Professional Documents
Culture Documents
Canada
In a sample of 25 adult male subjects with moderate to severe obstructive sleep apnea, the interaction
among craniofacial, airway, tongue, and hyoid variables was quantified by means of a canonical
correlation analysis. One lateral cephalometric radiograph with the teeth in occlusion was obtained
for each subject together with overnight polysomnographic measurements before the initiation of
therapy. A principal component analysis reduced the data base and one significant canonical
correlation (r, = 0.994) was identified for the 22 variables. Sleep apnea subjects showed a posteriorly
positioned maxilla and mandible, a steep occlusal plane, overerupted maxillary and mandibular
teeth, proclined incisors, a steep mandibular plane, a large gonial angle, high upper and lower facial
heights, and an anterior open bite in association with a long tongue and a posteriorly placed
pharyngeal wall. A multivariate statistical analysis extracted clinically significant associations among
craniofacial, tongue, and airway variables. Subjects with sleep apnea demonstrated several
alterations in craniofacial form that may reduce the upper airway dimensions and subsequently
impair upper airway stability. (AM J ORTHOO DENTOFAC ORTHOP 90: 484-491, 1988.)
Although the pharynx has been visualized with several techniques including cineradiography,g fiberoptic
bronchoscopy,g acoustic reflectance, and forced expiratory maneuvers, the techniques of CT scanning2s3
and lateral cephalometry4.15 are more commonly used.
While most patients with OSA do not have the pathologic lesions described above, recent reports with CT
scanning have suggested that the airway in such subjects
is narrower than in normal contro1s.2,13Although the
cause of these changes remains obscure, the importance
of craniofacial morphology upon the upper airway dimensions remains undefined.
Traditional cephalometry is used extensively in clinical orthodontics to quantify skeletal and soft-tissue dysplasias before the initiation of therapy.16 Numerous
authors7-1ghave quantified specific airway parameters,
although the obvious limitations of any two-dimensional cephalometric study are recognized. The significant advantages of cephalometry are its easy access,
low cost, and minimal radiation.
The purpose of this study was to investigate the
relationship among craniofacial, airway, tongue, and
hyoid variables obtained from lateral cephalometric radiographs in a sample of 25 adult male subjects with
OSA. In addition, the study was designed to reduce the
number of variables by a principal component analysis
and to test, on the pooled data base, for significant
canonical correlations. Clarification of the interaction
between craniofacial morphology, and airway and
tongue parameters in subjects with OSA could permit
the development of revised postulates with regard to
Volume 90
Number 6
Morphology
485
UPW
\/
Mean
Age
Weight
Body mass index
Apnea index
Total apnea time
SaO, mean
TcCO* mean
48
91.2
29.2
36.6
28.8
92.2
51.2
+
+
2
2
2
k
+
1 SD
11 yr
13.0 kg
3.2 kg/m
19.1 apneas/hr
17.7% total sleep time
4.3%
4.5 mmHg
*Nicholas
Laboratories
466
Lowe et al.
Am. J. Orthod.
Dent&c.
Orthop.
December 1986
Table II. Summary of the 22 variables used in the canonical correlation analysis
Variable
Interpretation
Craniofacial
SNA
MxUL
SNB
MdUL
ANB
[SNMP + Go A]
[l to NA + 1 to NA mm]
[ADH + MxMH]
[l toNB + 1 toNBmm]
[l to MP mm + MdMH]
SNOP
OB
OJ
UPH
LPH
[PI-H + RH]
Airway
Tongue
TGL
TGH
Tongue length
Tongue height
Hyoid
[MP-H + H-H]
Cd
C,H = ANTEROPOSTERIOR
POSITION
tance between C, and H.z
OF HYOID.
Linear dis-
The data matrix of 25 subjects with 32 cephalometric variables was computed and univariate and multivariate analyses of the data were undertaken. To perform multivariate analysis, it was essential to reduce
the dimensions of the data base by decreasing the number of variables. A principal component analysis (PCA)
can be used to reduce the number of cephalometric
variables when it is known that a set of variables is not
independent. A smaller set of new variables that represent linear combinations of the original variables can
be determined.23 The first principal component is a linear combination of the original variables, which accounts for the largest possible proportion of the variance. The PCA was applied to logical subgroups until
a group was found for which the first principal component explained more than 75% of the variance. All
original values in the raw data matrix were normalized
to determine the principal components and the canonical
correlations.
To investigate the association between craniofacial
variables, and airway, tongue, and hyoid variables, canonical correlations were determined for the normalized
Volume 90
Number 6
data matrix for the 25 subjects. By means of this analysis, the dependence between two sets of variables can
be summarized in the correlation between the pair of
canonical variables. The significance of the canonical
correlations can be tested by Bartletts chi-square test.23
If a canonical variable is significant, there is a linear
combination of variables in the first group that is significantly correlated with a linear combination of variables in the second group. Canonical variables can be
interpreted by evaluating the loadings on the original
variables.
RESULTS
Morphology
487
Linear
combination
Craniofacial
[SNMP + Go A]
[l toNA
+ 1 toNAmm]
[ADH +
[I toNB
[I to MP
[PFH +
MxMH]
+ 1 toNBmm]
mm + MdMH]
RH]
Airway
[UPW,
+ MPW,
+ LPW,]
UPW. vs. MPW,
UPW, vs. LPW,
MPW, vs. LPW,
[AA-PNS
+ PAS]
Hyoid
[MP-H
+ H-H]
Correlation
coejjicient
Proportion
explained
of variance
by first PC
(%)
0.6963
0.8015
84.07
89.57
0.7163
0.7968
0.7829
0.7666
85.81
89.55
89.15
88.89
79.11
0.7766
0.4629
0.7475
0.5348
76.92
0.4112
89.41
value was larger than 0.3994. Canonical variable loadings less than 0.400 were neglected.
The first (and only significant canonical correlation
at the 0.05 level of significance) showed a linear combination of negatively loaded variables-SNA,
SNB
and OB-with positively loaded variables-[SNMP
+
GoA], [ 1 to NA + 1 to NA mm], [ADH + MxMH],
[l toNB + 1 toNBmm],
[I toMPmm + MdMH],
SNOP, UFH, and LFH-for the craniofacial group, and
a linear combination of negatively loaded variables[UPW, + MPW, + LPWJ-and
a positively loaded
TGL variable in the airway, tongue, and hyoid group.
The eigenvalue for the first canonical correlation (see
Table IV) indicated that approximately 98.8% of the
variation in the first craniofacial variable could be accounted for by an interdependence with the first airway
and tongue soft-tissue variable. A diagrammatic representation of this craniofacial type is provided in
Fig. 2. Subjects with OSA demonstrated a posteriorly
positioned maxilla and mandible, a steep occlusal
plane, overerupted maxillary and mandibular teeth,
proclined maxillary and mandibular incisors, a steep
mandibular plane, a large gonial angle, an anterior open
bite, and high upper and lower facial heights in association with a long tongue and a posteriorly positioned pharyngeal wall. The second through sixth
canonical correlations for this sample of 25 adult subjects with OSA were not significant at the 0.05 level,
488
Lowe
et al.
Dentofac. Orthop.
December 1986
Table IV. Canonical correlations (and their significance) between the craniofacial variables and the airway,
Canonical
correlation
0.98835
0.81242
0.14173
0.63861
0.43202
0.31440
0.99416
0.93404
0.86124
0.79913
0.65729
0.56071
No. of
eigenvalue
test for
remaining
Chi-square
DF
127.74
69.86
43.09
25.49
12.26
4.91
90
70
52
36
22
1
2
3
4
5
eigenvalues
Tail
probability
0.0055
0.4823
0.8059
0.9039
0.9518
0.8973
10
Table V. Canonical variable loadings for the craniofacial variables, and the airway, tongue, and
hyoid variables*
Canonical
Variable
Craniojacial
SNA
MxUL
SNB
MdUL
ANB
[SNIyfP + Go A]
[l toNA + 1 toNamm]
[ADH + MxMH]
[l toNB + 1 toNBmm]
[l toMPmm + MdMH]
SNOP
OB
OJ
uw
LFH
[PFiH + RH]
pizq
-0.125
-0.257
-0.281
0.026
-0.285
variable
loadings
-0.093
0.286
0.058
-0.259
-0.095
0.273
0.237
-0.147
-0.147
-0.077
0.128
0.093
0.003
0.078
0.242
0.468
El-0.617
-0.003
0.042
-0.104
-0.227
0.124
0.069
0.019
-0.321
0.233
z
-0.014
0.190
pYzq
0.119
1-0.5451
0.283
-0.260
0.114
0.160
-0.088
0.187
-0.216
0.387
0.201
-0.114
0.164
z
PI
0.71
-0.511
0.051
-0.087
0.217
-0.023
0.145
0.181
0.101
0.123
0.011
0.211
0.282
-0.041
-0.138
-0.267
-0.219
0.188
0.223
-0.093
-0.029
(0.4851
-0.263
-0.095
pXio.4481
-0.285
-0.373
-0.128
-0.130
0.247
0.286
0.120
0.061
0.001
0.027
f--?i3q
Airway
TGL
TGH
0.081
10.541)
pmij
-0.174
[0.5691
0.191
0.216
-0.001
0.076
-0.072
-0.341
pxq
0.238
0.340
0.237
-0.420
-0.172
-0.214
El
Hyoid
[MP-H + H-H]
CsH
0.013
-0.165
pEi%j
0.007
0.036
0.365
-0.288
~0.710(
0.530
0.786
*Significant loadings larger than 0.400 (absolute values) are enclosed in boxes.
loss of information by means of the PCA. The combination of variables was carefully designed so that the
first principal components could be easily interpreted
from a biologic viewpoint. Usual statistical methods are
often inappropriate in this setting because the measured
variables are numerous and often interrelated so that
Volume 90
Number 6
Morphology
499
SNA
SLEEP
APNEA
r so.994
n= 25
Fig. 2. Diagrammatic representation of the variable loadings for the first canonical correlation. For a
complete description of each variable, see RESULTS section.
individual correlations may be meaningless. Furthermore, it is often more convenient to describe a pattern.
Canonical correlation analyses have been applied
to evaluate the complexity of relationpreviously zz23~25
ships between craniofacial and soft-tissue structures.
However, prior to this report, a multivariate approach
has not been undertaken in a sample of adult subjects
with OSA. Factor analytic methods have been used
extensively to extract the latent components that account for observable craniofacial morphology and to
summarize its overall characteristics. The influence of
many interrelated and confounding variables can be
minimized and a group of important or major variables
defined. Unlike prediction equations, which produce
normal values, the results from multivariate analysesin this case canonical correlations-are used to describe
the data and the interconnections in an objective
manner.
The results from this study demonstrate an interaction between craniofacial morphology and pharyngeal
soft tissues. A previous reportz2 quantified an association between a forward tongue posture and an anterior
open bite, a large overjet, a steep mandibular plane,
and large gonial angle. In the current sample, a posterior
position of the pharyngeal wall and a lengthening of
the tongue occurred with the following craniofacial features: a posterior maxilla and mandible, a clockwiserotated mandible, a steep occlusal plane, an increased
height of the lower and upper face, extruded teeth,
proclined incisors, and an open bite. These findings
490
Lowe et al.
Am. J. Orthod.
Dentofac. Orthop.
December 1986
effect relationships cannot be proved. However, documentation of the interaction among specific variables
does enhance our understanding of the possible interplay between craniofacial variables, and airway,
tongue, and hyoid variables in subjects with OSA.
The interactions described in this report may have
important implications for treatment of OSA. Uvulopalatopharyngoplasty is effective in treating many patients with OSA3 but relatively contraindicated in patients with marked retrognathia. Recently, it has been
suggested that patients in whom the PAS measurement
is abnormal should also have a mandibular advancement
performed to correct the skeletal abnormality.31 In view
of the results described herein, the mandibular changes
may be the result of other bony abnormalities for which
specific therapy may be required. For example, consideration of a maxillary advancement procedure may
be indicated in some subjects in whom a posteriorly
placed maxilla is documented.
On the basis of the findings in this report, it is
hypothesized that patients with obstructive sleep apnea
have several atypical craniofacial structures. These abnormalities, which may be associated with several aberrations in craniofacial form rather than just a deficient
mandible, reduce the upper airway dimensions and impair upper airway stability. A further analysis, based
on CT scans of OSA patients to create three-dimensional reconstructions of craniofacial, airway, and
tongue structures is underway that could provide a better understanding of the interaction among these factors.
The authors thank Drs. J. Petkauand K. Takadafor their
advice and assistancewith the statistical analysis, Mr. R.
DeCou for his assistancein figure preparation, and Mrs. B.
Crockford for typing the manuscript.
REFERENCES
1. Guilleminault C, Dement WC: Sleep apnea syndromes. KROC
Foundation series, vol 11. New York, 1978, Alan R Liss Inc.
2. Remmers JE, DeGroot WJ, Sauerland EK, Anch AM Pathogenesis of upper airway occlusion during sleep. J Appl Physiol
44: 931-938, 1978.
3. Zorick F, Roth T, Kramer M, Fless H: Exacerbation of upperairway sleep apnea by lymphatic lymphoma. Chest 77: 689-690,
1980.
4. Orr WC, Martin RJ: Obstructive sleep apnea associated with
tonsillar hypertrophy in adults. Arch Intern Med 141: 990-992,
1981.
5. Mezon BJ, West P, MaClean JP, Kryger MH: Sleep apnea in
acromegaly. Am J Med 69: 615618, 1980.
6. Perks WH, Horrocks PM, Cooper RA, Bradbury S, Allen A,
Baldock N, Prowse K, Vant Hoff W: Sleep apnea in acromegaly.
Br Med J 280: 894-897, 1980.
I. Perks WH, Cooper RA, Bradbury S, Horrocks P, Baldock N,
Allen A, Vant Hoff W, Weidman G, Prowse K: Sleep apnea in
Scheies syndrome. Thorax 35: 85-91, 1980.
Volume 90
Number 6
Morphology
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
491
Reprint requests:
Dr. Alan A. Lowe
Department of Grthodontics
Faculty of Dentistry
University of British Columbia
2199 Wesbrook Mall
Vancouver, B.C.
V6T 127