Previous investigators [3-7] have developed mathematical formulas for the clinical prediction of OSAS. These models used various measures, such as body mass index (BMI), neck circumference, oxygen saturation, witnessed apneas, and questionnaire data. Unfortunately, these models were limited because they did not account for craniofacial dysmorphism or disproportionate craniofacial anatomy [8-14], which are risk factors for the development of OSAS independent of obesity.

The craniofacial dysmorphism that leads to OSAS may involve a delayed growth of the mandible, producing the mandibular retroposition [9] commonly found in patients with OSAS. Mandibular retroposition is associated with posterior displacement of the tongue base. This narrows the upper airway, predisposing it to collapse and contributing to the development of OSAS. A high arched palate is also common in patients with OSAS because the posterior tongue displacement may force the lateral palatine processes to expand over the abnormally placed tongue before fusing at the midline. Evidence that these events may occur in patients with OSAS is found in the Robin sequence (the Pierre Robin syndrome) [15-17] and the Treacher Collins syndrome [17, 18], in which early mandibular hypoplasia from maldevelopment of the first branchial arch results in the above sequence of events in infants. The mandibular and palatal abnormalities seen in these infants invariably result in OSAS; not only do these cases provide an embryologic substrate for the craniofacial dysmorphism leading to OSAS, but they represent extremes of the craniofacial dysmorphism seen in adults with OSAS. These observations provide a rationale for including mandibular size and palatal height in our model for OSAS prediction.

The utility of any model is directly proportional to its accuracy and ease of use. We present a predictive morphometric model for OSAS that has a high sensitivity, high specificity, and good intermeasurer and test-retest reliability. The necessary measurements and calculations can be completed in an ambulatory setting (such as a clinician's office) within 5 minutes.

Methods

Top Methods Results Discussion Author & Article Info References

Model Derivation

The morphometric model was developed from a test sample of 30 representative patients who were studied with polysomnography: Fifteen had documented OSAS, and 15 had no evidence of OSAS. The criteria used to diagnose OSAS were 1) characteristic presenting symptoms (such as snoring, nocturnal breathing pauses, and excessive daytime sleepiness [19]; an Epworth Sleepiness Scale score > 10 was used to confirm the presence of excessive daytime sleepiness [20] and 2) a respiratory disturbance index (RDI) score (number of apneas and hypopneas per hour of sleep) of 5 or more on polysomnography.

Body mass index and neck circumference were incorporated into the model because univariate linear regression analyses in this study and previous studies [21, 22] indicated that these variables correlated significantly with the RDI. We noted correlation coefficients of 0.54 (P = 0.002) for the RDI compared with BMI and 0.53 (P = 0.002) for the RDI compared with neck circumference. These values are similar to those reported by previous investigators (0.51 for RDI compared with BMI and 0.53 for RDI compared with neck circumference [21]). A model that combines both BMI and neck circumference or that includes oral cavity dimensions has not been previously proposed. Various methods for measuring mandibular size and palatal height were tested for comfort, reproducibility, and rapidity and ease of use. These measurements, plus BMI and neck circumference, ultimately made up the model.

Separate prototype versions of the model were developed from different arrangements of these variables on the basis of trial-and-error experimentation and clinical experience, and they were combined with integer weights and mathematical operands. The model divided the test sample of 30 patients into OSAS and non-OSAS groups. The model cutoff value of 70 was calculated from the critical region containing the highest 5% of the non-OSAS group. This value was derived by adding the mean of the non-OSAS group to the product of the SE of the mean and the z-score of the highest 5% of the non-OSAS group by a one-tailed test [23]. An RDI of 5 or more was used as a cutoff value for OSAS; thus, the model treats the RDI as a dichotomous variable.

Model Measurements

The following morphometric data were incorporated in the model, and diagrams and detailed descriptions of the measurements are shown in Figure 1. Oral cavity measurements were made with electrocardiographic calipers after removal of the sharp tips and sterilization. These measurements were obtained by placing the caliper tips between the oral cavity structures to be measured, carefully removing the caliper, and measuring the distance between the caliper tips with a ruler.

View larger version (0K): [in this window] [in a new window]   Figure 1. Oral cavity measurements for the morphometric model. Top. Measurement of the palatal height by two separate calipers, labeled ABC and DEF. Angle ABC is a 20-degree angle between the maxillary and mandibular incisor tips with the vertex (B) of an externally placed caliper at the mandibular condyle. Angle DEF represents a caliper placed inside the oral cavity. The vertical distance DF is equal to P (palatal height) in millimeters. Bottom Left. Measurement from the mesial surfaces of the crowns of the second molars to obtain either Mx (maxillary intermolar distance) or Mn (mandibular intermolar distance) in millimeters. Bottom Right. Measurement of OJ (overjet), or the horizontal overlap of the crowns of the maxillary and mandibular right central incisors in millimeters.

The morphometric model is as follows: (Equation 1) where P is palatal height (in millimeters), or the distance from the dorsum of the tongue at the median lingual sulcus to the highest point of the palate, measured with the tongue in a relaxed position and the maxillary and mandibular incisor tips subtending an angle of 20 degrees from the mandibular condyle; Mx is the maxillary intermolar distance (in millimeters) between the mesial surfaces of the crowns of the maxillary second molars; Mn is the mandibular intermolar distance (in millimeters) between the mesial surfaces of the crowns of the mandibular second molars; OJ is the overjet (in millimeters), or the horizontal overlap of the crowns of the maxillary and mandibular right central incisors; BMI [24] is the body mass index (kg/m²; ideal BMI 25); and NC is neck circumference (in centimeters) measured at the level of the cricothyroid membrane.

(1)

Precise measurement of these anatomic distances is critical for the accuracy of the model. The bordered area of the formula reflects the contribution of craniofacial dysmorphism, as measured from the oral cavity, to the prediction of OSAS. The nonbordered area reflects the contribution of obesity, as measured by BMI and neck circumference, to the prediction of OSAS. The fraction NC ÷ BMI was selected to scale neck circumference relative to body size. The segment of the model enclosed within square brackets is limited to the larger of the two quantities: BMI –25, or zero. For example, if BMI is 25 or less, then [Max (BMI –25, 0)] is zero: That is, if BMI is 23, then (23 –25) = –2 and, because 0 is greater than –2,the maximum is zero. Therefore, if a patient is not obese (BMI 25), the contribution of the second part of the model to the final index value is nil; the final index value reflects only the degree of craniofacial dysmorphism.

Model Testing

After developing the model from the test sample of 30 patients, we prospectively tested it on 423 patients during a 6-month period to investigate its clinical utility and accuracy in predicting OSAS. The patients were self-referred or were referred from outside clinics; they had diverse primary symptoms, including fatigue, sleeplessness or sleepiness, rest-less legs, periodic limb movements, sleepwalking, and witnessed snoring or apneas, or they were recruited for research projects. Only patients who were visiting the clinic for the first time were included. Each patient had a clinical evaluation, which consisted of a discussion of symptoms and medical history as well as a general physical examination. The oral cavity, body weight, height, and neck circumference were measured during the examination.

At the completion of the clinical evaluation, a physician who was not involved in the study decided whether the patient should be scheduled for polysomnography; this physician was blinded to the results of the model measurements. Patients with unequivocal insomnia or a previous diagnosis (established on the basis of symptoms and polysomnography) of OSAS or another sleep disorder (for example, the restless legs syndrome, periodic limb movements, or sleepwalking) were not scheduled for polysomnography. The polysomnograms of the 300 patients included in the final study were independently reviewed by experienced physicians and technologists who were blinded to the results of the model measurements. The diagnosis of OSAS was then established in each patient on the basis of the criteria for the syndrome, as described above. Patients without symptomatic and polysomnographic evidence of OSAS made up the non-OSAS comparison group. The morphometric model results were matched to each patient; these data, combined with the polysomnographic data, were analyzed. Statistical analyses were done with Stat-View (Abacus Concepts, Inc., Berkeley, California), BMDP (BMDP Statistical Software, Inc., Los Angeles, California), and Splus [Math Soft, Cambridge, Massachusetts]) computer programs. To treat the model as objectively as possible, we cross-validated it as if it had been developed from the prospective 300 patients instead of from the initial sample of 30 patients. We used a linear logistic regression model and bootstrap methods [25], which use random distortions of an observed data set to make statements about the accuracy of statistical estimates. The 0.632 method [26], a bootstrap-based refinement of the cross-validation estimate for the accuracy of a prediction rule, was also used in the analysis. Receiver-operating characteristic (ROC) curves, area under the curve values, and error rates (number of errors ÷ number of patients) were used to compare the various models.

Intermeasurer reliability was tested by comparing the oral cavity measurements obtained by two physicians who measured the same 20 participants. These OSAS and non-OSAS patients made up a representative prospective sampling of our clinic population and were not included in the main study. One physician was experienced in performing the morphometric model measurements; the other was new to the procedure. The latter viewed a 5-minute videotape on the model and directly observed the experienced examiner measuring a test patient. Each examiner was blinded to the oral cavity measurements obtained by the other examiner and to the model results for each patient.

Test-retest reliability was studied on 10 patients by a physician who was experienced in obtaining the model measurements. The measurements were done on 10 patients who were measured at their initial evaluation and at a follow-up visit. At the follow-up visit, the physician performing the measurements was blinded to the measurements obtained during the initial evaluation.

Results

Top Methods Results Discussion Author & Article Info References

Data Analysis

Measurements of body weight, height, BMI, neck circumference, palatal height, maxillary and mandibular intermolar distances, and overjet in patients with OSAS and patients without evidence of a sleep-related breathing disorder are compared in Table 1. The two groups had significant differences in the individual components of the model as well as expected differences in the Epworth Sleepiness Scale score, RDI, and minimum oxygen saturation.

Of the 423 patients who were prospectively tested by using the model, the first 300 who were studied by polysomnography were selected for inclusion in the study (222 men and 78 women; age range, 15 to 75 years). One hundred seventy-four men and 66 women were white, 7 men and 3 women were African American, 24 men and 4 women were Asian or Pacific Islander, and 17 men and 5 women were Native American. Of the remaining 123 patients, 57 deferred polysomnography and 66 had a previous diagnosis of OSAS or another sleep disorder or had a history of upper airway surgery or use of orthodontic sleep apnea devices, nasal continuous positive airway pressure (CPAP), or supplementary oxygen therapy.

The model cutoff value of 70 divided the prospective sample of 300 patients into two groups. This bimodal distribution is shown in Figure 2. The means (±SD) for the OSAS and non-OSAS groups were 95.3 ± 21.2 and 61.6 ± 6.2, respectively. The sensitivity (248 of 254) was 97.6% (95% CI, 95% to 98.9%), the specificity (46 of 46) was 100% (CI, 92% to 100%), the positive predictive value (248 of 248) was 100% (CI, 98.5% to 100%), and the negative predictive value (46 of 52) was 88.5% (CI, 77% to 96%) (Table 2).

View larger version (0K): [in this window] [in a new window]   Figure 2. Frequency histogram of morphometric model values for the study groups with and without the obstructive sleep apnea syndrome (OSAS). The white bars represent the OSAS group; the striped bars represent the non-OSAS group. The dashed vertical line indicates the model cutoff value of 70. The OSAS group has 254 patients with a mean morphometric value of 95.3 ± 21.2 and an age range of 24 to 77 years. The non-OSAS group has 46 patients with a mean morphometric value of 61.6 ± 6.2 and an age range of 15 to 71 years.

In general, morphometric model values of 70 or more corresponded to patients with OSAS; values less than 70 corresponded to patients without evidence of OSAS (Table 2). The 46 patients in the non-OSAS group had the following primary diagnoses: insomnia (n = 17), primary snoring (n = 13), periodic limb movement disorder (n = 9), sleep disorder associated with a mood or anxiety disorder (n = 3), sleepwalking (n = 2), idiopathic hypersomnia (n = 1), and rhythmic movement disorder (n = 1). The model failed to predict OSAS accurately in 6 patients who had OSAS and the following RDI scores and model values: RDI, 8.1 and model, 69.2; RDI, 5.3 and model, 61; RDI, 5.7 and model, 67; RDI, 6 and model, 69.8; RDI, 13.5 and model, 69.2; and RDI, 6.1 and model, 67. Of these 6 patients, 3 were obese (BMI > 25 kg/m²) and 3 were not obese.

We tested our 300 patients by using either an ideal BMI or neck circumference cutoff value to predict OSAS; these measures have been used to predict OSAS in the past. Two-by-two contingency tables (Table 2) show the OSAS and non-OSAS groups separated by a BMI of 25 kg/m² and a neck circumference of 40 cm (estimated cutoff value [22]). It is recognized that populations with different base rates require adjusted cutoff values for BMI and neck circumference. Figure 3 shows ROC curves [27] representing different degrees of discrimination capacity for the morphometric model, BMI, and neck circumference. Area under the curve values are given in the legend of Figure 3; nonparametric bootstrap analysis [25] of the area under the curve values showed the following means: 0.996 ± 0.002 for the morphometric model, 0.938 ± 0.018 for BMI, 0.898 ± 0.023 for neck circumference, 0.058 ± 0.017 for the morphometric model – BMI, and 0.098 ± 0.023 for the morphometric model – neck circumference. These results strongly suggest that BMI or neck circumference alone is a poor predictor of OSAS compared with the morphometric model.

View larger version (0K): [in this window] [in a new window]   Figure 3. Comparisons of receiver-operating characteristic curves. Different discriminant capacities to distinguish patients with the obstructive sleep apnea syndrome (OSAS) from patients without OSAS by using the morphometric model (MM) (a), body mass index (BMI) (b), and neck circumference (NC) (c) are shown. The true-positive and false-positive rates for the entire locus of possible points for the selected discriminant capacity are shown in each graph. The area under the curve values are 0.996 for the morphometric model, 0.938 for BMI, and 0.898 for neck circumference.

An error rate of 0.02 was calculated for the model. A linear logistic regression model to y based on the predictors of neck circumference, neck circumference divided by BMI, maxillary and mandibular intermolar distances, palatal height, and overjet yielded an error rate of 0.03. Cross-validation by bootstrap analysis with the 0.632 rule [26] produced a conservative maximum error rate of 0.043. As these data show, the morphometric model offers a higher predictive value for distinguishing OSAS from non-OSAS conditions compared with a linear logistic regression model.

Intermeasurer and Test-Retest Reliability

The results of the intermeasurer reliability study were analyzed by intraclass correlation coefficients [28-30]. The means of the calculated model results for the 20 patients (12 men and 8 women; age range, 20 to 69 years) measured by the experienced and the nonexperienced examiner were 78.2 ± 21.9 and 79.1 ± 22.6, respectively. There was a high degree of reliability between the model results obtained by the experienced and the nonexperienced examiner: intraclass correlation coefficient [1, 2], 0.992 (CI, 0.981 to 0.997).

The results of the test-retest reliability study were also analyzed by intraclass correlation coefficients. The measurements were obtained a mean of 246 ± 184.6 days apart. The means of the calculated model results for the 10 patients (9 men and 1 woman; age range, 31 to 77 years) were 81.7 ± 21.3 during an initial evaluation and 81.2 ± 22.1 at a follow-up visit. There was a high degree of reliability between the model results obtained during the initial evaluation and those obtained during the follow-up visit: intraclass correlation coefficient [1, 2], 0.994 (CI, 0.975 to 0.998). Some patients showed increases in BMI and neck circumference between the initial evaluation and the follow-up visit, but these differences were minimal. Six of the 10 patients used nasal CPAP for the first time between the initial evaluation and the follow-up visit.

Discussion

Top Methods Results Discussion Author & Article Info References

The morphometric model enables a clinician to predict the presence or absence of OSAS rapidly and accurately in a given patient during an initial office visit. The sensitivity (97.6%) and specificity (100%) of the morphometric model are high compared with findings in previous studies done by using various mathematical formulas [4-7]. The morphometric model missed cases of OSAS in only six patients; that is, it gave six false-negative results. However, the mean calculated model result was 67.3 ± 3.3 for these six patients, which was close to the separation value of 70. In addition, the mean RDI for the six patients was low at 7.4 ± 3.1 (range, 5.3 to 13.5). Thus, calculated model values below but near the cutoff of 70 (for example, between 65 and 70) should be interpreted with caution.

Table 2 shows that some nonobese patients with craniofacial dysmorphism (which predisposes to OSAS) were correctly identified as having OSAS by the morphometric model. This contribution of the craniofacial component of the model to nonobese patients is shown in Figure 4. Conversely, Table 2 reveals that patients who are obese but do not have OSAS can also be identified by this model. These results suggest that the morphometric model was better at distinguishing patients with OSAS from patients without OSAS than were models that used BMI alone. In addition, our model fared better than models that use linear logistic regression or neck circumference alone.

View larger version (0K): [in this window] [in a new window]   Figure 4. The craniofacial versus the obesity component of the model. This scatterplot shows the craniofacial component (P + [Mx – Mn] + 3 x OJ) in the abscissa compared with the obesity component (3 x [Max (BMI –25, 0)] x [NC ÷ BMI]) in the ordinate. Plus signs indicate patients with a diagnosis of the obstructive sleep apnea syndrome (OSAS); circles indicate patients without OSAS. The solid line indicates the decision boundary at a craniofacial component cutoff value of 70, revealing the contribution of the craniofacial component of the model to nonobese patients. BMI = body mass index; Mx = maxillary intermolar distance; Mn = mandibular intermolar distance; NC = neck circumference; OJ = overjet.

The high degree of reliability between an experienced and a nonexperienced examiner indicates that clinicians can apply this model to their patients and obtain similar measurements. Repeated measurements of the same oral cavities by the same physician at different times were also similar, indicating that the selected oral cavity distances are stable over time and that the measurement procedure can be standardized. Six patients were prescribed nasal CPAP during the interim between measurements; although neck circumference mildly fluctuated, the change in the morphometric model measurements for pre- and postnasal CPAP application was not significant. However, given that the morphometric model is used to predict OSAS in patients before treatment, this latter observation is less important.

Certain conditions limit the accuracy of the morphometric model because of extreme values in one or more of the model's variables: 1) persons younger than 15 years of age or older than 80 years of age (because the permanent maxillary and mandibular second molars used in the model are present by the age of 15 years [31] and body mass is reduced in persons older than 80 years of age); 2) patients with the Marfan syndrome or major muscle disorders (because the weight-height disproportion found in these conditions affects body mass); 3) oral abnormalities, such as cleft palate, severe malocclusion, or reconstructive surgery; 4) coexisting serious or complicated medical conditions )because healthy adults in an ambulatory setting were used to derive this model); and 5) ethnicity. The patient sample was primarily white; few African Americans and Native Americans were involved in our study. However, the six patients with OSAS in whom OSAS was not predicted by our model were white.

Despite these limitations, the morphometric model seems to have clinical utility and predictive value for patients in whom OSAS is suspected at the initial evaluation. We do not suggest that this model be used to replace the gold standard of polysomnography in the evaluation of patients with OSAS. Rather, we envision primary care physicians using the model as a screening tool to help them decide which patients should be referred to sleep centers for further evaluation and treatment of OSAS. Additional evaluation is important because the model does not assess the severity of OSAS; polysomnography is necessary to distinguish patients with mild cases of OSAS, who would benefit from a dental appliance and weight loss, from those with moderate-to-severe OSAS, who require nasal CPAP or upper airway surgery. In addition to being useful for primary care physicians, the values obtained from this model provide additional data in the sleep specialist's armamentarium that may be useful in certain situations, for example, when the clinical history is ambiguous, the preliminary diagnosis of OSAS is uncertain, and the need for or urgency of a polysomnogram is in question.

Dr. Efron: Sequoia Hall, Room 106, Stanford University, Stanford, CA 94305.