Use and Performance of the STOP-Bang Questionnaire for Obstructive Sleep Apnea Screening Across Geographic Regions

1Obstructive sleep apnea (OSA) is an increasingly prevalent health condition globally. Its prevalence varies depending on geographic and demographic factors. Nevertheless, worldwide, an estimated 425 million individuals aged 30 to 69 years have moderate to severe OSA, and in the general adult population, 80% to 90% of OSA is untreated and undiagnosed. Unrecognized OSA is a significant health concern, associated with various diseases,, public safety hazards, and all-cause mortality. Thus, it is imperative to consider strategies focusing on early diagnosis and treatment of OSA.

2The criterion-standard test for OSA diagnosis is laboratory polysomnography (PSG), but it is costly and inconvenient. Home sleep apnea testing (HSAT) is an acceptable alternative; however, accessibility is lacking in resource-limited areas. Since waitlists are long for patient assessment at a sleep clinic, an effective screening tool is necessary to triage patients.

3The STOP-Bang questionnaire was developed as an OSA screening tool consisting of 4 self-reportable (STOP: snoring, tiredness, observed apnea, and high blood pressure) and 4 demographic (Bang: body mass index [BMI; calculated as weight in kilograms divided by height in meters squared], age, neck circumference, and gender) items. In the initial validation study, at a score of at least 3, the STOP-Bang questionnaire demonstrated a sensitivity of 84%, 93%, and 100% to detect all OSA (apnea-hypopnea index [AHI] ≥5), moderate OSA (AHI ≥15), and severe OSA (AHI ≥30), respectively. Because of its high diagnostic accuracy, ease of use, and clear thresholds for risk stratification,, the STOP-Bang questionnaire has been used worldwide. Two STOP-Bang questionnaire items, BMI and neck circumference, are influenced by region-specific body characteristics, which may affect the performance of STOP-Bang questionnaire in different geographic areas. The objective of this systematic review and meta-analysis was to determine the utility of the STOP-Bang questionnaire as an OSA screening tool to assist in triaging patients referred to sleep clinics in different global regions.

4The protocol of this systematic review and meta-analysis was registered in the International Prospective Register of Systematic Reviews (PROSPERO) (CRD42020196952) and followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. One of us, a medical information specialist (M.E.), designed a literature search strategy using free-text and index terms (ie, stop-bang or stopbang) and systematically searched the following databases from January 2008 to March 2020 with no language restrictions: MEDLINE, MEDLINE In-process, Embase, EmCare Nursing, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, PsycINFO and Journals@Ovid with full-text searching using the Ovid search interface; Web of Science (Clarivate Analytics), Scopus (Elsevier), and CINAHL. A Web of Science citation search was run on the initial STOP-Bang validation article to capture publications in which it has been cited. A manual citation search was performed. Continued literature surveillance was done through August 2020. The electronic search strategy appears in eTable 1 in the Supplement.

5Two of us (B.P. and L.C.) independently screened titles and abstracts of identified studies. After initial exclusion, full-text articles were assessed for the following inclusion criteria: (1) assessment of the STOP-Bang questionnaire to screen for OSA in adults (age ≥18 years); (2) patients referred to sleep clinic; (3) laboratory PSG or HSAT results confirmed the OSA diagnosis; and (4) AHI or respiratory disturbance index (RDI) was used to diagnose and grade the severity of OSA. Exclusion criteria were (1) pregnant populations; (2) use of a modified STOP-Bang questionnaire; (3) no analysis of test characteristics at a STOP-Bang score of at least 3; and (4) inadequate description of methods (ie, no report of PSG device used and OSA diagnosed without an AHI cutoff) or insufficient data for meta-analysis. Disagreements were resolved by discussion and consensus among 4 of us (B.P., L.C., M.N., and F.C.).

6Two of us (B.P. and L.C.) independently extracted clinical and demographic data using a predesigned form. Internal and external validity of the included studies was assessed independently according to the Cochrane Methods group’s guidelines on screening and diagnostic tests. Discrepancies were addressed with another author (M.N.).

7Studies were organized into groups dependent on their geographical location (ie, North America, South America, Europe, Middle East, East Asia, South or Southeast Asia). Asian countries were separated into East Asia and South or Southeast Asia to match population demographic characteristics and mitigate factors affecting OSA prevalence (ie, craniofacial characteristics and body habitus differences).,

8Meta-analysis was performed using Review Manager version 5.3 (The Cochrane Collaboration) and Stata/SE version 14.2 (StataCorp). Summary statistics were computed for all variables of interest. Mean and standard deviation were used as appropriate for descriptive statistics. When applicable, frequencies and percentages were reported. Sample size was considered when calculating the means and standard deviations for age and BMI from individual studies.

9The following AHI cutoffs were adopted: all OSA (AHI ≥5), moderate to severe OSA (AHI ≥15), and severe OSA (AHI ≥30). A STOP-Bang score of 3 or greater was adopted as a threshold. A 2 × 2 contingency table was reconstructed for each AHI cutoff at a STOP-Bang score of 3 or greater for each study. The random-effects bivariate analysis model was used to combine results from individual studies to obtain the following log-transformed summary estimates with 95% CIs: sensitivity, specificity, and log scale diagnostic odds ratio (DOR).,, This method analyzed paired outcomes, ie, sensitivity and specificity values from individual studies, while incorporating correlations between outcomes. For cells with a 0 value, a correction factor of 0.5 was added to prevent problems associated with sensitivity and specificity equaling. The following test characteristics were recalculated with 95% CIs: prevalence, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and log scale DOR. The summary area under the receiver operating characteristic (AUC) curves were calculated by logistic regression. Combined test characteristics were recalculated for each regional group at each OSA severity cutoff. Forest plots were designed using a random-effects model, while log scale DOR and AUC curve analysis were presented to assess diagnostic ability. Heterogeneity (I2) was assessed using the χ2 test, with P <.05 indicating that heterogeneity was present.

10To determine the association between STOP-Bang score and probability of moderate to severe and severe OSA, posttest probabilities were calculated as previously described and combined from studies that assessed the performance of the STOP-Bang questionnaire at scores from 3 to 8. Results were produced as a bar graph.

11Metaregression and sensitivity analyses were performed on various subgroups for each factor at each OSA severity using the Open MetaAnalyst software for continuous variables (ie, age, sex, BMI, neck circumference, and prevalence) and categorical variables (sample size, study design, validation tool, OSA criteria, and regional/ethnic groups). This aimed to measure these variables’ associations with the combined estimates of sensitivity, specificity, and log scale DOR. Robustness of the combined estimates was checked by leave-one-out meta-analysis to assess individual study association with the combined estimates and heterogeneity. A 2-tailed P <.05 was considered statistically significant.

12Study characteristics and demographic data are summarized in Table 1 and eTable 3 in the Supplement, respectively.,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, The initial search yielded 3871 studies, with 162 additional studies identified through citations (eFigure 1 in the Supplement). After screening titles and abstracts, 2309 articles were excluded, and 58 full-text articles were assessed for eligibility. Forty-seven studies, with 26 547 participants, were included. Mean (SD) age and BMI among participants were 50 (5) years and 32 (3), respectively, with 16 780 (65%) men. Reasons for article exclusion are listed in eTable 2 in the Supplement. Results of internal and external study validity assessment are presented in eTable 4 and eTable 5 in the Supplement. Included studies showed low to moderate risk of bias after validity assessment and were used to answer our review question. Studies were organized into 6 groups: (1) North America (9 studies,,,,,,,,, 3507 participants); (2) South America (6 studies,,,,,, 10 709 participants); (3) Europe (10 studies,,,,,,,,,, 5679 participants); (4) the Middle East (11 studies,,,,,,,,,,, 3468 participants); (5) East Asia (4 studies,,,, 1665 participants); and (6) South or Southeast Asia (7 studies,,,,,,, 1519 participants).

13The combined test characteristics in the sleep clinic at a STOP-Bang threshold of 3 or greater are presented in Figure 1, Table 2, and eFigure 2 in the Supplement. The prevalence rates of all OSA, moderate to severe OSA, and severe OSA were 80% (95% CI, 80%-81%), 58% (95% CI, 58%-59%), and 39% (95% CI, 38%-39%), respectively. A STOP-Bang score of at least 3 had excellent sensitivity (combined sensitivity, 91.4; 95% CI, 86.4-94.6) to detect all severities of OSA. The false-negative rate was 8% (6%-11%). Moreover, the STOP-Bang questionnaire demonstrated a high discriminative power, indicated by NPVs, to exclude moderate to severe (77%, 95% CI, 75%-78%) and severe (91%, 95% CI, 90%-92%) OSA. Moderately low specificity was seen, with 28% (95% CI, 22%-34%) for moderate to severe OSA and 24% (95% CI, 19%-30%) for severe OSA.

14For all OSA, the combined prevalence ranged between 75% (95% CI, 73%-77%) in East Asia and 87% (95% CI, 86%-88%) in the Middle East (Table 2; eFigure 3 and eTable 6 in the Supplement). The combined sensitivities were excellent in all regions (>90%) and combined specificity values were moderately low, except in South or Southeast Asia (sensitivity: 81%; 95% CI, 68%-89%; specificity: 60%; 95% CI, 46%-73%). Similarly, PPVs were consistently high among all regional groups, ranging between 81% (95% CI, 80%-82%) in Europe and 87% (95% CI, 86%-88%) in South America. The AUC curves ranged between 0.72 (95% CI, 0.68-0.76) in North America and 0.78 (95% CI, 0.74-0.81) in Europe, with an exception in East Asia (0.56; 95% CI, 0.52-0.60).

15For moderate to severe OSA, the STOP-Bang questionnaire demonstrated an excellent combined sensitivity (>93%; eg, North America, 94%; 95% CI, 93%-96%) and low specificity (<33%) in North America, South America, Europe, and the Middle East. In East Asian, sensitivity was 93% (95% CI, 84%-97%); in South or Southeast Asia, sensitivity was 89% (95% CI, 73%-96%). In South or Southeast Asia, specificity was 45% (95% CI, 30%-61%) (Table 2; eFigure 3 and eTable 6 in the Supplement). A STOP-Bang score of at least 3 had a moderate discriminative power to exclude moderate to severe OSA, with NPVs ranging between 66% (95% CI, 65%-67%) in South America and 78% (95% CI, 74%-81%) in North America. DOR values were comparable across groups for moderate to severe OSA, ranging from 1.62 (95% CI, 1.27-197) in South America to 2.24 (95% CI, 1.11-3.37) in the Middle East. The diagnostic accuracy of a STOP-Bang score of at least 3 to detect moderate to severe OSA indicated by the AUC curve was high in North America, Europe, and the Middle East (eg, North America: 0.89; 95% CI, 0.85-0.91) but lower in the East Asian (0.52; 95% CI, 0.48-0.56) and South or Southeast Asian (0.70; 95% CI, 0.66-0.74) groups.

16The STOP-Bang questionnaire demonstrated an excellent combined sensitivity, ranging between 90% (95% CI, 87%-93%) in East Asia and 99% (95% CI, 97%-99%) in Europe, with lower combined specificities (Table 2; eFigure 3 and eTable 6 in the Supplement). The STOP-Bang questionnaire demonstrated a high discriminative power to exclude severe OSA in all regional groups, as NPVs ranged between 82% (95% CI, 76%-87%) in East Asia and 96% (95% CI, 94%-98%) in Europe. The AUC curve value was highest in the European group (0.96, 95% CI, 0.93-0.97).

17Metaregression and sensitivity analysis of continuous variables slightly changed the combined estimates but did not affect overall inference of the results (Table 3; eTable 7 in the Supplement). Similarly, analysis of the categorical variables slightly changed the combined estimates but did not impact the final inference of our results. Leave-one-out meta-analysis showed no individual study greatly affected the results.

18For all OSA, as the STOP-Bang score increased from 3 to 8, the specificity increased from 40.6% (95% CI, 39.3%-42.0%) to 99.7% (95% CI, 99.0%-99.9%), while sensitivity decreased from 89.2% (95% CI, 88.7%-89.6%) to 3.6% (95% CI, 3.0%-4.3%) (eTable 8 in the Supplement). The PPVs and NPVs showed similar trends. This was similar for moderate to severe OSA and severe OSA.

19In all included studies, as the STOP-Bang score increased from 3 to 6, the probability of moderate to severe OSA increased from 65% to 75% (Figure 2). Similarly, the probability of having severe OSA with a STOP-Bang score of 3 was 45%, and with a score of 6, it was 57%.

20This systematic review and meta-analysis found that the STOP-Bang questionnaire was a useful screening tool to triage patients with suspected OSA in sleep clinics in different geographic regions. Its high sensitivity would help to identify those at risk for all, moderate to severe, and severe OSA, while high NPVs would allow clinicians to rule out severe OSA in individuals with a STOP-Bang score of 2 or less. The AUC was clinically significant at each AHI threshold, indicative of its global screening utility.

21The prevalence for all, moderate to severe, and severe OSA at sleep clinics globally was high at 80% (95% CI, 80%-81%), 58% (95% CI, 58%-59%), and 39% (95% CI, 38-39). Patients referred to sleep clinics are already suspected of having OSA, which explains the high prevalence and diagnostic accuracy in our study. In this setting, the STOP-Bang questionnaire fulfills the role of a screening tool that may allocate resource-limited care according to severity. While STOP-Bang scores of 3 to 5 had similar posttest probabilities (ie, 65%-66%), individuals who scored 6 or greater had a 75% posttest probability of being diagnosed with moderate to severe OSA and should be investigated more urgently. The implications of our findings are that the STOP-Bang questionnaire is a helpful tool when triaging patients who were referred to sleep clinics.

22Groups were organized relative to country and geographic proximity. Asian countries were separated into East Asia and South or Southeast Asia to minimize external factors that influence OSA prevalence (ie, sleep clinic referral patterns and burden of other comorbidities) as well as population-specific anatomic and phenotypic features that influence OSA development and severity. Although OSA is prevalent globally, certain regions report higher prevalences, ie, China, the United States, Brazil, and India. These differences influence prevalence-dependent variables, such as pretest and posttest probabilities. We have addressed this by organizing studies based on geography to accurately represent OSA prevalence in each global region. Moreover, DOR values, which are independent of prevalence, were comparable across groups for moderate-to-severe OSA, ranging from 1.62 (95% CI, 1.27-197) in South America to 2.24 (95% CI, 1.11-3.37) in the Middle Eastern group.

23The North American group showed a high sensitivity of 94% (95% CI, 93%-96%) and AUC of 0.89 (95% CI, 0.85-0.91) to detect moderate to severe OSA with lower specificities. Similar results were found for the South American, European, and the Middle Eastern groups. Asian groups showed lower sensitivity and AUC values at the same AHI cutoff, ie, 93% (95% CI, 84%-97%) and 0.52 (95% CI, 0.48-0.56) in the East Asian group; 89% (95% CI, 73%-96%) and 0.70 (95% CI, 0.66-0.74) in the South or Southeast Asian group; both groups had higher specificities.

24Variation in the STOP-Bang questionnaire performance among these geographic groups may be explained by obesity and craniofacial characteristics. Obesity prevalence and patterns of adiposity vary among ethnic groups. In European, North American, South American, and Asian populations, AHI increases with BMI; however, BMI is most associated with OSA severity in the latter 2 regions., In Asian populations, higher body fat percentages and visceral fat are associated with a lower BMI.,, This fat distribution pattern is associated with increased tongue size and lower lung volumes that may contribute to nocturnal upper airway collapse, such that Asians present with OSA at lower BMIs and are more sensitive to BMI increases. Changing the STOP-Bang BMI cutoff to 30, closer to the obesity cutoff for Asian populations, has shown increases in sensitivity without compromising specificity for OSA detection.,,

25The craniofacial features are important in OSA pathogenesis and are unique depending on location. Chinese individuals have a smaller, narrower retropalatal airway and soft tissues with greater susceptibility to pharyngeal collapse. Differences are found between ethnic groups, such that Asian populations have more craniofacial restriction, while European groups have larger upper airway soft tissues (ie, larger tongue and parapharyngeal fat pads) with smaller airway dimensions. Similarly, North American and South American populations also showed increased neck circumference, which is associated with greater odds of a higher Mallampati class, indicative of oropharyngeal crowding, while Asian groups do not. This may explain the excellent clinical utility seen in each regional group other than the East Asian population. Further research is necessary to investigate the contribution of ethnic and genetic factors to OSA prevalence and severity.

26Ideally, a diagnostic test has high sensitivity with sufficient specificity, is inexpensive, and allows for early disease identification. High sensitivity is helpful to rule out OSA and/or to prioritize patients referred to sleep clinics; however, it is an incomplete study estimate, especially in the presence of study heterogeneity and bias. False-negativity is a component of sensitivity that provides a robust summary estimate and may be considered independent of prevalence. In our meta-analysis, the false-negative rate was 8% (6%-11%). Considering specificity may improve accuracy. A moderate to low specificity at a STOP-Bang score of at least 3 may subject patients to false-positive results, increasing unnecessary costs and caseloads at sleep clinics. Nevertheless, with the goal of diagnosing and preventing mortality and morbidity associated with OSA as well as the low risk of investigating potential OSA, false-positives are of secondary importance.

27This study has limitations. First, included studies used various laboratory PSG and HSAT devices to diagnose OSA. Although often comparable, some inconsistences may exist. Second, risk of bias for internal validity was occasionally unclear, specifically whether PSG and STOP-Bang results were masked and interpreted independent of clinical information. Third, metaregression analysis showed positive confounders contributing to differences between groups, thus enhancing combined estimates of our results. Some confounders were related to components of the STOP-Bang questionnaire (age, sex), while others were not (OSA definition). Next, although the STOP-Bang questionnaire had high NPVs for moderate to severe and severe OSA, it is possible to miss patients with OSA who have lower AHIs because of lower NPVs in patients with mild OSA. Additionally, our results are unique to the sleep clinic and do not apply to other populations owing to inflated OSA prevalence, given that patients were screened before referral for further OSA investigation. Furthermore, moderate-to-high heterogeneity exists. This may be attributed to methodological heterogeneity and variability in study location; however, all studies were grouped with efforts to unify populations. A random-effects model for meta-analysis was performed because of the suspicion of high heterogeneity. Despite these limitations, our study provides an interpretation of the available literature on the STOP-Bang questionnaire as an OSA screening tool among different global populations in the sleep clinic setting.

28In this study, the STOP-Bang questionnaire showed high sensitivity and NPVs at all AHI cutoffs. These findings indicate that it is a useful OSA screening tool for triaging patients in sleep clinics in various global regions.