Published ahead of print on January 24, 2003, doi:10.1164/rccm.200112-110OC
© 2003 American Thoracic Society A Decision Rule for Diagnostic Testing in Obstructive Sleep ApneaDepartment of Medicine, Division of Respiratory Medicine; Department of Community Health Sciences; and Department of Anesthesia, University of Calgary, Calgary, AB, Canada Correspondence and requests for reprints should be addressed to Dr. Willis H. Tsai, Rockyview General Hospital, 7007 14th St SW Calgary, AB, Canada T2V 1P9. E-mail: tsai{at}ucalgary.ca
Obstructive sleep apnea (OSA) is traditionally diagnosed using overnight polysomnography. Decision rules may provide an alternative to polysomnography. A consecutive series of patients referred to a tertiary sleep center underwent prospective evaluation with the upper airway physical examination protocol, followed by determination of the respiratory disturbance index using a portable monitor. Seventy-five patients were evaluated with the upper airway physical examination protocol. Historic predictors included age, snoring, witnessed apneas, and hypertension. Physical examination–based predictors included body mass index, neck circumference, mandibular protrusion, thyro–rami distance, sterno–mental distance, sterno–mental displacement, thyro–mental displacement, cricomental space, pharyngeal grade, Sampsoon-Young classification, and over-bite. A decision rule was developed using three predictors: a cricomental space of 1.5 cm or less, a pharyngeal grade of more than II, and the presence of overbite. In patients with all three predictors (17%), the decision rule had a positive predictive value of 95% (95% confidence interval [CI], 75–100%) and a negative predictive value of 49% (95% CI, 35–63%). A cricomental space of more than 1.5 cm (27% of patients) excluded OSA (negative predictive value of 100%, 95% CI, 75–100%). Comparable performance was obtained in a validation sample of 50 patients referred for diagnostic testing. This decision rule provides a simple, reliable, and accurate method of identifying a subset patients with, and perhaps more importantly, without OSA.
Key Words: obstructive sleep apnea • decision rule • diagnostic testing • physical examination Obstructive sleep apnea (OSA) occurs in 2% and 4% of middle-aged women and men, respectively (1). Traditionally, OSA has been diagnosed using overnight polysomnography (PSG), which is costly in terms of personnel, time, and money. Decision rules are sets of prospectively validated criteria that predict a clinical outcome, thus facilitating clinical decision-making. They are appealing as diagnostic instruments because of their low cost. Flemons and colleagues randomly selected a series of 180 patients referred to a tertiary sleep center (2). Increased neck circumference, hypertension, habitual snoring, and reports of nocturnal gasping/choking were predictive of OSA (PSG-apnea–hypopnea index of 10 hour-1 or more) using logistic regression modeling. Individuals with the highest clinical score (i.e., all four characteristics) had a likelihood ratio and post-test probability of OSA (apnea–hypopnea index of 10 hour-1 or more) of 5.17% and 81%, respectively. In contrast, patients with the lowest clinical score had a likelihood ratio of 0.25 and a post-test probability of OSA of 17%. A morphometric model developed by Kushida and colleagues had an OSA diagnostic sensitivity and specificity of 98% and 100%, respectively; however, selection bias was a potential concern (3). Nevertheless, the model illustrated the potential value of physical examination–based decision rules in clinical decision-making. Current decision rules have only intermediate diagnostic characteristics and are frequently too cumbersome, either arithmetically or logistically, for bedside implementation (2, 4–10). The objective of this study was to develop a standardized approach toward patient assessment in the OSA setting, with a specific emphasis on ease of use for the bedside clinician. Predictors of OSA were identified, and a decision rule was developed.
Subjects were recruited from the Alberta Lung Association Sleep Centre (ALASC), which is the major sleep center in Southern Alberta. Referrals to the sleep center were assigned to one of four sleep physicians on a consecutive basis; that is, there was no systematic physician–patient allocation. The two physicians (W.H.T. and J.E.R.) participating in this study managed approximately 40% of patients seen at the center. All referrals not meeting exclusion criteria were eligible for study. Exclusion criteria consisted of a refusal to undergo diagnostic testing, a previous assessment for a primary sleep disorder, insomnia, or a referral for a sleep disorder other than OSA. The diagnostic criteria for insomnia and other sleep disorders are standardized in the International Classification of Sleep Disorders (11). The study was divided into three distinct phases: feasibility, model development, and validation. The Conjoint Ethics Committee of the University of Calgary approved the protocol, and all patients provided informed consent. The selection of measurement variables was based on expert opinion (J.E.R., W.H.T., and J.M.D.) and upper airway scoring systems described in the anesthesia literature (12–15). Also included were known clinical predictors of OSA: hypertension, habitual snoring, nocturnal choking/gasping, witnessed apneas, age, alcohol use, and smoking history (2, 4–10). Clinical history was obtained by self-report or from the subject's bed partner. During the feasibility phase, patients underwent routine clinical assessment plus the upper airway physical examination protocol (UAPP), performed by two investigators (W.H.T. and J.E.R.). Unreliable or time-consuming measurements were eliminated from the UAPP based on a consensus view (W.H.T., J.E.R., and J.M.D.). During the subsequent model development phase, patients underwent a structured physical examination using the reduced UAPP, followed by assessment of the respiratory disturbance index (RDI) using a portable monitor. This portable monitor has been described previously and has excellent correlation and agreement with PSG (16). Data were collected prospectively, and a decision rule was developed using multiple logistic regression. The final predictive model was validated in a consecutive sample of all patients undergoing portable monitor-based testing at the ALASC.
Portable Monitoring
This monitor has been validated in a previous study and demonstrates excellent agreement with PSG (
UAPP The facial profile was categorized as retrognathic, neutral, or prognathic. To classify a profile, an imaginary line was created, joining the brow and maxilla. If the anterior chin was behind the line, retrognathia was said to exist. If the chin lay in front of the line, prognathia was present. The cricomental space was determined using a thin ruler to connect the cricoid cartilage to the inner mentum, with the head in the neutral position. The cricomental line was bisected, and the perpendicular distance to the skin of the neck was measured (Figure 1) . The use of a thin ruler (1 mm or less) was considered essential because thicker devices (e.g., tongue depressors) might influence measurement. The pharyngeal space (pharyngeal grade) was assessed using a four-point ordinal scale and is graphically presented in Figure 2 .
Palatal position or tongue size was assessed using the Sampsoon-Young classification system (12): grade 1 = good visualization of the soft palate, fauces, uvula, and tonsillar pillars; grade 2 = pillars obscured by the base of the tongue, but posterior pharyngeal wall clearly visible below the soft palate; grade 3 = soft palate and base of uvula visible; and grade 4 = soft palate not visible. Tonsillar enlargement (tonsillar grade) was assessed using a four-point ordinal scale: class I = tonsils absent; class II = tonsils do not extend beyond the palatopharyngeal arch; class III = tonsils at the palatopharyngeal arch; and class IV = tonsils extend beyond the palatopharyngeal arch.
Statistics A decision rule was created using the binary predictors derived from the reduced logistic regression model. Sensitivity, specificity, and positive and negative predictive values were then determined. Statistical analysis was performed using Stata 5.0 (Stata Corporation, College Station, TX).
Sample Size Determination
Feasibility Phase Twenty consecutive patients were assessed using the UAPP. Because the UAPP had to be acceptable to bedside clinicians, items were removed based on a consensus (W.H.T., J.E.R., and J.M.D.) impression of unreliability or excessive complexity. The reduced UAPP was then used for decision rule development. Physical examination measurements included mandibular length, thyro–rami distance, mastoid–medial clavicle distance, temporal mandibular joint–rami distance, rami–rami distance, thyro–mental distance, thyro–mental displacement, sternal–mental distance, sterno–mental displacement, inter–incisor distance, cricomental space, mandibular advancement, facial profile, pharyngeal class, Sampsoon-Young classification, and the presence of overbite.
Model Development Phase The prevalence of OSA among the 99 patients was 61%, 48%, 43%, or 33%, depending on whether an RDI diagnostic criterion value of greater than 5, 10, 15, or 20 hour-1 was employed. Clinical characteristics are presented in Table 1 . More detailed clinical characterization and physical examination findings are summarized in Tables E1 and E2 of the online supplement.
Univariate Predictors of OSA Univariate predictors of OSA were identified by simple logistic regression using clinical and physical examination features as independent variables and OSA (RDI of 10 hour-1 or more) as the dependent variable. The univariate predictors of OSA were age, snoring history, witnessed apneas, and hypertension (Table 2) . The physical examination measurements predictive of OSA were body mass index, neck circumference, mandibular length, thyro–ramus distance, thyro–mental displacement, sterno–mental displacement, cricomental space, pharyngeal grade, Sampsoon-Young class, and overbite.
No new predictive variables were identified when the data were independently analyzed using an RDI diagnostic criterion value of greater than 15 hour-1 to define OSA.
Model Development Cricomental space and pharyngeal grade were continuous variables. To obtain binary cut points, these measurements were cross-tabulated against a diagnosis of OSA, and optimal cut points were visually selected. A cricomental space of more than 1.5 cm and a pharyngeal grade of more than II were chosen.
Diagnostic Performance of the Decision Rule
Reliability Twenty patients underwent two independent assessments using the UAPP predictive variables, and agreement was determined: cricomental space of more than 1.5 cm (  = 1.0), the presence of overbite ( = 0.61), the presence of retrognathia ( = 0.22), tonsil enlargement ( = 0.73), pharyngeal narrowing (pharyngeal grade of more than II, = 0.78), and thyro–mental displacement ( = 0.58). The inter-rater agreement was high for all variables ( coefficient range: 0.58–1.00) except retrognathia ( = 0.22).
Validation Sample
In a consecutive series of 75 patients referred to a tertiary sleep center, a number of predictors of OSA were identified. This study confirms the results of previous investigators by identifying age, snoring history, witnessed apneas, hypertension, body mass index, and neck circumference as predictive of OSA. A number of physical examination–based predictors were also identified, and a decision rule was subsequently developed: a cricomental space of 1.5 cm or less, a pharyngeal grade of more than 2, and the presence of overbite. In patients with all three predictors, the decision rule had a positive predictive value, 95% (95% CI, 75–100%); negative predictive value, 49% (95% CI, 35–63%); sensitivity, 40% (95% CI, 27–56%); and specificity, 96% (95% CI, 82–100%). A cricomental space or more than 1.5 cm excluded the possibility of OSA (negative predictive value of 100% [95% CI, 75–100%]). Comparable performance was obtained in a validation sample of 50 patients referred for diagnostic testing. The interrater reliability of UAPP measurement variables was high ( = 0.58–1.00), other than for retrognathia. It was not possible to identify any single combination of variables that simultaneously had a high sensitivity and specificity for OSA. However, the use of a three-variable model to rule in a diagnosis of OSA and a cricomental space or more than 1.5 cm to exclude OSA holds considerable promise. Patients with a cricomental space of more than 1.5 cm or those meeting all criteria of the three-variable model accounted for 17% and 27% of the study population, respectively. Clearly, most (approximately 60%) patients fell into a diagnostic gray zone. Although this might appear to be a seemingly high number of nondiagnostic assessments, because of the high cost of diagnostic testing, if even a subset of patients either avoid diagnostic testing or are referred directly for initiation of CPAP therapy, important economic gains may be realized. A diagnostic instrument need not have a simultaneously high sensitivity and specificity to be of clinical value. For example, for diagnosing clinically significant ankle fractures, the Ottawa Ankle Rule has a specificity of only 50% but a sensitivity of 100%. Not all patients will meet the decision rule criteria, but in those who do, the need for an ankle radiograph can be eliminated. The Ottawa Ankle Rule has a diagnostic gray zone of approximately 70%, but in field testing, it is estimated that the rule has reduced the need for ankle radiography by 30% (19). Similarly, a cricomental space of more than 1.5 cm has been demonstrated to have a very high negative predictive value with respect to excluding patients with OSA. Several physical examination features that have been presumed predictive of OSA were subjected to formal evaluation. The predictive value of pharyngeal grade, Sampsoon-Young class, and overbite supports the suspicion that pharyngeal narrowing, a low-lying palate, and overbite are associated with OSA. In contrast, despite the commonly held belief, retrognathia, tonsil size, and change in palatal elevation with phonation (change in Sampsoon-Young classification with phonation) were not predictive of OSA. Moreover, measurements such as retrognathia could not be reliably determined between investigators. A recent study by Schellenberg and colleagues supports these findings. After controlling for body mass index and neck circumference, only lateral narrowing of the pharyngeal walls were predictive of OSA. Low-lying palate, retrognathia, and overjet were not found to be predictive (20). Although both clinical and physical examination–based predictors were incorporated into the initial regression model, only physical examination–based predictors formed the final decision rule. This suggests that for patients referred to a tertiary sleep center, the inclusion of clinical features adds minimal predictive value for diagnosing OSA beyond that of physical examination alone. In a consecutive sample of 300 patients referred to the Stanford University Sleep Centre, Kushida and colleagues developed a physical examination–based prediction index with a sensitivity and specificity of 98% and 100%, respectively (3). Body mass index, neck circumference, and intermolar distance were identified as predictive variables. However, the prevalence of OSA was 85%, which is considerably higher than the approximately 50% prevalence rate observed at most sleep centers. More significantly, BMI had a diagnostic sensitivity and specificity of 93% and 74%, respectively. Criticism could be raised with respect to the subject selection process in this study, particularly as the study population consisted only of patients referred to the study investigators. However, no systematic triaging of referrals existed, and the study prevalence of OSA (63% at an RDI of 10 hour-1 or more) was similar to the institutional prevalence reported in a previous study (16). Referral bias may exist; however, the ALASC is the only major referral site for sleep disorders in Southern Alberta. Patients seen at the center range from highly specialized cases to uncomplicated snorers. Moreover, the diagnostic performance of the decision rule in the model development and validation samples was virtually identical. As with most studies evaluating OSA, the choice of instrument and the criteria used to determine RDI could come under criticism. Redline and colleagues have clearly demonstrated that the choice of RDI definition can contribute to substantial variability in the identification of the disorder (21). Similarly, the choice of RDI diagnostic cutoff value may also influence the prevalence of disease. Until there is methodologic standardization of RDI determination and diagnostic cutoff values, this will be difficult to address. However, we attempted to deal with this issue by providing decision rule performance characteristics at a variety of cutoff values. A key feature of this decision rule is its ease of implementation. From a practical standpoint, a decision rule is only of value if it is adopted into routine clinical practice. To achieve widespread acceptability, a decision rule must be easy to interpret and executable without extraneous equipment or complex mathematic algorithms. This decision rule makes use of only three clinical predictors, all of which can be assessed with no more than a ruler. Measurements are categorical so as to avoid the need for arithmetic calculations. Its simplicity may derive from the ability to combine several independently predictive variables into a single measurement. Specifically, the cricomental space is a novel multidimensional measurement that probably incorporates diverse characteristics such as neck circumference, body mass index, hyoid bone position, neck posture, mandibular positioning, and possibly pharyngeal length. However, the decision rule requires prospective evaluation in different settings, specifically, at the primary care level. This decision rule is likely to have the largest clinical impact in settings where other sleep disorders are not under consideration (i.e., no further testing is necessary once OSA is excluded).
Conclusion
Supported by the Alberta Heritage Foundation for Medical Research. This article has an online supplement, which is accessible from this issue's table of contents online at www.atsjournals.org Received in original form December 4, 2001; accepted in final form January 16, 2003
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
METHODS
RDI [PSG - monitor] = 2.18 ± 12.34 [SE] hour-1). The sensitivity and specificity for OSA were 98% and 88%, respectively. 
