95% CI, 0.005, 0.13; P = .034). After adjustment of sex, race, and BMI z-score, the association between HOMA-IR and oAHI remained significant (β coefficient = 0.06; 95% CI, 0.003, 0.13). After these same adjustments, the association between HOMA-IR and ODI was no longer significant (β coefficient = 0.06; 95% C, −0.01, −0.13). After taking into account sex, race, and WC z-score, and WHR, the relationships between HOMA-IR and oAHI, as well as HOMA-IR and ODI, were no longer statistically significant. Age group was assessed to have a modification effect in the relationship of both oAHI and ODI with respect to HOMA-IR. When WC z-score and WHR were included in the model, older and younger groups had statistically significant differential relationships of OSA indices and HOMA-IR (statistical interaction term for WC z-score: P = .020 and .029 for oAHI and ODI, respectively; for WHR: P = .034 and .040 for oAHI and ODI, respectively). However, when adjusting for BMI z-score, findings were no longer statistically significant (Figure 1 and Figure 2).Table 3 Unadjusted and adjusted models of the association of homeostatic model assessment-insulin resistance with primary predictors: continuous oAHI and ODI.PredictoroAHIODICoefficient (95% CI)PCoefficient (95% CI)PModel 1 (unadjusted)oAHI or ODI (per 5 units)0.07 (0.01, 0.13).0150.07 (0.005, 0.13).034Model 2oAHI or ODI (per 5 units)0.06 (0.003, 0.13).0410.06 (−0.01, 0.13).082Sex−0.09 (−0.33, 0.23).53−0.10 (−0.35, 0.23.48Race0.008 (−0.26, 0.36).990.01 (−0.26, 0.39).94BMI z-score0.11 (−0.19, 0.51).510.10 (−0.20, 0.53).54Model 3oAHI or ODI (per 5 units)0.04 (−0.02, 0.11).210.04 (−0.04, 0.11).34Sex−0.09 (−0.35, 0.26).55−0.10 (−0.36, 0.26).53Race−0.04 (−0.31, 0.34).81−0.03 (−0.31, 0.37).86WC z-score0.05 (−0.24, 0.45).770.05 (−0.25, 0.47).77Model 4oAHI or ODI (per 5 units)0.02 (−0.04, 0.08).560.01 (−0.06, 0.09).82Sex−0.12 (−0.35, 0.20).42−0.12 (−0.36, 0.22).43Race−0.02 (−0.29, 0.34).88−0.01 (−0.29, 0.37).93WHR (per 0.1 increase)0.20 (0.01, 0.44).0430.22 (0.01, 0.47).036Model 5aoAHI or ODI (per 5 units); age < 12 years0.01 (−0.09, 0.12).81−0.004 (−0.12, 0.13).95oAHI or ODI (per 5 units); age ≥ 12 years0.07 (−0.01, 0.14).0710.06 (−0.02, 0.14).15Sex−0.13 (−0.36, 0.17).35−0.14 (−0.37, 0.18).36Race−0.05 (−0.30, 0.29).72−0.03 (−0.29, 0.33).83BMI z-score0.24 (−0.10, 0.72).180.27 (−0.10, 0.78).17Age ≥ 12 vs < 12 years0.26 (−0.12, 0.81).210.28 (−0.13, 0.87).20Model 6aoAHI or ODI (per 5 units); age ≥ 12 years−0.10 (−0.20, 0.02).093−0.12 (−0.24, 0.01).078oAHI or ODI (per 5 units); age ≥ 12 years0.06 (−0.01, 0.14).0880.05 (−0.03, 0.14).20Sex−0.13 (−0.36, 0.18).37−0.11 (−0.36, 0.22).46Race−0.11 (−0.35, 0.22).46−0.10 (−0.35, 0.25).53WC z-score0.37 (−0.03, 0.93).0730.38 (−0.03, 0.97).076Age ≥ 12 vs < 12 years0.23 (−0.16, 0.81).280.25 (−0.17, 0.88).28Model 7aoAHI or ODI (per 5 units); age ≥ 12 years−0.09 (−0.19, 0.02).097−0.12 (−0.23, 0.01).066oAHI or ODI (per 5 units); age ≥ 12 years0.05 (−0.02, 0.12).200.03 (−0.04, 0.12).39Sex−0.16 (−0.37, 0.13).26−0.14 (−0.37, 0.17).33Race−0.09 (−0.33, 0.23).52−0.08 (−0.33, 0.25).58WHR (per 0.1 increase)0.22 (0.03, 0.45).0220.24 (0.04, 0.48).015Age ≥ 12 vs < 12 years0.13 (−0.21, 0.60).500.14 (−0.22, 0.65).50Models 2 and 5 adjusted for BMI z-score; models 3 and 6 adjusted for WC z-score; models 4 and 7 adjusted for WHR; and models 5, 6, and 7 include the interaction of age groups. aModel 5, P = .41; model 6, P = .020 for oAHI and P = .029 for ODI; model 7, P = .034 for oAHI and P = .040 for ODI. BMI = body mass index, oAHI = obstructive apnea-hypopnea index, ODI = oxygen desaturation index, WC = waist circumference, WHR = waist-height ratio.Figure 1: Scattered plots showing the relationship of HOMA-IR and oAHI along with interactions of age and BMI z-score.The relationship between HOMA-IR and oAHI had no significant difference (P = .41) among age groups when using the interaction of BMI z-score. BMI = body mass index, HOMA-IR = homeostatic model assessment of insulin resistance, oAHI = obstructive apnea-hypopnea index.Download FigureFigure 2: Scatter plots showing the relationship of HOMA-IR and oAHI along with interactions of age and waist circumference z-score.The relationship between HOMA-IR and ODI was significant (P = .029), when factoring waist circumference z-score, where HOMA-IR increased in the older group but decreased in the younger group when oAHI increased. BMI = body mass index, HOMA-IR = homeostatic model assessment of insulin resistance, oAHI = obstructive apnea-hypopnea index, ODI = oxygen desaturation index.Download FigureAdditional secondary analyses were focused on the evaluation of oAHI and ODI cutoffs by assessment of ROC curves. ROC curves identified the oAHI cut-point of ≥4.9 as predicting HOMA-IR ≥ 3 with a sensitivity 0.76, specificity of 0.66, and AUC of 0.70 (95% CI, 0.57, 0.83; P = .028). The ODI cut-point of ≥4.6 showed a sensitivity of 0.70, specificity of 0.67, and AUC of 0.68 (95% CI, 0.55, 0.80) for predicting HOMA-IR ≥ 3, without significant associations between ODI and high HOMA-IR (P = .11) (Figure 3).Figure 3: Receiver operating characteristic curve (ROC) of sleep indices on HOMA-IR ≥ 3.Top: ROC of oAHI on HOMA-IR ≥ 3. The association of oAHI and HOMA-IR ≥ 3 was significant at P = .028. The oAHI cutoff value was 4.9 (sensitivity, 0.76; specificity, 0.66) and AUC was 0.70 (95% confidence interval, 0.57–0.83). Bottom: The association of ODI and HOMA-IR ≥ 3 was not significant (P = .11). The ODI cutoff value was 4.6 (sensitivity, 0.70; specificity, 0.67, and AUC was 0.68 (95% confidence interval, 0.55–0.80). AUC = area under the receiver operating characteristic curve, HOMA-IR = homeostatic model assessment of insulin resistance, oAHI = obstructive apnea-hypopnea index, ODI = oxygen desaturation index.Download FigureDISCUSSIONIn this study of patients presenting to a pediatric obesity clinic (median BMI = 34.6 kg/m2), we identified a significant association of OSA (both defined by oAHI dichotomized at 5 events/h and also as a continuous measure) and HOMA-IR even after adjustment for obesity. This study extends current knowledge of OSA and association with insulin resistance in pediatric obesity to a clinical cohort with sex- and race-based diversity, as well as careful consideration of anthropometric measures. When considering a 5-unit increase in oAHI or ODI as a continuous predictors, a 7% increase in HOMA-IR was observed even after accounting for obesity (based on BMI z-score). The statistical interaction of both oAHI and ODI with respect to age on HOMA-IR was significant even after adjustment for waist z-circumference.Our study contrasts with a similar study investigating children without obesity with habitual snoring, in which BMI was a significant predictor of fasting insulin and HOMA-IR and not the severity of sleep-disordered breathing.40 Because BMI may not precisely reflect adiposity and obesity in children, we chose to examine associations with BMI z-score, which may allow for more accurate discernment of influence of obesity and adiposity. In a study of snoring in children with and without obesity, obesity was defined as the principal determinant of IR, whereas snoring appeared to have a lesser role.25 In both these studies, snoring, and not PSG measuring oAHI, was used. Snoring alone is on the spectrum of sleep-disordered breathing and may by itself not be associated with IR. Our results indicate that OSA, as measured by the gold standard PSG, is associated with IR in an obese pediatric population beyond the influence of weight or BMI alone.We also examined adiposity in the relationship between IR and OSA in children because of previous findings that adiposity may be the predominant determinant of IR compared with OSA or sleep architecture.41 We examined adiposity by including both the BMI z-score and WC z-score to measure clinically derived measures, with the latter being more strongly associated with cardiometabolic risks because of the relationship with central adiposity.35 When taking waist circumference z-score into account, oAHI and ODI were no longer significantly associated with HOMA-IR. Similarly, in our secondary analysis, we examined WHR in the relationship with IR and OSA. When taking WHR into account, oAHI and ODI were no longer associated with HOMA-IR. In fact, WHR appeared more sensitive to IR than BMI z-score and WC z-score in the different models (Table 3). With every 0.1-unit increase in WHR, HOMA-IR would increase 22% (95% CI, 0.036, 0.44; P = .018).However, we de