This systematic review was carried out using the recommended methods established by the Cochrane Methods Group on Screening and Diagnostic Tests1 and by other authors.13,14
Literature search
Obstructive sleep apnea (OSA) is a highly prevalent sleep disorder; however, it remains underdiagnosed and undertreated. Although screening tools such as the Berlin questionnaire (BQ), STOP-BANG questionnaire (SBQ), STOP questionnaire (STOP), and Epworth sleepiness scale (ESS) are widely used for OSA, the findings regarding their diagnostic accuracy are controversial.
In order to include all available evidence, a systematic search of the literature was carried out through the Cochrane Library, MEDLINE (from 1950 to April 2009), EMBASE (from 1980 to April 2009), and CINAHL (from 1990 to 2009) using the search strategy that was designed for each database. The search strategy was developed and executed by an expert librarian and included the following free-text and index terms: “obstructive sleep apnoea or apnea”, “hypopnea or hypopnoea”, “OSA or SHS or OSAHS”, “sleep related respiratory disorder”, “sleep disordered breathing”, “Sleep Apnea Syndromes”, “Risk Assessment”, “Mass Screening”, “validation studies”, “questionnaire”, “sensitivity”, “specificity”, “screen”, “risk”, “score or scale”, and “mass screening” (Appendix). The search was extended to checking the reference lists of the included papers.
The search results were evaluated by two independent reviewers (A.A., A.K.) to find the eligible articles for inclusion. First, obviously irrelevant items were excluded by reviewing the title and/or abstract of the records. Next, the full-text articles of the remaining papers were retrieved and carefully evaluated to determine if they met the following eligibility criteria: 1) The study used a patient-based questionnaire as a screening tool for OSA in adult subjects (≥18 yr); 2) The questionnaire’s accuracy was evaluated by comparing its results with the results of a PSG as the “gold standard”15 for diagnosing OSA; 3) OSA was clearly defined as apnea/hypopnea index (AHI), apnea index (AI), or respiratory disturbance index (RDI) ≥ 5; 4) Information was adequately presented to allow the construction of a 2 by 2 contingency table; 5) The questionnaire and full text paper were written in English. The studies that were found ineligible and excluded from our study are listed in Table 2.
Assessment of methodological quality
The methodological quality of each paper was assessed independently by the authors (A.A., A.K.), and disagreements were resolved by arbitration of the senior author (F.C.). Validity criteria assessing internal and external validity were explicitly described and coded according to Cochrane Methods Group on Screening and Diagnostic Tests.A Internal validity included the following factors: study design, definition of the disease, blind execution of the index test (questionnaire) and the reference test (polysomnography), valid reference test, avoidance of verification bias, independent interpretation of test results. External validity consisted of the following items: disease spectrum, clinical setting, demographic information, previous screening or referral filter, explicit cut-offs, percentage of missing patients, missing data management, subject selection for polysomnography.
Data extraction and analysis
Data were extracted by two reviewers (A.A., A.K.) independently using standard data collection forms. In each study, the true positive, false positive, true negative, and false negative values were extracted for each AI, AHI, or RDI cut-off, and 2 by 2 contingency tables were constructed accordingly. The AI/AHI or RDI ≥ 5 were considered as diagnosis cut-offs for the existence of OSA. The AI/AHI or RDI ≥ 15 and 30 were considered as diagnosis cut-offs for moderate and severe OSA, respectively. Using the 2 by 2 contingency tables, we recalculated the following predictive parameters in each study: sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and prevalence of OSA in each study. Results were not extracted that would require extrapolations from equations, graphs, or derivations from figures or tables. Studies were excluded from the review if there was inadequate information to draw the 2 by 2 contingency tables (seven papers).11,16-20
A validated computer program for meta-analysis of test accuracy data (Meta-DiSc,21 version 1.4, Hospital Ramony Cajal, Madrid, Spain) was used to describe the overall accuracy of the questionnaires and to assess inconsistencies in accuracy parameters (sensitivity and specificity) across studies (heterogeneity). Accuracy parameters with a similar target population were analyzed together (sleep-disorder patients vs patients without history of sleep disorders). Inconsistency (12) > 50% was considered as significant. Subgroup analysis was carried out on studies using the same questionnaire to explore the reasons of heterogeneity. Meta-analysis was not carried out on other predictive parameters (PPV and NPV), as they are not related to the intrinsic quality of the questionnaires. However, all parameters were presented in the review for descriptive analysis.