Introduction
Obstructive sleep apnea (OSA) is highly common in the adult population with an incidence that is rising globally [1, 2]. We have shown OSA to be highly prevalent among adult patients with hypertrophic cardiomyopathy (HCM) [3]. Under normal conditions, uninterrupted sleep is a highly regulated physiological state essential for maintaining autonomic balance, metabolic homeostasis, and broader physiological integrity [4–8].
Characterised by intermittent upper airway collapse during sleep, OSA causes hypoxemia, hypercapnia, sleep fragmentation, and large swings in intrathoracic pressure that result in increased sympathetic tone and significant recurrent transmural pressure differences [9, 10]. OSA is associated with several severe cardiovascular and metabolic consequences that are of direct relevance to patients with hypertrophic cardiomyopathy including a higher likelihood of obesity, heart failure, arrhythmia, and mortality [3, 11–18]. Indeed, the high prevalence of comorbid sleep apnea, associated increased symptom burden, risk of arrhythmia, and potential benefit from risk factor modulation and treatment of OSA are highlighted in the AHA/ACC guidelines on management of HCM [19].
Importantly, a variety of interventions offer effective treatment for OSA, suggesting potential clinical benefit from screening of at-risk populations [20]. It is also well-established that treatment significantly alleviates sleepiness attributable to OSA, the most common symptom and a key factor affecting quality of life [21]. Although overnight attended polysomnography remains the gold standard diagnostic tool for the diagnosis of sleep apnea, it is costly, time consuming and requires trained sleep technologists that are in short supply, emphasizing the importance of straightforward, reliable methods to identify and triage patients at highest risk of OSA [22].
While several screening tools are currently employed among the general population to identify patients at risk of OSA likely to benefit from formal assessment, none have thus far been validated in the HCM population.
The primary aim of this study was to validate the predictive utility of three commonly used screening tools for OSA (Berlin questionnaire, Epworth Sleepiness Scale, and STOP-Bang questionnaire) against gold-standard attended polysomnographic assessment within a large cohort of patients with HCM. The secondary aim was to use data from this cohort to construct an exploratory model capable of identifying patients at highest risk of comorbid OSA. While unvalidated and hypothesis-generating, this approach enabled a comparative assessment of model performance against commonly used screening tools, highlighting the potential value of HCM-specific risk stratification.
Methods
Patient population and polysomnographic evaluation
This prospective cohort study was approved by the Mayo Clinic Institutional Review Board. Inclusion criteria included age > 18 years, a diagnosis of HCM, and the ability to answer screening questionnaires and undergo polysomnography. HCM diagnostic criteria included left ventricular wall thickness ≥ 15 mm (or ≥ 13 mm in first-degree relatives of patients with HCM) when assessed by echocardiography or cardiac magnetic resonance imaging, in the absence of other systemic or cardiac disease capable of explaining the degree of hypertrophy.
Exclusion criteria included a prior diagnosis of sleep apnea or pregnancy, with no specific body mass index (BMI) or comorbidity cut-offs applied. To minimize selection bias, a baseline cohort of all adult patients with known HCM evaluated at Mayo Clinic who were residents of Minnesota, Wisconsin, or Iowa was identified. This geographic restriction was applied to ensure study feasibility for overnight polysomnography and to mitigate the extreme phenotype bias associated with national and international tertiary referrals.
A statistician randomized the master list of medical record numbers, and eligible patients were contacted by telephone sequentially based on this randomized list to invite them to participate. Participants were asked to abstain from alcohol, stimulants, and exercise the day prior to attendance and during their visit and written informed consent was obtained from all patients. All patients underwent standardized anthropometric assessment and administration of screening tools prior to polysomnography.
Overnight polysomnography was carried out and scored by a registered sleep technologist according to American Academy of Sleep Medicine (AASM) guidelines and definitions [10]. The technologist was blinded to the results of the screening tools. An apnea hypopnea index (AHI) ≥ 5 but < 15 events/hour was defined as mild OSA, ≥ 15 but < 30 events/hour as moderate OSA, and ≥ 30 events/hour as severe OSA [10].
Screening tools
STOP-bang questionnaire
The STOP-Bang questionnaire consists of eight ‘yes or no’ questions on four subjective items (STOP: snoring, tiredness, observed apnea, and high blood pressure) and four demographic items (Bang: body mass index [BMI] >35 kg/m2, age > 50 years, neck circumference ≥ 40 cm, and male sex) [23]. Final score ranges from zero to eight, classifying the risk of OSA [24]. A systematic review and meta-analysis of nine studies (n = 1894 patients) validated the STOP-Bang questionnaire as a screening tool for OSA in patients with cardiovascular risk factors.
With a score ≥ 3, pooled sensitivity was consistently high at 89%, 91%, and 94% in detecting any OSA, moderate-to-severe OSA, and severe OSA, respectively. The pooled specificities were moderately low at 32.3%, 22.5%, and 18.3% for all OSA, moderate-to-severe OSA, and severe OSA, respectively. The area under the receiver operator characteristic (ROC) curve (AUC) was 0.86 for detecting OSA of any severity [25].
Alternatively, a score ≥ 5 has been used to identify patients at greatest risk of moderate-severe OSA [24, 26]. For context, a comparative summary of prior studies assessing the performance of these screening tools in other cohorts is detailed in Supplemental Table 1.
Epworth sleepiness scale (ESS)
The ESS lists eight everyday situations and assesses the probability of falling asleep using a four-point Likert-type scale ranging from 0 (no chance of dozing) to 3 (high chance of dozing). Individual item scores are summed to a total score ranging from 0 to 24, with high values indicating more severe daytime sleepiness [26]. A cutoff ≥ 11 has typically been used to define excessive daytime sleepiness and identify those patients at greatest risk of OSA.
Using the cutoff ≥ 11, ESS had reasonable specificity (79%, 81% and 83% for mild, moderate and severe OSA respectively) with low sensitivity (16%, 15%, 19% for mild, moderate and severe OSA respectively) [27].
Berlin questionnaire
The Berlin questionnaire is composed of 3 categories. The first category pertains to snoring behavior, the second to symptoms of daytime sleepiness and fatigue, and the last category evaluates the presence of hypertension or obesity. The range of scores for category 1 is 0–6 and category 2 is 0–3, with value of 2 considered as the cutoff for category positivity.
Category 3 is considered positive if the patient has a history of hypertension or if the BMI is > 30 kg/m2. The patient is considered at high risk of OSA if two or more categories are positive [28]. When administered in the general population, the sensitivity and specificity of the questionnaire to predict an AHI ≥ 15events/hour were 59%, and 78% respectively.
The AUC for moderate-to-severe OSA was 0.68. When used to predict an AHI ≥ 30, the sensitivity of the questionnaire increased to 77% with a specificity of 73%. Corresponding negative predictive value (NPV), positive predictive value (PPV), and AUC of the questionnaire to predict severe OSA were 96%, 25%, and 0.75, respectively [29].
Statistical analysis
Most continuous variable data were not normally distributed hence are presented as median (25th, 75th percentile). Categorical variable data are presented as frequency and percent. The Kruskal-Wallis test was used for between-group comparisons of continuous data and Pearson’s χ2 test used to compare categorical data.
Sensitivity, specificity, AUC, and positive and negative predictive values were calculated for each of the screening tools’ individual performance in correctly identifying patients with any severity of sleep apnea (AHI ≥ 5) and in correctly identifying patients with moderate to severe OSA (AHI ≥ 15) using their standard cutoffs based on the literature. Finally, exploratory multivariable logistic regression with stepwise selection (requiring p < 0.1 for model inclusion) was used to identify the clinical variables most predictive of sleep apnea within this cohort of patients with HCM. Because this model was intended strictly as an exploratory, hypothesis-generating tool based on a limited pool of traditional clinical variables, conventional stepwise selection was utilized.
However, future prospective studies validating these predictive metrics will necessitate the use of standardized assessment technologies and rigorous algorithmic approaches to ensure generalizability and reproducibility [30–32]. BlueSky Statistics v7.4 (BlueSky Statistics, Chicago, USA) was used for statistical analysis, with a 2-sided p < 0.05 considered statistically significant (Fig. 1).
Fig. 1Reliability of common screening tools for Obstructive Sleep Apnea (OSA) in patients with Hypertrophic Cardiomyopathy (HCM). A cohort of 159 HCM patients underwent baseline measurements and attended polysomnography.
The prevalence of OSA was approximately 60% and was associated with older age, male sex, higher BMI, elevated blood pressure, and higher NYHA class. Standard screening tools (Epworth Sleepiness Scale, STOP-Bang, and Berlin Questionnaire) demonstrated low sensitivity and poor diagnostic accuracy (AUC < 0.65). A custom regression model incorporating age, BMI, and waist-to-hip ratio (WHR) significantly outperformed standard tools (AUC 0.82), though further validation is required
Results
Clinical and polysomnographic characteristics
In total, 242 patients were invited, from which 159 participated in the study, and all volunteers successfully completed screening tool and polysomnographic assessment. Median (25th, 75th ) age was 60 (48, 68) years, 101 (64%) participants were male. Ninety (57%) patients were diagnosed with SDB; of whom 42 (46%) were diagnosed with mild OSA; 23 (26%) moderate OSA; and 25 (28%) with severe OSA.
Increasing severity of diagnosed OSA was associated with male sex (p = 0.022), increasing age (p < 0.001), and increasing BMI (p < 0.001). Subjects diagnosed with OSA were more likely to report NYHA class II or III rather than class I dyspnea (p = 0.006) and were more likely to be hypertensive (p = 0.010), to describe their sleep quality as “poor” (p = 0.016), and more likely to have been told by their partner that they stop breathing at night (p = 0.018). Clinical and polysomnographic characteristics are summarized in Table 1.
Table 1Clinical & polysomnographic characteristicsNo OSA(N = 69)Mild OSA(N = 42)Moderate OSA(N = 23)Severe OSA(N = 25)Total(N = 159)p-valueAHI, (events/hour)1.7 (0.8, 3.2)9.6 (7.5, 11.6)22.1 (16.3, 25.1)47.9 (40.1, 61.6)7.6 (2.5, 16.5)N/AAge, (y)54 (41, 63)63 (54, 69)66 (59, 71)63 (53, 69)60 (48, 68)< 0.001Male37 (54%)28 (67%)14 (61%)22 (88%)101 (64%)0.022BMI, (kg/m2)28 (25, 32)30 (28, 34)31 (30, 35)33 (31, 38)30 (27, 34)< 0.001Waist circ. (cm)97 (90, 106)107 (99, 114)105 (99, 116)106 (103, 116)103 (95, 113)< 0.001Hip circ. (cm)107 (102, 116)111 (103, 119)108 (106, 116)114 (105, 119)109 (103, 117)0.402Waist-hip ratio0.92 (0.86, 0.96)0.96 (0.93, 1.00)0.98 (0.93, 1.00)0.97 (0.93, 1.02)0.94 (0.89, 0.99)< 0.001NYHA class I52 (75%)27 (64%)12 (52%)11 (44%)102 (64%) II12 (17%)12 (28%)10 (43%)7 (28%)41 (25%)0.006 III5 (7%)3 (7%)1 (4%)7 (28%)16 (10%)Hypertension38 (55%)27 (64%)16 (69%)23 (92%)104 (65%)0.010Diabetes3 (4%)5 (11%)3 (13%)6 (24%)17 (10%)0.052Smoking3 (4%)5 (11%)1 (4%)1 (4%)10 (6%)0.383Average SpO2 during sleep95.7 (1.2)94.2 (1.3)93.6 (1.9)92.9 (2.2)94.5 (1.9)< 0.001Minimum SpO2 during sleep89.4 (4.0)85.6 (3.8)81.7 (5.7)78.5 (8.3)85.5 (6.5)< 0.001Subject clinical and polysomnographic characteristics grouped according to OSA status.
Values presented are mean (standard deviation), median (25th, 75th percentiles), or n (%). OSA: obstructive sleep apnea; BMI: body mass index; NYHA: New York Heart Association; AHI: Apnea-hypopnea Index
Screening tool characteristics
STOP-bang questionnaire
In predicting any form or severity of sleep apnea, using a cutoff score ≥ 3 led to a sensitivity of 77% (95% CI: 68%–86%), a specificity of 40% (95% CI: 28%–52%), and an AUC of 0.59 (95% CI: 0.50–0.68), and an odds ratio of OSA [OR (95% CI)] of 2.3 (1.1, 4.7, p = 0.020). In predicting moderate to severe sleep apnea, a cutoff score ≥ 3 led to a sensitivity of 78% (95% CI: 66%–90%), a specificity of 41% (95% CI: 32%–50%), and an AUC of 0.61 (95% CI: 0.51–0.71). Screening tool performance outcomes are summarized in Table 2.
Table 2Screening tool characteristicsScreening ToolCutoffOSA SeverityAUCSensitivitySpecificitySTOP-Bang≥ 3Any OSAModerate-Severe OSA0.59 (0.50–0.68)0.61 (0.51–0.71)77% (68–86)78% (66–90)40% (28–52)41% (32–50)≥ 5Any OSAModerate-Severe OSA0.68 (0.59–0.77)0.71 (0.62–0.80)46% (36–56)53% (39–67)88% (80–96)88% (82–94)Berlin≥ 2 category positivityAny OSAModerate-Severe OSA0.62 (0.53–0.71)0.65 (0.55–0.75)65% (55–75)73% (60–86)57% (45–69)57% (48–66)ESS≥ 11Any OSAModerate-Severe OSA0.58 (0.49–0.67)0.59 (0.49–0.69)24% (15–33)27% (14–40)92% (86–98)60% (51–69)Custom ModelContinuous ScaleAny OSAModerate-Severe OSA0.82 (0.74–0.88)0.84 (0.77–0.91)60% (48–72)76% (65–85)84% (75–91)69% (56–81)Performance of individual screening tools and the predictive model obtained from our study for various degrees of OSA based on different cutoff. Sensitivity, specificity, and predictive values for the exploratory model were calculated using a standard default predicted probability cut-point of > 0.50. OSA: Obstructive Sleep Apnea; AUC: Area under the receiver operator characteristic curve; ESS: Epworth Sleepiness Scale
Epworth sleepiness scale (ESS)
In predicting any form or severity of sleep apnea, using a cutoff score ≥ 11 led to a sensitivity of 24% (95% CI: 15%–33%), a specificity of 92% (95% CI: 86%–98%), and an AUC of 0.58 (95% CI: 0.49–0.67), and an odds ratio of OSA [OR (95% CI)] of 3.3 (1.2, 10.3, p = 0.027). In predicting moderate to severe sleep apnea, a cutoff score ≥ 11 led to a sensitivity of 27% (95% CI: 14%–40%), a specificity of 60% (95% CI: 51%–69%), and an AUC of 0.59 (95% CI: 0.49–0.69).
Berlin questionnaire
In predicting any form or severity of sleep apnea, using the ‘high-risk’ cutoff led to a sensitivity of 65% (95% CI: 55%–75%), a specificity of 57% (95% CI: 45%–69%), and an AUC of 0.62 (95% CI: 0.53–0.71), and an odds ratio of OSA [OR(95% CI)] of 2.4 (1.3, 4.7, p = 0.007). In predicting moderate to severe sleep apnea, the ‘high-risk’ cutoff led to a sensitivity of 73% (95% CI: 60%–86%), a specificity of 57% (95% CI: 48%–66%), and an AUC of 0.65 (95% CI: 0.55–0.75).
Custom predictive model
Multivariable logistic regression modelling with stepwise selection was applied, with P < 0.1 required for inclusion within the final exploratory model. This found age, BMI, and waist-hip circumference ratio to be most predictive of SDB in this sample. In predicting the presence of OSA of any severity within this cohort, this model had an AUC of 0.82 (95% CI: 0.74–0.88), a sensitivity of 60% (95% CI: 48%–72%), a specificity of 84% (95% CI: 75%–91%), and positive and negative predictive values of 0.70 (95% CI: 0.61–0.79) and 0.72 (95% CI: 0.60–0.84) respectively.
The resulting exploratory logistic regression equation (from which the predicted probability can be calculated by converting these log-odds) for the presence of OSA of any severity was:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}Log-odds(Any\;OSA)=-\\15.63(95\%\;CI:-21.99\;to\;-9.27)+\\0.07(Age\;in\;year,\;95\%\;CI:0.04\;to\;0.11)+\\0.12(BMI\;in\;Kg/m^2,95\%\;CI:0.05\;to\;0.19)+\\8.58(Waist-Hip\;Ratio,95\%\;CI:2.79\;to\;14.37)\end{array}$$\end{document}
Finally, in predicting moderate to severe sleep apnea, the model led to an AUC of 0.84 (95% CI: 0.77–0.91), a sensitivity of 71% (95% CI: 58%–83%), a specificity of 78% (95% CI: 67%–87%), and positive and negative predictive values of 0.77 (95% CI: 0.65–0.89) and 0.67 (95% CI: 0.58–0.76) respectively. The exploratory model to predict the presence of moderate-to-severe OSA was defined by the following logistic regression equation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}Log-odds(Moderate-Severe\;OSA)=-\\17.96(95\%CI:-26.13\;to-9.79)\:+\\\:0.08(Age\;in\;years,\;95\%\;CI:0.04\;to\;0.12)\:+\\\:0.15(BMI\;in\;Kg/m^{2},\;95\%\;CI:0.07\;to\;0.23)\:+\\\:8.82(Waist-Hip\;Ratio,\;95\%\;CI:1.91\;to15.74)\end{array}$$\end{document}
Responses received using the screening tools are reported in Supplementary Table 2. Screening tool receiver operating characteristics in identifying any OSA and moderate-severe OSA are compared in Fig. 2.
Fig. 2OSA screening tool receiver operating curve comparison
Discussion
This study found significant limitations and limited utility of three commonly used screening tools for OSA in patients with HCM and underscores the importance of developing tailored predictive strategies for this high-risk population. Several key findings emerged from the results, with important implications for clinical practice. First, the prevalence of OSA in this cohort was remarkably high, with 57% of patients diagnosed with OSA.
The substantial burden of OSA observed aligns with prior studies in HCM populations, which have demonstrated a strong association between OSA and adverse outcomes such as arrhythmia, heart failure progression, and increased symptom burden [3, 7–12]. Patients with OSA were older, had a higher BMI, and reported worse sleep quality, reinforcing the need to consider these characteristics in risk stratification. These findings highlight the importance of systematically evaluating the possibility of comorbid OSA in patients with HCM as a means of addressing a modifiable risk factor.
Second, the predictive accuracy of established screening tools such as the STOP-Bang questionnaire, Epworth Sleepiness Scale (ESS), and Berlin questionnaire was modest at best when compared to prior studies assessing their use in other cohorts [19, 23, 26]. While the STOP-Bang questionnaire (score ≥ 3) demonstrated a high sensitivity (77–78%) for detecting OSA, its low specificity (40–41%) somewhat limits its practical utility in this population due to the potential for over-referral from a high false positive rate. In a high-risk cardiac population, prioritizing high sensitivity is often necessary to prevent missed diagnoses that could exacerbate arrhythmia and sudden death risks.
However, the cost of the STOP-Bang’s associated high false-positive rate is substantial; over-referral can overwhelm limited polysomnography resources and delay time-to-treatment for truly positive patients [33]. This delicate cost-benefit ratio highlights the clinical necessity for screening tools that offer a better balance of sensitivity and specificity tailored to the HCM phenotype [34]. Conversely, the ESS had excellent specificity (92%) but poor sensitivity (24–27%), suggesting ESS using the ≥ 10 vs. < 10 threshold is less useful for identifying OSA in patients with HCM, many of whom may not exhibit excessive daytime sleepiness despite significant OSA.
The Berlin questionnaire performed better overall, with moderate sensitivity (65–73%) and specificity (57%), but still fell short of providing a robust predictive capability with the specified cutpoint.
Third, this study highlights the potential utility of a custom exploratory model tailored to patients with HCM. Incorporating easily obtainable variables; age, BMI, and waist-to-hip circumference ratio resulted in a model with a sensitivity of 60%, specificity of 84%, and an area under the curve (AUC) of 0.82 for predicting OSA of any severity. These findings emphasize the value of targeted screening strategies that account for the unique characteristics and risk factors of patients with HCM, rather than relying solely on tools validated in the general population.
That said, these estimates reflect apparent metrics from the training data and no internal or external validation assessment was performed. Performance in future validation studies is likely to be worse.
An ideal screening tool for OSA should balance ease of use, feasibility, reliability, reproducibility, cost, and diagnostic accuracy. While the ESS was developed to assess daytime sleepiness, it is important to note that a significant proportion of patients diagnosed with OSA did not complain of typical symptoms of sleepiness, suggesting the absence of an important marker of the disease seen in other populations [22]. Furthermore, while both the STOP-Bang and Berlin questionnaires incorporate subjective symptom assessments such as daytime fatigue, tiredness, or observed apneas, their predictive utility remains poor in this specific cohort, potentially because these general symptoms overlap heavily with the baseline manifestations of hypertrophic cardiomyopathy itself.
Importantly, there is a lack of non-invasive treatment options offering prognostic benefit in HCM, making the diagnosis of OSA particularly relevant. Patients with HCM are especially vulnerable to atrial fibrillation, which they tolerate poorly due to their reliance on atrial systolic function in the setting of a stiff, non-compliant ventricle [12]. Additionally, the sympathetic activation associated with OSA is concerning, as it may further exacerbate ventricular hypertrophy and increase the risk of arrhythmias [27].
Given these risks, and the potential for treatment of OSA to improve outcomes in this population, prioritizing sensitivity in a screening tool may be justified, even at the cost of a higher referral rate, to minimize the number of undiagnosed cases.
The implications of these findings are significant. Current American Heart Association (AHA)/American College of Cardiology (ACC) guidelines for HCM management emphasize the importance of addressing comorbidities such as OSA, given their impact on clinical outcomes and quality of life [13]. However, the lack of effective screening tools for OSA in this population presents a critical gap in care.
The results of this study suggest that existing tools may be insufficient for widespread use in clinical practice and support the development of customized approaches that integrate clinical and patient-reported data to improve screening accuracy. Our findings add to the growing body of evidence suggesting that OSA may be a critical, yet underrecognized, comorbidity in patients with HCM. Moreover, the observed associations between OSA and adverse outcomes, such as arrhythmias, heart failure progression, and overall mortality, underline the need for targeted diagnostic and therapeutic interventions in this high-risk population.
Strengths and limitations
Several limitations of this study merit consideration. The cohort consisted of a single-center population, and the generalizability of the findings to broader HCM populations warrants further investigation. Additionally, while the study demonstrates the utility of the custom exploratory model, external validation in independent cohorts is necessary before widespread clinical implementation.
Furthermore, while our exploratory analysis utilized traditional logistic regression, the complexity of phenotypic clusters and body habitus variations in patients with hypertrophic cardiomyopathy suggests that future validation efforts would greatly benefit from advanced computational strategies [35, 36]. Specifically, the integration of machine learning and deep learning applications which have increasingly demonstrated utility in complex pathology prediction and risk stratification across broader scientific disciplines, may be required to develop truly robust, automated screening algorithms for this high-risk population [37]. Finally, the reliance on polysomnography as the gold standard, while necessary, limits the scalability of screening efforts, particularly in resource-limited settings.
Conclusions
In conclusion, this study underscores the high prevalence of OSA in patients with HCM and the limitations of commonly used screening tools in accurately identifying affected individuals. The development and validation of tailored predictive models, such as the one proposed here, represent a promising step toward improving the identification and management of OSA in this high-risk population. Further research is needed to validate these findings in larger, diverse cohorts and to explore the impact of early detection and treatment of OSA on clinical outcomes in HCM.
Electronic Supplementary Material
Below is the link to the electronic supplementary material.
Supplementary Material 1 (DOCX 26.8 KB)