Obstructive sleep apnea subtypes by cluster analysis Masafumi Tsuchiya, DDS, DDSc," Alan A. Lowe, DMD, PhD, FRCD(C), b Eung-Kwon Pae, DDS, MSc, c and John A. Fleetham, MD, FRCP(C) a Vancouver, British Columbia, Canada
A sample of 84 adult male patients with obstructive sleep apnea (OSA) were classified by a cluster analysis on the basis of apnea index (AI) and body mass index (BMI). Demographic, cephalometric, tongue, soft palate, and upper airway-size data were evaluated for the two subgroups of OSA patients and for 18 control subjects. One OSA group consisted of 43 patients with a high AI and low BMI ratio, the other group was comprised of 41 patients with a low AI and high BMI ratio. The patients with a high AI and low BMI ratio had retruded mandibles with high mandibular plane angles and proclined lower incisors. The patients with a low AI and high BMI ratio had inferior hyoid bones and large soft palates. A multiple regression analysis was performed between AI (the dependent variable) and the other variables (independent variables) for each of the subgroups. In the patients with a high AI and low BMI ratio, a high AI was related to a large skeletal anteroposterior disqrepancy, a steep manidbular plane, and an inferoanterior position of the hyoid bone. In the pat!ents with a low AI and high BMI ratio, a high AI was related to a large tongue and a small upper airway. In both groups, BMI was the major contributor to AI. In conclusion, these two groups may represent distinct subgroups of OSA patients and provide some insight into the contribution of obesity to the pathogenesis of OSA. The patients wit h & high AI and low BMI ratio have a skeletal mismatch, whereas the patients with a low AI and high BMI have atypical soft tissue structures. (AM J ORTHOD DENTOFACORTHOP 1992;101:533-42.)
O b s t r u c t i v e sleep apnea (OSA) occurs because of recurrent occlusion of the upper airway during sleep. The majority of patients with OSA are obese; however, some patients with OSA are not obese, and only a small proportion of oyerweight subjects develop OSA. A previous overemphasis on obesity may have caused some investigations to overlook other potential factors that may predispose to this condition. Recently, several studies '3 have emphasized that upper airway collapse during sleep is a result of interrelated structural and functional factors. Obesity plays a significant and important role in the occurrence of the upper airway occlusion during sleep. 24 Intuitively, OSA patients with and without obesity are likely to have a different pathogenesis. However, there have been no previous reports that classify the OSA patient population on the basis of weight and the influence of the various anatomic factors. The purpose of this study is to use cluster analysis techniques to classify patients with OSA on the basis of the amount of obesity and the severity of their OSA,
Fromthe Universityof BritishColumbia. This projectwas supportedby grants fromthe MedicalResearchCouncilof Canadaand the BritishColumbiaLungAssociation. 'VisitingClinicalAssistantProfessor.Departmentof ClinicalDentalSciences. bProfessorand }lead, Departmentof Clinical DentalSciences. ~GraduateStudent, Departmentof ClinicalDentalSciences. aAssociateProfessor,Departmentof Medicine, UniversityHospital. 8/1/27742
,r_.._
Fig. 1. Reference points, contours, and planes used to identify hyoid bone measurements. H, The most anterior and superior point on the hyoid bone. C3, The inferior anterior position on the third cervical vertebrae. RGn, The most inferior posterior point on the mandibular symphysis. 1, MPH; 2, HH1; 3, HRGn; 4, C3H. For complete definitions, please see METHODS section. and to evaluate craniofacial structure and demographic data by means of multiple regression analyses.
.MATERIAL AND METHODS Patients A group of 84 patients with OSA (obstructive sleep apnea) and 18 control subjects without symptoms were evaluated.
533
Tsuchiya et al.
534
Am. J. Orthod. Dentofac. Orthop. June 1992
I. Differences o f d e m o g r a p h i c variables b e t w e e n the two subgroups and b e t w e e n each subgroup and controls
Table
G-A (n = 43) Variables
Mean
AI (apneaslhr) TAT (%) Weight (kg) BMI (kg/m 2) Age (years)
40.0 28.9 86.6 28.2 47.3
]
G-B (n = 41)
SD
Mean
20.2 15.7 15.0 5.1 13.8
19.4 12.7 98.2 34.6 46.7
]
Control (n = 18)
SD
Mean
I SD
13.9 10. I 18.3 9.0 9.4
--77.0 24.8 31.8
--8.0 1.9 8.2
Probability G-A vs G-B
[ G-A vs Cont
0.000"* 0.000"* 0.002** 0.000"* 0.833
--0.013" 0.011 * 0.00(3**
] G-B vs Cont --0.000"* 0.000'* 0.000'*
*Significant at the 5% level. **Significant at the 1% level.
T a b l e II. A data matrix o f p variables determined for n o b s e r v a t i o n s for cluster analysis
Variables
1
xu
x12
~ 9 . .
xlT,
.....
xlp
2
X21
X22
....
X21~
.....
X2p
r
Xtl
xl2
....
Xrk
.....
Xrp
n
Xal
Xr
....
Xnk
.....
Xi~
.0
To obtain more homogeneous data, female subjects and subjects under the age of 17 were excluded. Edentulous subjects were also excluded from this study as complete cephalometric analyses were not possible. Patients were considered 1o have OSA if they had an apnea index (AI) of > 5 apneas per hour of total sleep time during an overnight polysomnogram. A detailed description of the current overnight polysomnographic monitoring techniques has been completed .5 The control subjects were men with Class 1 malocclusions who revealed n o symptoms suggestive of OSA, such as loud snoring, repeated nocturnal apnea, or excessive daytime sleepiness, when evaluated by the attendhzg physician. Demographic data for the two groups of OSA patients and control subjects are provided in Table I. Cephalometric and upper airway computed tomography (CT) examinations were performed with identical equipment on all subjects.
Cephalometry A single cephalometric radiograph'was taken for each of the subjects before treatment. The subject was seated in an upright position with the head fixed by earrods in the Frankfort horizontal plane parallel to the floor. The dorsum of the tongue was coated with Microtrast Oesophageal Cream (Nicholas Laboratpries Ltd., S!ough, United Kingdom) to enhance-the
outline of the tongue and pharyngeal soft tissues. Each patient was instructed to swallow, lightly contact the back teeth with the lips in repose, and to breathe normally. Tracings were constructed of the lateral head films and traditional contours and points were digitized. In addition, several hyoid bone measurements were determined (Fig. I)? 1. MPH: Linear distance along a perpendicular.from H (the most anterior and superior point on the hyoid bone) to the mandibular plane. 2. tlii1: Linear distance between H and a perpendicular to the C3 (inferior anterior position on the third cervical vertebrae) to retrognathion (RGn: most inferior posterior point on the mandibular symphysis). 3. HRGn: Linear distance between H and RGn. 4. C3H: Linear distance between H and C3.
Upper airway computed tomography Each subject underwent a Siemens DR 2 CT scan analysis. Before the scan, the subject was positioned supine on the examination table in such a position that the Frankfort horizontal plane was perpendicular to the floor. Each subject was instructed to relax (teeth apart) and to continue regular breathing, but not to swallow during any one scan. The patient was evaluated while awake. Scans were obtained at 8 mm
Volume 101
Obstructive sleep apnea subO'pes 535
Number 6
loo:
80A
G-A o
~60. C
~ o o
t~ o~
o o
o
o
o o oo [] o
.r 40-
o
o~
o
9
9 9
,1= u9 e w9 ~ ~0
O
20
o
oo O
o
0 o a...... t,= 9 i
20
.
00 9
9
9 9
G-B
9
00
i
|
4b BMI ( k g / m 2) 30
so
r,o
Fig. 2. Scattergram for two subgroups segregated by cluster analysis. [ ] = G-A; 9 = G-B.
intervals from the Frankfort plane to a level below the sixth cervical vertebrae. Linear attenuation coefficients for each pixel within each scanned section were stored digitally on a magnetic disk and were available for video display as an image analog to the digital distribution. A fixed window level of 30 was selected to emphasize muscle and a window width of 210 was used. Since variations in window level and width are known to change the appearance of any image by altering the edges of a particular structure, these parameters were fixed for all subjects. A magnification factor of 3.5 was used, and a center point was fixed before obtaining the first scan. Tracings were made on acetate paper of each of the slices for airway, tongue, and soft palate structures by one investigator and confirmed by another. The tongue was delineated to include all instrinsic and extrinsic muscles (genioglossus, hyoglossus, and styloglossus). Tracings were also made of the outlines of the nasopharynx, the oropharynx, and the hypopharynx. The nasopharynx was defined to extend from
the roof of the airway to the level of the hard palate. The oropharynx extended down to the uppermost tip of the epiglottis, and the hypopharynx was traced to the slice that most closely corresponded to the most inferior point of the sixth cervical vertebrae. The contour of each structure was digitized, and the data were stored in the computer (1000 E series, Hewlett-Packard Co., Andover, Mass.). Calculations of surface area and volume for all structures were completed. Partial airway volume (PAirVol) represented a combination of nasopharynx, oropharynx, and hypopharynx. A three-dimensional reconstruction of each of the structures was also performed. On the basis of the CT slices, details of the threedimensional analyses for these structures have been previously documented.6 Statistical
analysis
Univariate and multivariate analysis of the data were perf0rmed on the data matrix of 102 subjects with 5 demographic,
536
Ts,tchiya et al.
Am. J. Orthod. Dentofac. Orthop. June
1992
Table III. Differences o f c e p h a l o m e t r i c variables b e t w e e n the two subgroups and b e t w e e n each subgroup and controls
G-A (n = 43)
G-B (n = 41)
Variables
Mean
I SD
Mean
I SD
SNA (*) MxUL (mm) SNB (o) MdUL (ram) ANB (~ ULD (mm) SNMP (~ GoA (~ OB (mm) OJ (mm) UltoNA (*) UltoNA (ram) ~LltoNB (~ !LltoNB (ram) MPH (mm) HH1 (ram) HRGn (mm) C3H (mm)
79.7 96.7 76.1 124.5 3.6 27.8 37.9 130.3 2.6 4.8 20.7 5.2 24.8 6.1 23.3 12.2 39.8 43.9
4.8 5.9 4.2 6.9 2.9 5.8 5.9 6.8 2.9 2.5 7.5 3.7 5.8 3.1 8.2 7.9 6.8 4.6
81.2 97.5 78.0 124.8 3.2 27.3 34.5 127.6 3.4 4.9 19.1 4.4 22.8 4.5 27.3 18.1 43.7 45.7
4.2 4.9 4.4 6.2 3.1 5.9 6.5 5.9 2.7 2.4 8.1 3.6 7.4 2.9 8.6 8.2 7.9 412
Control (n = 18) Mean
I
83.0 99.8 80.9 127.9 2.1 27.8 31.0 127.2 3.1 4.2 23.4 4.5 20.2 3.5 19.1 9.6 42:5 41.9.
Probability
SO
G-A vs G-B
3.8 3.8 4.3 7.0 3.2 6.5 7.1 7.3 3.1 2.2 14.1 3.3 7.0 2.9 7. ! 6.1 9.2 3.6
0.136 0.506 0.043* 0.820 0.544 0.714 0.013" 0.058 0.185 0.947 0.342 0.310 0.185 0.018" 0.034* 0.001"* 0.019" 0.079
I G-a vs Cont 0.012' 0.049* 0.000'* 0.114 0.080 0.966 0.000"* 0.I 1! 0.506 0.304 0.325 0.482 0.010' 0.004** 0.075 0.229 0.214 0. ! 10
I G-B vs Cont 0.121 0.088 0.022* 0.130 0.225 0.754 0.072 0.768 0.747 0.297 0.138 0.917 0.197 0.238 0.001"* 0.000"* 0.629 0.002**
*Significant at the 5% level. **Significant at the !% level.
18 cephalometric, and 6 upper airway CT variables. The data were standardized before a cluster analysis since variables with large variances tended to have more effect on the resultant clusters than those with small variances. To classify patients with OSA according to the degree of obesity and the severity of OSA, a cluster analysis was performed on the basis of standardized AI and body mass indices (BMI). A cluster analysis is a multivariate procedure used to detect natural data groupings on the basis of the similarity of d a t a / ' The investigation can classify a set of objects into subgroups although neither the number of subgroups nor the members of the subgroups are known. Initially each object is considered as a separate cluster, then the two most similar objects are joined to form a cluster. The amalgamating process continues in a stepwise fashion (joining objects or clusters of objects) until a single cluster is formed that contains all the objects. Table II shows a data matrix of p variables, [x~,x.,. . . . . x~. . . . . xp] determined for n observations. The value of a correlation coefficient (R) was specified as the measurement of similarity for clustering observations in this report. The similarity "S,.," between two cases "r,s" is defined as follows: S,=R, R,,, = X,,,/X/2 X, 9 X , where E (x,,, -
X,., = E (x~ - -x,) (xa - -x,), X,, = -x,) 2. The similarity between clusters is calculated
in accordance with a linkage method. There are various linkage methods to combine clusters, i.e. single linkage, complete linkage, average linkage. In this study the average linkage rule was used. The average similarity is the arithmetic average of the similarities Su with all possible pairings of the observations between the two clusters: 1~ ~ SkdKL where observations k are in the first cluster and observations 1 are in the second cluster, and K and L are the number of observations in the two clusters. To investigate the reproducibility of the cluster analysis in this report, a subsample, which consisted of 66 patients randomly selected from the original 84 patients with OSA, was subjected to the same clustering procedure as the original one.
To evaluate differences in craniofacial structure and demographic data, comparisons of means for cephalometric, upper airway CT, and demographic variables were completed by the Student's t test. The subdivided groups and differences between each subgroup and a comparable control group were evaluated. A stepwise method, which is one of the linear multiple regression analyses, was also performed between AI (the dependent variable) and cephalometric, upper ai~vay CT, and demographic variables (independent variables) for each of the subdivided patient groups to evaluate any interactions between the severity of OSA and craniofacial structure and
Volume101
Obstructivesleepapneasubtypes
Number 6
537
Table IV. Differences of upper airway CT variables between the two subgroups and between each subgroup
and controls G-A (n = 43) Variables TngVol (cc) TngSA (cm:) SPVol (cc)
SPSA (cm") TAirVol (co) PAirVol (cc)
Mean 87.8 106.5 9.1 30.6 39.7 21.1
I
SD 12.6 12.6 2.9 7.0 10.4 6.6
G-B (n = 41) Mean 93.7 110.3 10.2 33.7 38.8 19.8
I
SD
Control (n = 18) Mean
17.6 18.3 2.3 5.9 9.7 6.3
79.6 89.0 7.0 25.5 34.6 16.1
I
Probability
SD
G-A vs G-B
15.9 11.6 2.5 6.2 10.8 7.5
I
I
i
i
] G-A vs Cont
.179 .268 .077
.030* .687 .358
[ G-B vs Cont
.035* .000"* .009**
.005** .000"* .000"*
.010" .093 .011"
.000"* .150 .051
Definition of abbreviations: TngVol = tongue volume; TngSA = tongue surface area; SPVol = soft palate volume; SPSA = soft palate surface area; TAirVol = total airway volume; PAirVol = partial airway volume. *Significant at the 5% level. **Significant at the 1% level.
demographic data. The stepwise multiple regression analysis is a statistical method for predicting the value of a response (dependent) variable from a collection of predictor (independent) variable values. It can also be used for assessing the effects of the predictor variables on the response. An adjusted squared multiple correlation coefficient (adj. R~) indicates a proportion of the variation in a dependent variable explained by a combination of independent variables. Statistical evaluations were all completed at the 5% level of significance with SAS/STAT and SYSTAT software programs. RESULTS
An intracluster similarity of 30.04% was determined when the patient population under study was divided into two clusters. It was 48.02% when the population was divided into three clusters and >50% when it was divided into more than three. Consequently, on the basis of an intracluster similarity of 30.04%, the patients with OSA were subdivided into two groups. One group consisted of 43 patients (G-A), and the other was comprised of 41 patients (G-B). The sample distribution is illustrated in Fig. 2. Cluster analysis of the subsamples produced two very similar clusters of patients at the same intracluster similarity as the original clustering; only five patients mismatched with the original clusters. The procedure of cluster analysis employed in this report was confirmed to be reproducible. Table I provides the summary of the differences for demographic variables between the two subgroups and between each subgroup and the control group. The AI (40.0 • 20.2 apneas per hour total sleep time) of G-A was significantly larger (p < 0.000) than that of
Table VA. Regression analysis for G-A
(Deperident variable: AI) R 0.833
I
R" 0.694
Adjusted I R" 0.652
Probability
0.000
G-B (19.4 • 13.9 apneas per hour). Similarly, the Total Apnea Time (TAT, total apnea time per total sleep time: 28.9% +-- 15.7%) of G-A was significantly larger (p < 0.000) than that of G-B (12.7% • 10.1%). In contrast, both BMI (28.2 • 5.1 k g / m 2) and weight (86.6 • 15.0 kg) in G-A were significantly smaller (BMI: p < 0.000, weight: p < 0.002) than those (BMI: 34.6 _ 9.0 kg/m2, weight: 98.2 • 18.3 kg) in G-B. No significant difference was determined for age between the two subgroups. The BMI, weight, and age in both subgroups were significantly larger than those of the control subjects. Judging from these results, G-A may represent a group of patients with relatively high AI and low BMI and G-B may represent a group of patients with relatively low AI and high BMI respectively. Groups A and B will subsequently be referred to as the high Al/Iow BMI ratio or the low AI/high BMI ratio group. A summary of cephalometric variable differences between the two subgroups and between each subgroup and the control group are provided in Table III. When compared with the low AI/high BMI ratio group (G-B), the high AI/low BMI ratio group (G-A) had .retruded mandibles (SNB: 76.1 ~ - 4.2~ high man-
538
Tsuchiya et al.
Am. J. Orthod.Dentofac.Orthop. June 1992
Table VB. Regression analysis for G-A (Independent variables: cephalometric, upper airway CT and
demographic variables)
Const. BMI ANB SNMP MPH HHl
Coefficient
Standard coefficient
- 62.284 2.449 1.157 0.766 -0.927 i.537
-0.649 0.175 0.231 -0.393 0.624
I
T vahte
-
-6.036 1.705 1.659 1.740 2.441
]
Probability -0.000"* 0.097 0.106 0.090 0.020*
*Significant at the 5 % level. **Significant at the 1% level.
Table VIA. Regression analysis for G-A
(Dependent variable: AI)
Probability
RlR'-lZdjustedR'-[ 0.620
0.385
0.318
0.001
dibular plane angles (SNMP: 37.9 ~ - 5.9~'), and proclined lower incisors (LI to NB: 6.1 __+ 3.0 mm). In comparison, the low AI/high BMI ratio group revealed inferior hyoid bones (MPH: 27.3 "+" 8.6 mm, HHI: 18.1 - 8.2 mm) when compared with the other group. Six statistically significant differences were found between G-A and control subjects. The SNA angle, maxillary unit length, and SNB angle were significantly smaller in G-A. The SNMP angle of G-A was significantly larger than that of the control samples. Both the L1 to NB angle and LI to NB distance were significantly larger in G-A. Ilowever, no significant differences were determined for hyoid bone variables. In contrast, four statistically significant differences were determined between G-B and control subjects. The SNB angle was significantly smaller in G-B. The MPH, ttH1, and C3H were significantly larger in G-B. The upper airway CT variables for the two subgroups and the control group are compared in Table IV. The low Al/high BMI group (G-B) had significantly larger (p < 0.03) soft palate surface areas (SPSA: 3 3 . 7 - 5.9 cm ~) when compared with the high AI/low BMI group (SPSA: 30.6 e,- 7.0 cm2). In comparison with control subjects, both patient groups revealed larger tongues and soft palates, whereas the high Al/low BMI group revealed larger partial airway volumes. A summary of the stepwise multiple regression ,analysis between AI and demographic, cephalometric, and
upper ainvay CT variables for the high AI/low BMI ratio group is provided in Table V. A significant correlation (Adj.R z = 0.652,p < 0.000) was determined between AI and a linear combination of BMI, ANB, SNMP, MPH, and HHI. An adjusted R 2 value of 0.652 indicated that approximately 65% of the variation in AI could be explained by a combination of BMI, ANB, SNMP, MPH, and HH 1, although the partial regression coefficients ofoANB, SNMP, and MPH were not significant at the 5% level. Severe OSA was seen in association with obesity, an anteroposterior discrepancy, a steep mandibular plane, and an inferior and anterior position of the hyoid bone. The largest standardized partial regression coefficient of 0.649 for BMI suggested that BMI was one of the major contributors to AI for the high AI/low BMI ratio group. It was believed that BMI was so strongly associated with AI that some other interdependency between the structural factors and AI might be obscured. Consequently, a regression analysis was repeated with only the cephalometric and CT variables as independent variables. The results of this regression analysis are summarized in Table VI. A significant equation with a linear combination of ANB, SNMP, MPH, and HHI (Adj.R 2 = 0.318, p < 0.001) was identified. An adjusted R ~ of 0.318 indicated that nearly 32% of the variation in AI could be accounted for by a combination of ANB, SNMP, MPH, and HHI. A high AI was related to an anteroposterior discrepancy between the maxilla and mandible, a skeletal open bite tendency and an inferior and anterior position of the hyoid bone. A similar regression analysis for the low AI/high BMI ratio group identified the equation summarized in Table VII. A significant correlation between AI and a combination of PAirVol and BMI (Adj.R z = 0.679, p < 0.000) was determined, although the partial regression coefficient for PAirVol was not significant. An adjusted R 2 of 0.679 indicated that approximately
Volume 101 Number6
Obstrttctive sleep apnea subtypes
539
Table VIB. Regression analysis for G-A (Independent variables: cephalometric and upper airway CT variables)
Const. ANB SNMP MPtI ttttl
Coefficient
Standard coefficient
T value
Probability
- 27.436 2.040 1.748 -2.162 3.414
-0.309 0.527 -0.917 1.386
-2.201 2.893 -3.141 4.453
-0.034* 0.006** 0.003** 0.000'*
*Significant at the 5 % level. **Significant at the 1% level.
Table VlIA. Regression analysis for G-B (Dependent variable: AI) R
IR'lAdjustedR21Probability
0.833
0.695
0.679
0.000
68% of the variation in AI could be explained by a combination of PAirVol and BMI. Subjects with high AI were observed to have small upper airways and obesity. BMI was again the major contributor to AI. In a similar context to the previous group, a regression analysis was repeated with only the cephalometric and CT variables as independent variables. Results of the regression analysis are provided in Table VIII. Tongue volume (TngVol) and PAirVol were identified in the significant regression equation (Adj.R2= 0.246, p < 0.003). An adjusted R2 of 0.246 indicated that nearly 25% of the variation in AI could be accounted for by combination of TngVol and PAirVol. A high AI was seen in association with large tongue and small upper airway volumes. DISCUSSION
Classification of OSA into different subtypes may improve our understanding of the pathogenesis of this disease. Few attempts to classify the OSA patient population from this viewpoint have been reported. Partien et al. 4 subdivided patients with OSA on the basis of BMI, respiratory disturbance index (RDI), airway size, and hyoid bone position. However, the categorization was carried out only on a subjective basis. Lowe et al. 3 divided patients into four skeletal subtypes according to conventional cephalometric criteria and evaluated airway and two- and three-dimensional skeletal structures. To our knowledge, this is the first report to classify patients with OSA on the basis of an interrelationship between obesity and the severity of OSA.
A cluster analysis is a unique statistical method designed to detect associations among a set of objects in a data matrix. 79 The purpose of this analysis is to place objects into clusters not defined deductively, such that objects in a given cluster tend to be similar to each other in:s0me sense, and objects in different clusters tend to be dissimilar. The cluster analysis can also be used to summarize data rather than to find natural or real clusters. In this sense,, it resembles a discriminant analysis. The discriminant analysis pertains to a known number of groups and group structures, and the operational objective is to assign new observhtions to one of these known groups.l~ The cluster method differs in how the distance between two clusters is calculated. In other words, the clustering depends on a linkage method. 79 Various linkage methods, such as single linkage, complete linkage, or average linkage, have been proposed. Since simulation studies, which compared various methods 0f cluster analysis, demonstrated that the average linkage was one of the methods with the best overall performance, s the average linkage was employed in this report. The number of clusters has beer. controversial. Although a rapid decrease in the intracluster similarity indicates a large loss of similarity in an amalgamating procedure and is used to determine the number of discrete clusters, satisfactory significant tests for determining the number of clusters do not exist. 7'8 It should be essential to determine the number of clusters in a realistic way, and also to interpret the structures of clusters from a biologic standpoint. To summarize the characteristics of the two distinct groups segregated by the cluster analysis, the patients with high AI and low BMI ratio had proclined mandibular incisors, retruded mandibles, and a skeletal open bite tendency. The other patients with low AI and high BMI ratio had inferior hyoid bones and large soft palates. The high AI/low BMI ratio patients were char-
540
Tsuchiya et el.
Am. J. Orthod. Dentofac. Orthop. June 1992
Table VIIB. Regression analysis for G-B (Independent variables: cephalometric, upper airway CT and demographic variables)
I Const PAirVol BMI
Standard coefficient
Coefficient
I
T value
- 18.144
--
--
-0.000
-0.137
- 1.520
1.260
0.811
9.022
Probability -0.137 0.000"*
* * S i g n i f i c a n t at the 1% level.
Table VillA. Regression analysis for G-B (Dependent variable: AI) RIR21adjustedR" 0.537
0.288
I 0.246
Probabiliq 0.003
aCterized by skeletal dei'ormities, and the low AI/high BMI ratio patients were characterized by soft tissfle abnormalities. It has already been described that patients with OSA have retrognathic mandibles or.steep mandibular planes. 3'5'"'12 Inferior positioned hyoid bones in patients with OSA have been also reported. 4"12 In addition, a recent study by our group suggests large soft palates in OSA subjects. 3 With regard to an interdependency between severity of OSA and craniofacial morphology and/or demographic variables for the high AI/low BMI ratio group, patients with severe OSA were observed to have skeletal d'iscrepancies such as a large anteroposterior discrepancy and a steep mandibular plane. AI was also highly correlated with BMI. In terms of the position of the hyoid bone, AI has a negative correlation with MPH and a positive correlation with ttHI. These opposite correlations appear inconsistent. However, taking high mandibular plane angles of this group into account, these results could be interpreted as inferior and anterior positional alterations of the hyoid bone. Regression analyses for the low AI/high BMI ratio group showed that severe OSA was associated with a large tongue volume and a small upper airway volume. These analyses confirmed again that the low AI/high BMI ratio group was affected by soft tissue abnormalities. In addition, BMI was the major contributor to AI for this group as well. The hyoid bone serves an important respiratory function. Van Lunteren and his coworkers t3"14demonstrated in a series of investigations that the coordinated-
activation of both of the suprahyoid and infrahyoid muscles displaced the hyoid arch outward, which resulted in a dilated upper airway. Furthermore, head position significantly affected both the resting length of the hyoid muscles and their passive responses to increases in upper airway volume. Winnberg et a1.15demonstrated that a more extended head posture dropped the hyoid apparatus inferiorly and anteriorly. A recent study in which Pae 16 investigated a relationship between airway size and body position suggested that patients with OSA had inferior-positioned hyoid bones and extended head postures. In addition, he reviewed earlier reports and presumed that a smaller than optimal airway induced an extended head posture for better airway patency, and that an extended head posture elicited an inferiorpositioned hyoid bone. In the present study, an inferior- and anteriorpositioned hyoid bone was strongly associated with a severe OSA in the high AI/low BMI ratio group, although no significant mean difference was found for hyoid bone variables between this group and the control group. In comparison, the low AI/high BMI ratio group revealed an inferior-positioned hyoid bone, whereas no significant correlation between AI and hyoid bone position was determined. In patients with high AI and low BMI ratio, the hyoid bone may migrate inferiorly and anteriorly to maintain airway patency as the AI increases. However, mean differences could not be detected statistically. In contrast, patients with low A1 and high BMI ratio tended to have a large tongue and soft palate, and a small upper airway which may be related to obesity. In compensation for these soft tissue abnormalities of the upper airway, the hyoid bone may migrate and reach an anatomic limitation. Consequently, mean differences were determined, but statistically significant correlations between AI and hyoid bone variables could not be found. With regard to the contribution of obesity to the pathogenesis of OSA, a weight-matched study by Hor-
Volume101 Number6
Obstructive sleep apnea subtypes
541
Table VlllB. Regression analysis for G-B (Independent variables: cephalometric and upper airway CT variables)
I Const. TngVol PAirVol
Coefficient
Standard coefficient
4.752 0.000 -0.001
-0.448 -0.369
I
T vahte
Probability
-3.063 -2.519
-0.004** 0.017"
*Significant at the 5% level. **Significant at the 1% level.
ner et al. '7 concluded that significantly more fat was present in the regions surrounding the collapsable segment of the pharynx in the patients with O S A , as compared with control subjects without OSA. Fat deposits in the soft p~late were also observed. Katz et al. 'g described that the external neck circumference was significantly larger in patients with O S A than those without O S A , and it x~'as an important predictor o f OSA. They also speculated that the static pharyngeal size, which is modulated by the dynamic loading of the airway and the weight o f fatty tissue of the neck, may contdbuie to the pathogenesis o f OSA. Our laboratory has identified a significant relationship between tongue and soft palate volume and BMI. 3 This analysis helps us understand the contribution o f obesity in the pathogenesis of O S A . In low A I / h i g h BMI ratio patients who are obese, a large tongue a n d / o r a large soft palate possibly related to obesity may result a small upper airway and consequently contribute to the development o f O S A . On the contrary, in high A I / I o w BMI ratio patients who are moderately obese, skeletal abnormalities may be more important etiologic factors in OSA. These two groups may represent different disease entities. It is suggested that these two distinct subgroups should not be considered as one when pathologic or therapeutic approaches a r e considered. Surgical procedures such as uvulopalatopharygoplasty (UPPP) or mandibular/maxillary osteotomy are frequently recommended to patients with O S A , but it is well known that all patients do not respond to surgery. '92' Both nasal continuous positive airway pressure (CPAP) therapy and some dental appliances are not always effectivc. 19'2''23 According to our results, the high A I / l o w BMI group could respond well to the procedures that advance the mandible forward. In contrast, weight reduction a n d / o r UPPP may better contribute to the improvement o f O S A symptoms for the low A l / h i g h BMI group. In conclusion, these two groups may represent two distinct O S A subgroups. The high AI and low BMI
group appears to have a skeletal mismatch that may contribute to the cause of their O S A , whereas the low AI and high BMI group appears to have atypical soft tissue structures. We thank Mr. S. Kita, Statistical Analysis Adviser of the University of British Columbia Computing Services, for his statistical.advice. We are also indebted to Mrs. M. Wong and Mr. B. Sinclair for their computer expertise. REFERENCES
I. Remmers JE, DeGroot WJ, s EK, Anch AM. Patho_ genesis of upper airway occlusion during sleep. J Appl Physiol 1978;44:931-8. 2.llarper RM. Sleep apnea: OSA mechanisms. CDAJ 1988;35: 35-9. 3. Lowe AA, Fleetham JA, Adachi S, Ryan CF. Cephalometric and CT predictors of apnea index severit)'. Sleep 199 l, [In press]. 4. Partinen M, Guilleminault C, Quera-Salva MA, Jamieson A. Obstructive sleep apnea and celphalometric roentgen~rams: the role of anatomic upper airway abnormalities in the definition of abnormal breathing during sleep. Chest 1988;93:1199-205. 5. Lowe AA, Santamaria JD, Fleetham JA, Price C. Facial morphol~y and obstructive sleep apnea. Ar,1J ORmoo DENTOFAC OR'rdOP 1986;90:484-91. 6. Lowe AA, Gionhaku N, Takeuchi K, Fleetham JA. Threedimensional CT reconstructions of tongue and airway in adult subjects with obstructive sleep apnea. A.~IJ OR'nIODDE:,rrorAc ORmOP 1986;90:364-74. 7. Johnson RA, Wichern DW. Applied multivariate statistical analysis. New Jersey: Prentice-Hall, 1988:543-89. 8. SAS/SYSTAT users guide: release 6.03 edition. Cary: SAS Institute, 1988:47-83. 9. Kawaguchi M. Tahem3'o-KaisekiNyumon 11 (Multivariate statistical analysis It). Tokyo: Morikita Publishers, 1983:26--44. 10. Johnson RA, Wichern DW. Applied multivariate statistical analysis. New Jersey: Prentice-ttall, 1988:470-542. ! 1. Jamieson A, Guilleminault C, Partinen hi, Quera-Salva MA. Obstructive sleep apnea patients have craniomandibular abnormalities. Sleep 1986;9:469-77. 12. LybergT, Kr~stad O, Djupesland G. Cephalometric analysis in patients with obstructive sleep apnea syndrome: skeletal morphology. J Laryngol Oto 1989;103:287-92. 13. Van Lunteren E, Haxhiu MA, Cherniack NS. Mechanical function of hyoid muscles during spontaneous breathing in cats. J'A~ppl Physiol 1987;62:582-90.
542
Tsuchiya et al.
14. Van Lunteren E, ltaxhiu MA, Chemiack NS. Relation between upper airway volume and hyoid muscle length. J Appl Physiol 1987;63:1443-9. 15. Winnberg A, Pancherz H, Westesson P-L. Head posture and hyomandibular function in man: a synchronized electromyographic and videofiuor~raphic study of the open-close-clench cycle. AM J ORTHOODE.',rrOFACORTHOP1988;94:393-404. 16. Pae E-K. A comparative study of the relationship between airway size, tongue activity and body position. [Thesis], University of British Columbia, 1989. 17. Homer RL, Mohiaddin Rtt, Lowell DG, Shea SA, Burman ED, Longmore DB, Guz A. Sites and sizes of fat deposits around the pharynx in obese paiients with obstructive sleep apnoea and weight matched controls. Eur Respir J 1989;2:613-22. 18. Katz I, Stradling J, Slutsky AS, Zamel N, Hoffstein V. Do patients with obstructive sleep apnea have thick necks. Am Rev Respir Dis 1990;14hl228-31. 19. Arand DL. Sleep Apnea: an overview, J Can Dent Assoc 1988;35:12-6. 20. Riley RW, Powell NB, Guilleminault C. Inferior mandibular osteotomy and hyoid myotomy suspension for obstructive sleep
Am. J. Orthod. Dentofac. Orthop. June 1992
apnea: a review of 55 patients. J Oral Maxillofac Surg 1989;47:159-64. 21. Ryan CF, Dickson RI, Lowe AA, Blokmanis A, Fleetham JA. Upper airway measurements predict response to uvulopalatopharyngoplasty in obstructive sleep apnea. Laryngoscope 1990;100:248-53. 22. Cartwright RD. Predicting response to the tongue retaining device for sleep apnea syndrome. Arch Otolaryngol 1985;111: 385-8. 23. Bonham PE, Currier GF, Orr WC, Othman J, Nanda RS. The effect of a modified functional appliance on obstructive sleep apnea. AM J ORTHODDEN'rOFACORTItOP 1988;94:384-92. Reprint requests to: Dr. Alan A. Lowe Department of Clinical Dental Sciences Faculty of Dentistry The University of British Columbia 2199 Wesbrook Mall Vancouver, B.C. Canada, V6T IZ7
BOUND VOLUMES AVAILABLE TO SUBSCRIBERS B o u n d v o l u m e s o f the AMERICAN JOURNAL OF ORTHODONTICS AND DENTOFACIAL ORTIIOPEDICS are available to subscribers (only) for the 1992 issues from the Publisher, at a cost of $53.00 ($67.71 C a n a d a and $64.00 international) for Vol. 101 (January-June) and Vol. 102 (July-December). Shipping charges are included. Each b o u n d v o l u m e contains a subject and author index and all advertising is removed. Copies are shipped within 60 days after publication o f the last issue in the volume. 'The b i n d i n g is durable b u c k r a m with the j o u r n a l name, v o l u m e n u m b e r , and year stamped in gold on the spine. Payment must accompany all orders. Contact M o s b y - Y e a r Book, Inc., Subscription Services, 11830 Westline Industrial Drive, St. Louis, M O 63146-3318, U S A ; telephone (314)453-4351 or (800)325-4177. S u b s c r i p t i o n s m u s t be in f6rcetu qualify. Bound volumes are not available
in place of a regular Journal subscription.
A sample of 84 adult male patients with obstructive sleep apnea (OSA) were classified by a cluster analysis on the basis of apnea index (AI) and body mass index (BMI). Demographic, cephalometric, tongue, soft palate, and upper airway-size data were evaluated for the two subgroups of OSA patients and for 18 control subjects. One OSA group consisted of 43 patients with a high AI and low BMI ratio, the other group was comprised of 41 patients with a low AI and high BMI ratio. The patients with a high AI and low BMI ratio had retruded mandibles with high mandibular plane angles and proclined lower incisors. The patients with a low AI and high BMI ratio had inferior hyoid bones and large soft palates. A multiple regression analysis was performed between AI (the dependent variable) and the other variables (independent variables) for each of the subgroups. In the patients with a high AI and low BMI ratio, a high AI was related to a large skeletal anteroposterior disqrepancy, a steep manidbular plane, and an inferoanterior position of the hyoid bone. In the pat!ents with a low AI and high BMI ratio, a high AI was related to a large tongue and a small upper airway. In both groups, BMI was the major contributor to AI. In conclusion, these two groups may represent distinct subgroups of OSA patients and provide some insight into the contribution of obesity to the pathogenesis of OSA. The patients wit h & high AI and low BMI ratio have a skeletal mismatch, whereas the patients with a low AI and high BMI have atypical soft tissue structures. (AM J ORTHOD DENTOFACORTHOP 1992;101:533-42.)
O b s t r u c t i v e sleep apnea (OSA) occurs because of recurrent occlusion of the upper airway during sleep. The majority of patients with OSA are obese; however, some patients with OSA are not obese, and only a small proportion of oyerweight subjects develop OSA. A previous overemphasis on obesity may have caused some investigations to overlook other potential factors that may predispose to this condition. Recently, several studies '3 have emphasized that upper airway collapse during sleep is a result of interrelated structural and functional factors. Obesity plays a significant and important role in the occurrence of the upper airway occlusion during sleep. 24 Intuitively, OSA patients with and without obesity are likely to have a different pathogenesis. However, there have been no previous reports that classify the OSA patient population on the basis of weight and the influence of the various anatomic factors. The purpose of this study is to use cluster analysis techniques to classify patients with OSA on the basis of the amount of obesity and the severity of their OSA,
Fromthe Universityof BritishColumbia. This projectwas supportedby grants fromthe MedicalResearchCouncilof Canadaand the BritishColumbiaLungAssociation. 'VisitingClinicalAssistantProfessor.Departmentof ClinicalDentalSciences. bProfessorand }lead, Departmentof Clinical DentalSciences. ~GraduateStudent, Departmentof ClinicalDentalSciences. aAssociateProfessor,Departmentof Medicine, UniversityHospital. 8/1/27742
,r_.._
Fig. 1. Reference points, contours, and planes used to identify hyoid bone measurements. H, The most anterior and superior point on the hyoid bone. C3, The inferior anterior position on the third cervical vertebrae. RGn, The most inferior posterior point on the mandibular symphysis. 1, MPH; 2, HH1; 3, HRGn; 4, C3H. For complete definitions, please see METHODS section. and to evaluate craniofacial structure and demographic data by means of multiple regression analyses.
.MATERIAL AND METHODS Patients A group of 84 patients with OSA (obstructive sleep apnea) and 18 control subjects without symptoms were evaluated.
533
Tsuchiya et al.
534
Am. J. Orthod. Dentofac. Orthop. June 1992
I. Differences o f d e m o g r a p h i c variables b e t w e e n the two subgroups and b e t w e e n each subgroup and controls
Table
G-A (n = 43) Variables
Mean
AI (apneaslhr) TAT (%) Weight (kg) BMI (kg/m 2) Age (years)
40.0 28.9 86.6 28.2 47.3
]
G-B (n = 41)
SD
Mean
20.2 15.7 15.0 5.1 13.8
19.4 12.7 98.2 34.6 46.7
]
Control (n = 18)
SD
Mean
I SD
13.9 10. I 18.3 9.0 9.4
--77.0 24.8 31.8
--8.0 1.9 8.2
Probability G-A vs G-B
[ G-A vs Cont
0.000"* 0.000"* 0.002** 0.000"* 0.833
--0.013" 0.011 * 0.00(3**
] G-B vs Cont --0.000"* 0.000'* 0.000'*
*Significant at the 5% level. **Significant at the 1% level.
T a b l e II. A data matrix o f p variables determined for n o b s e r v a t i o n s for cluster analysis
Variables
1
xu
x12
~ 9 . .
xlT,
.....
xlp
2
X21
X22
....
X21~
.....
X2p
r
Xtl
xl2
....
Xrk
.....
Xrp
n
Xal
Xr
....
Xnk
.....
Xi~
.0
To obtain more homogeneous data, female subjects and subjects under the age of 17 were excluded. Edentulous subjects were also excluded from this study as complete cephalometric analyses were not possible. Patients were considered 1o have OSA if they had an apnea index (AI) of > 5 apneas per hour of total sleep time during an overnight polysomnogram. A detailed description of the current overnight polysomnographic monitoring techniques has been completed .5 The control subjects were men with Class 1 malocclusions who revealed n o symptoms suggestive of OSA, such as loud snoring, repeated nocturnal apnea, or excessive daytime sleepiness, when evaluated by the attendhzg physician. Demographic data for the two groups of OSA patients and control subjects are provided in Table I. Cephalometric and upper airway computed tomography (CT) examinations were performed with identical equipment on all subjects.
Cephalometry A single cephalometric radiograph'was taken for each of the subjects before treatment. The subject was seated in an upright position with the head fixed by earrods in the Frankfort horizontal plane parallel to the floor. The dorsum of the tongue was coated with Microtrast Oesophageal Cream (Nicholas Laboratpries Ltd., S!ough, United Kingdom) to enhance-the
outline of the tongue and pharyngeal soft tissues. Each patient was instructed to swallow, lightly contact the back teeth with the lips in repose, and to breathe normally. Tracings were constructed of the lateral head films and traditional contours and points were digitized. In addition, several hyoid bone measurements were determined (Fig. I)? 1. MPH: Linear distance along a perpendicular.from H (the most anterior and superior point on the hyoid bone) to the mandibular plane. 2. tlii1: Linear distance between H and a perpendicular to the C3 (inferior anterior position on the third cervical vertebrae) to retrognathion (RGn: most inferior posterior point on the mandibular symphysis). 3. HRGn: Linear distance between H and RGn. 4. C3H: Linear distance between H and C3.
Upper airway computed tomography Each subject underwent a Siemens DR 2 CT scan analysis. Before the scan, the subject was positioned supine on the examination table in such a position that the Frankfort horizontal plane was perpendicular to the floor. Each subject was instructed to relax (teeth apart) and to continue regular breathing, but not to swallow during any one scan. The patient was evaluated while awake. Scans were obtained at 8 mm
Volume 101
Obstructive sleep apnea subO'pes 535
Number 6
loo:
80A
G-A o
~60. C
~ o o
t~ o~
o o
o
o
o o oo [] o
.r 40-
o
o~
o
9
9 9
,1= u9 e w9 ~ ~0
O
20
o
oo O
o
0 o a...... t,= 9 i
20
.
00 9
9
9 9
G-B
9
00
i
|
4b BMI ( k g / m 2) 30
so
r,o
Fig. 2. Scattergram for two subgroups segregated by cluster analysis. [ ] = G-A; 9 = G-B.
intervals from the Frankfort plane to a level below the sixth cervical vertebrae. Linear attenuation coefficients for each pixel within each scanned section were stored digitally on a magnetic disk and were available for video display as an image analog to the digital distribution. A fixed window level of 30 was selected to emphasize muscle and a window width of 210 was used. Since variations in window level and width are known to change the appearance of any image by altering the edges of a particular structure, these parameters were fixed for all subjects. A magnification factor of 3.5 was used, and a center point was fixed before obtaining the first scan. Tracings were made on acetate paper of each of the slices for airway, tongue, and soft palate structures by one investigator and confirmed by another. The tongue was delineated to include all instrinsic and extrinsic muscles (genioglossus, hyoglossus, and styloglossus). Tracings were also made of the outlines of the nasopharynx, the oropharynx, and the hypopharynx. The nasopharynx was defined to extend from
the roof of the airway to the level of the hard palate. The oropharynx extended down to the uppermost tip of the epiglottis, and the hypopharynx was traced to the slice that most closely corresponded to the most inferior point of the sixth cervical vertebrae. The contour of each structure was digitized, and the data were stored in the computer (1000 E series, Hewlett-Packard Co., Andover, Mass.). Calculations of surface area and volume for all structures were completed. Partial airway volume (PAirVol) represented a combination of nasopharynx, oropharynx, and hypopharynx. A three-dimensional reconstruction of each of the structures was also performed. On the basis of the CT slices, details of the threedimensional analyses for these structures have been previously documented.6 Statistical
analysis
Univariate and multivariate analysis of the data were perf0rmed on the data matrix of 102 subjects with 5 demographic,
536
Ts,tchiya et al.
Am. J. Orthod. Dentofac. Orthop. June
1992
Table III. Differences o f c e p h a l o m e t r i c variables b e t w e e n the two subgroups and b e t w e e n each subgroup and controls
G-A (n = 43)
G-B (n = 41)
Variables
Mean
I SD
Mean
I SD
SNA (*) MxUL (mm) SNB (o) MdUL (ram) ANB (~ ULD (mm) SNMP (~ GoA (~ OB (mm) OJ (mm) UltoNA (*) UltoNA (ram) ~LltoNB (~ !LltoNB (ram) MPH (mm) HH1 (ram) HRGn (mm) C3H (mm)
79.7 96.7 76.1 124.5 3.6 27.8 37.9 130.3 2.6 4.8 20.7 5.2 24.8 6.1 23.3 12.2 39.8 43.9
4.8 5.9 4.2 6.9 2.9 5.8 5.9 6.8 2.9 2.5 7.5 3.7 5.8 3.1 8.2 7.9 6.8 4.6
81.2 97.5 78.0 124.8 3.2 27.3 34.5 127.6 3.4 4.9 19.1 4.4 22.8 4.5 27.3 18.1 43.7 45.7
4.2 4.9 4.4 6.2 3.1 5.9 6.5 5.9 2.7 2.4 8.1 3.6 7.4 2.9 8.6 8.2 7.9 412
Control (n = 18) Mean
I
83.0 99.8 80.9 127.9 2.1 27.8 31.0 127.2 3.1 4.2 23.4 4.5 20.2 3.5 19.1 9.6 42:5 41.9.
Probability
SO
G-A vs G-B
3.8 3.8 4.3 7.0 3.2 6.5 7.1 7.3 3.1 2.2 14.1 3.3 7.0 2.9 7. ! 6.1 9.2 3.6
0.136 0.506 0.043* 0.820 0.544 0.714 0.013" 0.058 0.185 0.947 0.342 0.310 0.185 0.018" 0.034* 0.001"* 0.019" 0.079
I G-a vs Cont 0.012' 0.049* 0.000'* 0.114 0.080 0.966 0.000"* 0.I 1! 0.506 0.304 0.325 0.482 0.010' 0.004** 0.075 0.229 0.214 0. ! 10
I G-B vs Cont 0.121 0.088 0.022* 0.130 0.225 0.754 0.072 0.768 0.747 0.297 0.138 0.917 0.197 0.238 0.001"* 0.000"* 0.629 0.002**
*Significant at the 5% level. **Significant at the !% level.
18 cephalometric, and 6 upper airway CT variables. The data were standardized before a cluster analysis since variables with large variances tended to have more effect on the resultant clusters than those with small variances. To classify patients with OSA according to the degree of obesity and the severity of OSA, a cluster analysis was performed on the basis of standardized AI and body mass indices (BMI). A cluster analysis is a multivariate procedure used to detect natural data groupings on the basis of the similarity of d a t a / ' The investigation can classify a set of objects into subgroups although neither the number of subgroups nor the members of the subgroups are known. Initially each object is considered as a separate cluster, then the two most similar objects are joined to form a cluster. The amalgamating process continues in a stepwise fashion (joining objects or clusters of objects) until a single cluster is formed that contains all the objects. Table II shows a data matrix of p variables, [x~,x.,. . . . . x~. . . . . xp] determined for n observations. The value of a correlation coefficient (R) was specified as the measurement of similarity for clustering observations in this report. The similarity "S,.," between two cases "r,s" is defined as follows: S,=R, R,,, = X,,,/X/2 X, 9 X , where E (x,,, -
X,., = E (x~ - -x,) (xa - -x,), X,, = -x,) 2. The similarity between clusters is calculated
in accordance with a linkage method. There are various linkage methods to combine clusters, i.e. single linkage, complete linkage, average linkage. In this study the average linkage rule was used. The average similarity is the arithmetic average of the similarities Su with all possible pairings of the observations between the two clusters: 1~ ~ SkdKL where observations k are in the first cluster and observations 1 are in the second cluster, and K and L are the number of observations in the two clusters. To investigate the reproducibility of the cluster analysis in this report, a subsample, which consisted of 66 patients randomly selected from the original 84 patients with OSA, was subjected to the same clustering procedure as the original one.
To evaluate differences in craniofacial structure and demographic data, comparisons of means for cephalometric, upper airway CT, and demographic variables were completed by the Student's t test. The subdivided groups and differences between each subgroup and a comparable control group were evaluated. A stepwise method, which is one of the linear multiple regression analyses, was also performed between AI (the dependent variable) and cephalometric, upper ai~vay CT, and demographic variables (independent variables) for each of the subdivided patient groups to evaluate any interactions between the severity of OSA and craniofacial structure and
Volume101
Obstructivesleepapneasubtypes
Number 6
537
Table IV. Differences of upper airway CT variables between the two subgroups and between each subgroup
and controls G-A (n = 43) Variables TngVol (cc) TngSA (cm:) SPVol (cc)
SPSA (cm") TAirVol (co) PAirVol (cc)
Mean 87.8 106.5 9.1 30.6 39.7 21.1
I
SD 12.6 12.6 2.9 7.0 10.4 6.6
G-B (n = 41) Mean 93.7 110.3 10.2 33.7 38.8 19.8
I
SD
Control (n = 18) Mean
17.6 18.3 2.3 5.9 9.7 6.3
79.6 89.0 7.0 25.5 34.6 16.1
I
Probability
SD
G-A vs G-B
15.9 11.6 2.5 6.2 10.8 7.5
I
I
i
i
] G-A vs Cont
.179 .268 .077
.030* .687 .358
[ G-B vs Cont
.035* .000"* .009**
.005** .000"* .000"*
.010" .093 .011"
.000"* .150 .051
Definition of abbreviations: TngVol = tongue volume; TngSA = tongue surface area; SPVol = soft palate volume; SPSA = soft palate surface area; TAirVol = total airway volume; PAirVol = partial airway volume. *Significant at the 5% level. **Significant at the 1% level.
demographic data. The stepwise multiple regression analysis is a statistical method for predicting the value of a response (dependent) variable from a collection of predictor (independent) variable values. It can also be used for assessing the effects of the predictor variables on the response. An adjusted squared multiple correlation coefficient (adj. R~) indicates a proportion of the variation in a dependent variable explained by a combination of independent variables. Statistical evaluations were all completed at the 5% level of significance with SAS/STAT and SYSTAT software programs. RESULTS
An intracluster similarity of 30.04% was determined when the patient population under study was divided into two clusters. It was 48.02% when the population was divided into three clusters and >50% when it was divided into more than three. Consequently, on the basis of an intracluster similarity of 30.04%, the patients with OSA were subdivided into two groups. One group consisted of 43 patients (G-A), and the other was comprised of 41 patients (G-B). The sample distribution is illustrated in Fig. 2. Cluster analysis of the subsamples produced two very similar clusters of patients at the same intracluster similarity as the original clustering; only five patients mismatched with the original clusters. The procedure of cluster analysis employed in this report was confirmed to be reproducible. Table I provides the summary of the differences for demographic variables between the two subgroups and between each subgroup and the control group. The AI (40.0 • 20.2 apneas per hour total sleep time) of G-A was significantly larger (p < 0.000) than that of
Table VA. Regression analysis for G-A
(Deperident variable: AI) R 0.833
I
R" 0.694
Adjusted I R" 0.652
Probability
0.000
G-B (19.4 • 13.9 apneas per hour). Similarly, the Total Apnea Time (TAT, total apnea time per total sleep time: 28.9% +-- 15.7%) of G-A was significantly larger (p < 0.000) than that of G-B (12.7% • 10.1%). In contrast, both BMI (28.2 • 5.1 k g / m 2) and weight (86.6 • 15.0 kg) in G-A were significantly smaller (BMI: p < 0.000, weight: p < 0.002) than those (BMI: 34.6 _ 9.0 kg/m2, weight: 98.2 • 18.3 kg) in G-B. No significant difference was determined for age between the two subgroups. The BMI, weight, and age in both subgroups were significantly larger than those of the control subjects. Judging from these results, G-A may represent a group of patients with relatively high AI and low BMI and G-B may represent a group of patients with relatively low AI and high BMI respectively. Groups A and B will subsequently be referred to as the high Al/Iow BMI ratio or the low AI/high BMI ratio group. A summary of cephalometric variable differences between the two subgroups and between each subgroup and the control group are provided in Table III. When compared with the low AI/high BMI ratio group (G-B), the high AI/low BMI ratio group (G-A) had .retruded mandibles (SNB: 76.1 ~ - 4.2~ high man-
538
Tsuchiya et al.
Am. J. Orthod.Dentofac.Orthop. June 1992
Table VB. Regression analysis for G-A (Independent variables: cephalometric, upper airway CT and
demographic variables)
Const. BMI ANB SNMP MPH HHl
Coefficient
Standard coefficient
- 62.284 2.449 1.157 0.766 -0.927 i.537
-0.649 0.175 0.231 -0.393 0.624
I
T vahte
-
-6.036 1.705 1.659 1.740 2.441
]
Probability -0.000"* 0.097 0.106 0.090 0.020*
*Significant at the 5 % level. **Significant at the 1% level.
Table VIA. Regression analysis for G-A
(Dependent variable: AI)
Probability
RlR'-lZdjustedR'-[ 0.620
0.385
0.318
0.001
dibular plane angles (SNMP: 37.9 ~ - 5.9~'), and proclined lower incisors (LI to NB: 6.1 __+ 3.0 mm). In comparison, the low AI/high BMI ratio group revealed inferior hyoid bones (MPH: 27.3 "+" 8.6 mm, HHI: 18.1 - 8.2 mm) when compared with the other group. Six statistically significant differences were found between G-A and control subjects. The SNA angle, maxillary unit length, and SNB angle were significantly smaller in G-A. The SNMP angle of G-A was significantly larger than that of the control samples. Both the L1 to NB angle and LI to NB distance were significantly larger in G-A. Ilowever, no significant differences were determined for hyoid bone variables. In contrast, four statistically significant differences were determined between G-B and control subjects. The SNB angle was significantly smaller in G-B. The MPH, ttH1, and C3H were significantly larger in G-B. The upper airway CT variables for the two subgroups and the control group are compared in Table IV. The low Al/high BMI group (G-B) had significantly larger (p < 0.03) soft palate surface areas (SPSA: 3 3 . 7 - 5.9 cm ~) when compared with the high AI/low BMI group (SPSA: 30.6 e,- 7.0 cm2). In comparison with control subjects, both patient groups revealed larger tongues and soft palates, whereas the high Al/low BMI group revealed larger partial airway volumes. A summary of the stepwise multiple regression ,analysis between AI and demographic, cephalometric, and
upper ainvay CT variables for the high AI/low BMI ratio group is provided in Table V. A significant correlation (Adj.R z = 0.652,p < 0.000) was determined between AI and a linear combination of BMI, ANB, SNMP, MPH, and HHI. An adjusted R 2 value of 0.652 indicated that approximately 65% of the variation in AI could be explained by a combination of BMI, ANB, SNMP, MPH, and HH 1, although the partial regression coefficients ofoANB, SNMP, and MPH were not significant at the 5% level. Severe OSA was seen in association with obesity, an anteroposterior discrepancy, a steep mandibular plane, and an inferior and anterior position of the hyoid bone. The largest standardized partial regression coefficient of 0.649 for BMI suggested that BMI was one of the major contributors to AI for the high AI/low BMI ratio group. It was believed that BMI was so strongly associated with AI that some other interdependency between the structural factors and AI might be obscured. Consequently, a regression analysis was repeated with only the cephalometric and CT variables as independent variables. The results of this regression analysis are summarized in Table VI. A significant equation with a linear combination of ANB, SNMP, MPH, and HHI (Adj.R 2 = 0.318, p < 0.001) was identified. An adjusted R ~ of 0.318 indicated that nearly 32% of the variation in AI could be accounted for by a combination of ANB, SNMP, MPH, and HHI. A high AI was related to an anteroposterior discrepancy between the maxilla and mandible, a skeletal open bite tendency and an inferior and anterior position of the hyoid bone. A similar regression analysis for the low AI/high BMI ratio group identified the equation summarized in Table VII. A significant correlation between AI and a combination of PAirVol and BMI (Adj.R z = 0.679, p < 0.000) was determined, although the partial regression coefficient for PAirVol was not significant. An adjusted R 2 of 0.679 indicated that approximately
Volume 101 Number6
Obstrttctive sleep apnea subtypes
539
Table VIB. Regression analysis for G-A (Independent variables: cephalometric and upper airway CT variables)
Const. ANB SNMP MPtI ttttl
Coefficient
Standard coefficient
T value
Probability
- 27.436 2.040 1.748 -2.162 3.414
-0.309 0.527 -0.917 1.386
-2.201 2.893 -3.141 4.453
-0.034* 0.006** 0.003** 0.000'*
*Significant at the 5 % level. **Significant at the 1% level.
Table VlIA. Regression analysis for G-B (Dependent variable: AI) R
IR'lAdjustedR21Probability
0.833
0.695
0.679
0.000
68% of the variation in AI could be explained by a combination of PAirVol and BMI. Subjects with high AI were observed to have small upper airways and obesity. BMI was again the major contributor to AI. In a similar context to the previous group, a regression analysis was repeated with only the cephalometric and CT variables as independent variables. Results of the regression analysis are provided in Table VIII. Tongue volume (TngVol) and PAirVol were identified in the significant regression equation (Adj.R2= 0.246, p < 0.003). An adjusted R2 of 0.246 indicated that nearly 25% of the variation in AI could be accounted for by combination of TngVol and PAirVol. A high AI was seen in association with large tongue and small upper airway volumes. DISCUSSION
Classification of OSA into different subtypes may improve our understanding of the pathogenesis of this disease. Few attempts to classify the OSA patient population from this viewpoint have been reported. Partien et al. 4 subdivided patients with OSA on the basis of BMI, respiratory disturbance index (RDI), airway size, and hyoid bone position. However, the categorization was carried out only on a subjective basis. Lowe et al. 3 divided patients into four skeletal subtypes according to conventional cephalometric criteria and evaluated airway and two- and three-dimensional skeletal structures. To our knowledge, this is the first report to classify patients with OSA on the basis of an interrelationship between obesity and the severity of OSA.
A cluster analysis is a unique statistical method designed to detect associations among a set of objects in a data matrix. 79 The purpose of this analysis is to place objects into clusters not defined deductively, such that objects in a given cluster tend to be similar to each other in:s0me sense, and objects in different clusters tend to be dissimilar. The cluster analysis can also be used to summarize data rather than to find natural or real clusters. In this sense,, it resembles a discriminant analysis. The discriminant analysis pertains to a known number of groups and group structures, and the operational objective is to assign new observhtions to one of these known groups.l~ The cluster method differs in how the distance between two clusters is calculated. In other words, the clustering depends on a linkage method. 79 Various linkage methods, such as single linkage, complete linkage, or average linkage, have been proposed. Since simulation studies, which compared various methods 0f cluster analysis, demonstrated that the average linkage was one of the methods with the best overall performance, s the average linkage was employed in this report. The number of clusters has beer. controversial. Although a rapid decrease in the intracluster similarity indicates a large loss of similarity in an amalgamating procedure and is used to determine the number of discrete clusters, satisfactory significant tests for determining the number of clusters do not exist. 7'8 It should be essential to determine the number of clusters in a realistic way, and also to interpret the structures of clusters from a biologic standpoint. To summarize the characteristics of the two distinct groups segregated by the cluster analysis, the patients with high AI and low BMI ratio had proclined mandibular incisors, retruded mandibles, and a skeletal open bite tendency. The other patients with low AI and high BMI ratio had inferior hyoid bones and large soft palates. The high AI/low BMI ratio patients were char-
540
Tsuchiya et el.
Am. J. Orthod. Dentofac. Orthop. June 1992
Table VIIB. Regression analysis for G-B (Independent variables: cephalometric, upper airway CT and demographic variables)
I Const PAirVol BMI
Standard coefficient
Coefficient
I
T value
- 18.144
--
--
-0.000
-0.137
- 1.520
1.260
0.811
9.022
Probability -0.137 0.000"*
* * S i g n i f i c a n t at the 1% level.
Table VillA. Regression analysis for G-B (Dependent variable: AI) RIR21adjustedR" 0.537
0.288
I 0.246
Probabiliq 0.003
aCterized by skeletal dei'ormities, and the low AI/high BMI ratio patients were characterized by soft tissfle abnormalities. It has already been described that patients with OSA have retrognathic mandibles or.steep mandibular planes. 3'5'"'12 Inferior positioned hyoid bones in patients with OSA have been also reported. 4"12 In addition, a recent study by our group suggests large soft palates in OSA subjects. 3 With regard to an interdependency between severity of OSA and craniofacial morphology and/or demographic variables for the high AI/low BMI ratio group, patients with severe OSA were observed to have skeletal d'iscrepancies such as a large anteroposterior discrepancy and a steep mandibular plane. AI was also highly correlated with BMI. In terms of the position of the hyoid bone, AI has a negative correlation with MPH and a positive correlation with ttHI. These opposite correlations appear inconsistent. However, taking high mandibular plane angles of this group into account, these results could be interpreted as inferior and anterior positional alterations of the hyoid bone. Regression analyses for the low AI/high BMI ratio group showed that severe OSA was associated with a large tongue volume and a small upper airway volume. These analyses confirmed again that the low AI/high BMI ratio group was affected by soft tissue abnormalities. In addition, BMI was the major contributor to AI for this group as well. The hyoid bone serves an important respiratory function. Van Lunteren and his coworkers t3"14demonstrated in a series of investigations that the coordinated-
activation of both of the suprahyoid and infrahyoid muscles displaced the hyoid arch outward, which resulted in a dilated upper airway. Furthermore, head position significantly affected both the resting length of the hyoid muscles and their passive responses to increases in upper airway volume. Winnberg et a1.15demonstrated that a more extended head posture dropped the hyoid apparatus inferiorly and anteriorly. A recent study in which Pae 16 investigated a relationship between airway size and body position suggested that patients with OSA had inferior-positioned hyoid bones and extended head postures. In addition, he reviewed earlier reports and presumed that a smaller than optimal airway induced an extended head posture for better airway patency, and that an extended head posture elicited an inferiorpositioned hyoid bone. In the present study, an inferior- and anteriorpositioned hyoid bone was strongly associated with a severe OSA in the high AI/low BMI ratio group, although no significant mean difference was found for hyoid bone variables between this group and the control group. In comparison, the low AI/high BMI ratio group revealed an inferior-positioned hyoid bone, whereas no significant correlation between AI and hyoid bone position was determined. In patients with high AI and low BMI ratio, the hyoid bone may migrate inferiorly and anteriorly to maintain airway patency as the AI increases. However, mean differences could not be detected statistically. In contrast, patients with low A1 and high BMI ratio tended to have a large tongue and soft palate, and a small upper airway which may be related to obesity. In compensation for these soft tissue abnormalities of the upper airway, the hyoid bone may migrate and reach an anatomic limitation. Consequently, mean differences were determined, but statistically significant correlations between AI and hyoid bone variables could not be found. With regard to the contribution of obesity to the pathogenesis of OSA, a weight-matched study by Hor-
Volume101 Number6
Obstructive sleep apnea subtypes
541
Table VlllB. Regression analysis for G-B (Independent variables: cephalometric and upper airway CT variables)
I Const. TngVol PAirVol
Coefficient
Standard coefficient
4.752 0.000 -0.001
-0.448 -0.369
I
T vahte
Probability
-3.063 -2.519
-0.004** 0.017"
*Significant at the 5% level. **Significant at the 1% level.
ner et al. '7 concluded that significantly more fat was present in the regions surrounding the collapsable segment of the pharynx in the patients with O S A , as compared with control subjects without OSA. Fat deposits in the soft p~late were also observed. Katz et al. 'g described that the external neck circumference was significantly larger in patients with O S A than those without O S A , and it x~'as an important predictor o f OSA. They also speculated that the static pharyngeal size, which is modulated by the dynamic loading of the airway and the weight o f fatty tissue of the neck, may contdbuie to the pathogenesis o f OSA. Our laboratory has identified a significant relationship between tongue and soft palate volume and BMI. 3 This analysis helps us understand the contribution o f obesity in the pathogenesis of O S A . In low A I / h i g h BMI ratio patients who are obese, a large tongue a n d / o r a large soft palate possibly related to obesity may result a small upper airway and consequently contribute to the development o f O S A . On the contrary, in high A I / I o w BMI ratio patients who are moderately obese, skeletal abnormalities may be more important etiologic factors in OSA. These two groups may represent different disease entities. It is suggested that these two distinct subgroups should not be considered as one when pathologic or therapeutic approaches a r e considered. Surgical procedures such as uvulopalatopharygoplasty (UPPP) or mandibular/maxillary osteotomy are frequently recommended to patients with O S A , but it is well known that all patients do not respond to surgery. '92' Both nasal continuous positive airway pressure (CPAP) therapy and some dental appliances are not always effectivc. 19'2''23 According to our results, the high A I / l o w BMI group could respond well to the procedures that advance the mandible forward. In contrast, weight reduction a n d / o r UPPP may better contribute to the improvement o f O S A symptoms for the low A l / h i g h BMI group. In conclusion, these two groups may represent two distinct O S A subgroups. The high AI and low BMI
group appears to have a skeletal mismatch that may contribute to the cause of their O S A , whereas the low AI and high BMI group appears to have atypical soft tissue structures. We thank Mr. S. Kita, Statistical Analysis Adviser of the University of British Columbia Computing Services, for his statistical.advice. We are also indebted to Mrs. M. Wong and Mr. B. Sinclair for their computer expertise. REFERENCES
I. Remmers JE, DeGroot WJ, s EK, Anch AM. Patho_ genesis of upper airway occlusion during sleep. J Appl Physiol 1978;44:931-8. 2.llarper RM. Sleep apnea: OSA mechanisms. CDAJ 1988;35: 35-9. 3. Lowe AA, Fleetham JA, Adachi S, Ryan CF. Cephalometric and CT predictors of apnea index severit)'. Sleep 199 l, [In press]. 4. Partinen M, Guilleminault C, Quera-Salva MA, Jamieson A. Obstructive sleep apnea and celphalometric roentgen~rams: the role of anatomic upper airway abnormalities in the definition of abnormal breathing during sleep. Chest 1988;93:1199-205. 5. Lowe AA, Santamaria JD, Fleetham JA, Price C. Facial morphol~y and obstructive sleep apnea. Ar,1J ORmoo DENTOFAC OR'rdOP 1986;90:484-91. 6. Lowe AA, Gionhaku N, Takeuchi K, Fleetham JA. Threedimensional CT reconstructions of tongue and airway in adult subjects with obstructive sleep apnea. A.~IJ OR'nIODDE:,rrorAc ORmOP 1986;90:364-74. 7. Johnson RA, Wichern DW. Applied multivariate statistical analysis. New Jersey: Prentice-Hall, 1988:543-89. 8. SAS/SYSTAT users guide: release 6.03 edition. Cary: SAS Institute, 1988:47-83. 9. Kawaguchi M. Tahem3'o-KaisekiNyumon 11 (Multivariate statistical analysis It). Tokyo: Morikita Publishers, 1983:26--44. 10. Johnson RA, Wichern DW. Applied multivariate statistical analysis. New Jersey: Prentice-ttall, 1988:470-542. ! 1. Jamieson A, Guilleminault C, Partinen hi, Quera-Salva MA. Obstructive sleep apnea patients have craniomandibular abnormalities. Sleep 1986;9:469-77. 12. LybergT, Kr~stad O, Djupesland G. Cephalometric analysis in patients with obstructive sleep apnea syndrome: skeletal morphology. J Laryngol Oto 1989;103:287-92. 13. Van Lunteren E, Haxhiu MA, Cherniack NS. Mechanical function of hyoid muscles during spontaneous breathing in cats. J'A~ppl Physiol 1987;62:582-90.
542
Tsuchiya et al.
14. Van Lunteren E, ltaxhiu MA, Chemiack NS. Relation between upper airway volume and hyoid muscle length. J Appl Physiol 1987;63:1443-9. 15. Winnberg A, Pancherz H, Westesson P-L. Head posture and hyomandibular function in man: a synchronized electromyographic and videofiuor~raphic study of the open-close-clench cycle. AM J ORTHOODE.',rrOFACORTHOP1988;94:393-404. 16. Pae E-K. A comparative study of the relationship between airway size, tongue activity and body position. [Thesis], University of British Columbia, 1989. 17. Homer RL, Mohiaddin Rtt, Lowell DG, Shea SA, Burman ED, Longmore DB, Guz A. Sites and sizes of fat deposits around the pharynx in obese paiients with obstructive sleep apnoea and weight matched controls. Eur Respir J 1989;2:613-22. 18. Katz I, Stradling J, Slutsky AS, Zamel N, Hoffstein V. Do patients with obstructive sleep apnea have thick necks. Am Rev Respir Dis 1990;14hl228-31. 19. Arand DL. Sleep Apnea: an overview, J Can Dent Assoc 1988;35:12-6. 20. Riley RW, Powell NB, Guilleminault C. Inferior mandibular osteotomy and hyoid myotomy suspension for obstructive sleep
Am. J. Orthod. Dentofac. Orthop. June 1992
apnea: a review of 55 patients. J Oral Maxillofac Surg 1989;47:159-64. 21. Ryan CF, Dickson RI, Lowe AA, Blokmanis A, Fleetham JA. Upper airway measurements predict response to uvulopalatopharyngoplasty in obstructive sleep apnea. Laryngoscope 1990;100:248-53. 22. Cartwright RD. Predicting response to the tongue retaining device for sleep apnea syndrome. Arch Otolaryngol 1985;111: 385-8. 23. Bonham PE, Currier GF, Orr WC, Othman J, Nanda RS. The effect of a modified functional appliance on obstructive sleep apnea. AM J ORTHODDEN'rOFACORTItOP 1988;94:384-92. Reprint requests to: Dr. Alan A. Lowe Department of Clinical Dental Sciences Faculty of Dentistry The University of British Columbia 2199 Wesbrook Mall Vancouver, B.C. Canada, V6T IZ7
BOUND VOLUMES AVAILABLE TO SUBSCRIBERS B o u n d v o l u m e s o f the AMERICAN JOURNAL OF ORTHODONTICS AND DENTOFACIAL ORTIIOPEDICS are available to subscribers (only) for the 1992 issues from the Publisher, at a cost of $53.00 ($67.71 C a n a d a and $64.00 international) for Vol. 101 (January-June) and Vol. 102 (July-December). Shipping charges are included. Each b o u n d v o l u m e contains a subject and author index and all advertising is removed. Copies are shipped within 60 days after publication o f the last issue in the volume. 'The b i n d i n g is durable b u c k r a m with the j o u r n a l name, v o l u m e n u m b e r , and year stamped in gold on the spine. Payment must accompany all orders. Contact M o s b y - Y e a r Book, Inc., Subscription Services, 11830 Westline Industrial Drive, St. Louis, M O 63146-3318, U S A ; telephone (314)453-4351 or (800)325-4177. S u b s c r i p t i o n s m u s t be in f6rcetu qualify. Bound volumes are not available
in place of a regular Journal subscription.
