ONLINE ONLY

New model for surgical and nonsurgical therapy

in adults with Class III malocclusion
Janka Kochel,a Stefanie Emmerich,a Philipp Meyer-Marcotty,a and Angelika Stellzig-Eisenhauerb
Wuerzburg, Germany

Introduction: Not all adult Class III malocclusion patients are candidates for surgical correction. In patient
assessment and selection, major issues remain regarding diagnosis and treatment planning. The purpose of
this investigation was to ascertain whether adding a transverse parameter to a discriminant analysis could
improve the classification of adults with Class III malocclusion into 2 groups of patients: those who can effectively
be treated by orthodontic therapy and those who require orthognathic surgery. Methods: Cephalograms, plaster
casts, and extraoral photos of 69 adults with Class III malocclusion were analyzed. A discriminant analysis was
performed to identify the variables that best separate the 2 groups. Results: Stepwise variable selection resulted
in a new, highly significant (P \0.0001) model of 4 variables that provided the best discriminant function to
distinguish between patients with and without indications for surgical correction. The resulting equation was
the following: score 5 –10.988 1 0.243 * Wits 1 0.055 * M/M ratio 1 0.068 * NSAr – 0.589 * mand MLD. The
percentage of patients correctly classified by this equation was 91.3%. The sensitivity was 0.92, and the
specificity was 0.89. Conclusions: In the discriminant analysis, the mandibular midline deviation as a transverse
component was included. The addition of the transverse variable led to an improved model concerning the
predictive value in Class III malocclusion patients with surgical requirements. (Am J Orthod Dentofacial
Orthop 2011;139:e165-e174)

C
lass III malocclusion is a severe dentofacial anom-
aly. In most patients, there is no single feature
responsible for the anomaly.1-8 Those with Class
III malocclusion frequently show combinations of
skeletal and dentoalveolar components.9,10 Moreover,
there are complex interactions of genetic and
environmental factors that can act synergistically, in
isolation, or in opposition.11-15
Compared with Class I subjects, several aberrant
cephalometric measurements have been reported in
Class III malocclusion patients, such as shorter anterior
cranial base length, more acute cranial base angle,
shorter and more retrusive maxilla, more obtuse gonial
angle, excessive lower anterior face height, mandibular
prognathism or excessive growth, more proclined
a
Orthodontist, Department of Orthodontics, Dental Clinic, Medical Faculty of the
University of Wuerzburg, Wuerzburg, Germany.
b
Professor and head, Department of Orthodontics, Dental Clinic, Medical Faculty
of the University of Wuerzburg, Wuerzburg, Germany.
The authors report no commercial, proprietary, or financial interest in the
products or companies described in this article.
Reprint requests to: Janka Kochel, Department of Orthodontics, Dental Clinic,
Medical Faculty of the University of Wuerzburg, Pleicherwall 2, 97070 Wuerz-
burg, Germany; e-mail, kochel_j@klinik.uni-wuerzburg.de.
Submitted, April 2010; revised and accepted, September 2010.
0889-5406/$36.00
Copyright Ó 2011 by the American Association of Orthodontists.
doi:10.1016/j.ajodo.2010.09.024
maxillary incisors, and more retroclined mandibular
incisors.4-12 Studies have also shown that no single
morphologic feature indicates potential Class III
development.
Kerr et al16 presented cephalometric criteria for
classification of adult Class III patients to treat them
objectively. The pretreatment lateral cephalograms of
patients who had either surgical or orthodontic treat-
ment of their Class III malocclusion were compared by
using univariate statistical methods. Although signifi-
cant differences were found between both groups in
terms of ANB angle, maxillary-mandibular (M/M) ratio,
mandibular incisor inclination, and Holdaway’s angle,
in view of the complex interaction of skeletal and den-
toalveolar parameters, it seems highly improbable that
single variables could contain enough information to ex-
plain the anomaly.17 Furthermore, univariate statistical
techniques were insufficient for diagnosis, treatment
planning, and outcome prognosis.18 Therefore, recent
studies have recommended a multivariate approach for
analyzing the relationship between craniofacial structure
and Class III malocclusion.19-21
Based on a discriminant analysis (DA), Stellzig-
Eisenhauer et al1 developed a formula to classify Class III
adults into a group that is treatable solely orthodontically
and a group that requires orthognathic surgery. DA is
a multivariate procedure that has been especially
e165
e166
designed to differentiate between 2 groups of subjects
from the same population.22 In the orthodontic literature,
most studies with multivariate statistics explored the
potential of DA.23-25 The determining variables in the
aforementioned study1 were the following: Wits appraisal
(Wits),26 length of the anterior base (S-N), M/M ratio, and
lower gonial angle (Golower). With the multivariate model,
92% of the study patients could be classified correctly.
Consequently, the DA was highly significant (P\0.0001).
In addition to these results, DA had previously been
successfully applied to separate Class III patients from Class
I subjects.19,27,28 Moreover, a DA was used to determine the
prognosis for treatment outcome and relapse of
orthodontically treated Class III patients.21,25,29-34
In the study of Schuster et al,2 multivariate proce-
dures were used to identify the dentoskeletal variables
that provide the best differentiation between prepuber-
tal children with Class III malocclusion who could be
adequately treated by orthopedic or orthodontic therapy
alone and those who required orthognathic surgery. The
models were highly significant, classifying 93.2% to
94.3% of the patients correctly.
In the studies of Stellzig-Eisenhauer et al1 and Schus-
ter et al,2 the Wits appraisal was the most predictive
variable for differentiating between nonsurgery and
surgery patients.
However, the results of the former studies should be
regarded critically. Although multivariate techniques are
better than univariate ones, their limitations include the
following: for a sufficiently stable model that also
applies to patients outside the study, a large sample
size is a prerequisite, and the selection of parameters
might not include all variables required to accurately
differentiate the groups.17,19,23,35
Stellzig-Eisenhauer et al1 could correctly allocate
97.7% of the solely orthodontically treated adults
with Class III malocclusion. Those who required
orthognathic surgery could be classified in 86.4% of
the cases; only 2.3% of the nonsurgery patients were
misclassified, but 13.6% of those who needed orthog-
nathic surgery were misclassified. These findings led to
the hypothesis that, especially in borderline surgical
patients, additional factors are responsible for the
necessity of surgical intervention. Because Class III
patients frequently show skeletal deviations in the
transverse dimension, the predictive value of the mul-
tivariate model might improve if transverse compo-
nents are included.1,2
Therefore, the purpose of this investigation was to
apply DA to a new patient population to determine
whether the addition of a transverse parameter could im-
prove the correct classification of adults with Class III
malocclusion into patients who can effectively be
February 2011 Vol 139 Issue 2 American Journal of Orthodontics and Dentofacial Orthopedics
Kochel et al
treated by orthodontic therapy and those who require
orthognathic surgery.
MATERIAL AND METHODS
To obtain as large a sample size as possible for
a sufficiently stable model, our analysis was based on
data from the Department of Orthodontics of the Univer-
sity of Wuerzburg in Germany and from 7 cooperative
private practices. All patients were white adults (Table I)
and met the inclusion criteria matching those of the orig-
inal study1: (1) Class III molar relationship, (2) negative
overjet, (3) Wits appraisal #–2 mm, and (4) a negative dif-
ference between the ANB angle and the individualized
ANB angle. Patients with craniofacial disorders such as
cleft palate or craniosynostosis were excluded.
The classification of the patients into nonsurgery and
surgery groups was based on pretreatment records
(plaster casts, cephalograms, and extraoral pictures).
The patients were grouped by 3 independent and expe-
rienced orthodontists. The following treatment outcome
criteria for both the surgery and nonsurgery groups had
to be fulfilled: (1) stable occlusion in the sagittal, trans-
verse, and vertical dimensions; (2) correct overjet and
overbite; (3) proper incisal inclination; (4) satisfactory
facial esthetics; and (5) long-term stability. The nonsur-
gery group consisted of 28 patients, and the surgery
group consisted of 41 patients.
Because the lateral cephalograms were taken with
different x-ray devices, all linear measurements were
corrected by their respective magnification factors. One
investigator traced all films with the original 21 landmarks
(Fig) and digitized the data using appropriate software
(fr-win, computerkonkret, Falkenstein, Germany).
As in the previous study,1 the following linear, propor-
tional, and angular measurements were calculated: S-N
(anteroposterior length of the cranial base), PoOr-NBa
(cranial deflection), ML-NSL (divergence of the mandibu-
lar plane relative to the anterior cranial base), NSAr (saddle
angle), ArGoMe (gonial angle), SNB (anteroposterior man-
dibular position in the anterior cranial plane), Wits (length
of the distance AO-BO; AO (intersection between a per-
pendicular line from Point A and the occlusal plane); BO
(intersection between a perpendicular line from Point B
and the occlusal plane); ANB (anteroposterior relationship
of the maxilla and the mandible); ANB-ANBind (difference
between the ANB angle and an individualized ANB angle
according to the formula ANBind 5 –35.16 1 0.4 SNA
1 0.2 ML – NSL26); M/M ratio (ratio of the anteropos-
terior length of the maxilla to the anteroposterior length
of the mandible); ANPog (angle of convexity); 1/1 (angle
between the axis of the maxillary and mandibular inci-
sors); SNA (anteroposterior maxillary position to the
Kochel et al
Table I. Ages of the sample
Nonsurgery group Surgery group
Age (y)
Median 23.6 24.2
Mean 27.6 26.2
Maximum 52.4 56.5
Minimum 18.0 17.9
anterior cranial plane); and NL-NSL (inclination of the
palatal plane in relation to the anterior cranial base).
Thirty-five films were selected randomly, retraced,
and redigitized by the same examiner independently
on 2 separate occasions, 2 weeks apart. The method error
in locating and measuring was calculated as recommen-
ded by using the formula of Dahlberg36:
d2
ME 5 where
2n
d2 is the sum of the squared differences between the
first and second registrations of a pair, and n is the num-
ber of double registrations. Random errors ranged from
0.14 to 1.88 mm for the linear measurements and from
0.31 to 1.98 for all angular variables. Systematic error
was tested at the 10% level of significance, as recom-
mended by Houston,37 and no systematic errors were
found.
In the study of Stellzig-Eisenhauer et al,1 a DA was
used to identify the variables that best separated the
patients who needed orthognathic surgery for correction
from those who did not. In this regard, the discriminant
function was based only on lateral cephalometric land-
marks. Consequently, the skeletal transverse component
of Class III malocclusion was not considered.
The investigation of Stellzig-Eisenhauer et al1 resulted
in the following equation that gives a score for assigning
a new patient to 1 group: score 5 –1.805 1 0.209 Wits
1 0.044 S-N 1 5.689 M/M – 0.056 Golower. The
critical score was –0.023, which was the mean value of
the group centroids of the 2 groups.
Thus, each Class III malocclusion patient with a score
higher than the critical score can be treated successfully
by orthodontic therapy alone. In contrast, each Class III
patient with a more negative score than the critical score
must be treated by combined orthodontic-orthognathic
therapy.
In this study, a transverse parameter was additionally
implemented (mandibular midline deviation, mand
MLD). For this purpose, the degree of mand MLD was
diagnosed by analysis of the pretreatment plaster casts
as described by Schwarz (1936) and Grunberg
(1912).38 This is a routinely used procedure in our
American Journal of Orthodontics and Dentofacial Orthopedics
e167
Fig. Hard tissue landmarks used in the study: sella (S),
porion (Po), basion (Ba), articulare (Ar), gonial intersec-
tion (Go), menton (Me), pogonion (Pog), point B (B),
apex of the mandibular incisor (L1 apex), tip of the man-
dibular central incisor (L1 tip), tip of the maxillary central
incisor (U1 tip), apex of the maxillary central incisor
(U1 apex), point A (A), anterior nasal spine (Ans), poste-
rior nasal spine (Pns), pterygomaxillary fissure (Ptm),
orbitale (Or), nasion (N), ethmoid registration point
(ERP), posterior point of the occlusal plane (POcP), and
anterior point of the occlusal plane (AOcP).
pretherapeutic diagnostic evaluation. Here, the sagittal
jaw relationship is determined involving all tooth move-
ments. The differences between the Angle classes on
both sides after reconstruction (Schwarz) unmask the
mand MLD. To verify the finding of chin deviation, vis-
ible facial asymmetry was used. A mandibular deviation
from the facial midline of at least 2 mm was evaluated as
laterognathism.
Statistical analysis
The data analysis was performed by using SPSS PC1
(version 14.0, SPSS, Chicago, Ill). The arithmetic means,
standard deviations, medians, minimums, and maxi-
mums were calculated for each variable and group
(Table II).
To assess differences between the craniofacial features
of both groups, the data were compared by using the
Mann-Whitney test. This is a nonparametric significance
test that is suitable for small and not normally distributed
groups of patients to assess whether the 2 independent
samples have equally large values. The levels of signifi-
cance were set at P \0.05, P \0.01, and P \0.001
(Table III).
As in the original study, the DA was used to identify
the variables that best separated the patients who needed
February 2011 Vol 139 Issue 2
e168 Kochel et al

Table II. Means, standard deviations, medians, minimums, and maximums of the groups
Nonsurgery group (n 5 28) Surgery group (n 5 41)

Cephalometric variable Mean SD Median Minimum Maximum Mean SD Median Minimum Maximum
SNA ( ) 79.29 0.74 79.35 69.40 85.50 79.32 0.66 79.30 69.70 86.40
SNB ( ) 78.80 0.65 78.15 71.50 85.10 83.00 0.79 82.30 73.90 93.80
ANB ( ) 0.49 0.44 0.65 –7.10 4.20 –3.68 0.53 –3.70 –10.40 4.30
ANB-ANBind ( ) –3.75 0.28 –3.45 –7.00 –1.20 –7.52 0.36 –7.50 –14.30 –2.90
ANPog ( ) –0.51 0.54 –0.15 –9.10 3.80 –5.27 0.62 –4.90 –12.50 4.40
Wits (mm) –4.56 0.30 –4.50 –7.70 –1.10 –9.22 0.49 –9.50 –18.90 –3.00
PoOr-NBa ( ) 32.28 0.64 32.60 23.50 37.60 31.81 0.53 31.60 22.40 41.40
NSAr ( ) 127.06 1,20 127.20 112.70 137.60 121.83 0.78 123.10 111.00 130.70
ArGoMe ( ) 126.69 1,10 128.20 111.30 136.40 128.25 1.35 127.30 111.70 148.90
Goupper ( ) 51.29 0.75 51.35 44.00 61.20 52.00 0.68 52.00 42.10 62.60
Golower ( ) 75.40 1.02 74.60 65.60 85.40 76.25 1.17 76.00 59.90 92.00
NL-NSL ( ) 9.26 0.99 8.90 –4.30 22.40 7.60 0.50 8.20 –0.10 14.30
ML-NSL ( ) 34.51 1.03 34.65 23.40 45.10 32.42 1.50 31.70 8.90 54.60
1-NSL ( ) 102.95 1.12 102.50 92.40 114.30 107.83 1.18 107.90 91.30 123.30
1-ML ( ) 89.90 1.05 90.15 78.80 102.20 84.34 1.19 84.60 69.40 102.70
1/1 ( ) 132.64 1.55 132.50 109.20 149.50 135.42 1.48 134.40 116.10 159.30
S-N (mm) 66.18 1.21 66.50 46.00 76.00 68.70 0.76 68.00 61.00 80.00
M/M ratio (%) 92.71 1.35 93.25 78.90 109.00 81.26 1.20 80.20 68.50 102.70
Mand MLD (mm) 0.45 0.11 0.00 0.00 2.00 1.35 0.16 1.00 0.00 3.00

variables were included in the model according to the

Table III. Significant differences between the nonsur-
5% level of significance. The first variable to be selected
gery and surgery groups was that with the smallest value of the Wilks lambda ( ),
Mann-Whitney test where is the ratio of the within-group sum of squares
divided by the total sum of squares. Subsequent variables
Cephalometric variable Z P
were chosen by recalculating for each variable, and the
SNA ( ) –.079 0.937 NS
SNB ( ) –3.367 0.001 variable with the next lowest value was selected. For each
ANB ( ) –4.822 0.000 stage, a test was performed to ascertain whether the in-
ANB-ANBind ( ) –5.892 0.000 clusion of that variable in the model would significantly
ANPog ( ) –4.761 0.000 improve prediction.
Wits (mm) –5.806 0.000
Unstandardized discriminant function coefficients
PoOr-NBa ( ) –.941 0.347 NS
NSAr ( ) –3.318 0.001 were calculated with a constant for each selected vari-
ArGoMe ( ) –.556 0.578 NS able (Table IV). These calculations led to a new equation
Goupper ( ) –.947 0.344 NS that assigns a score to each patient. For each group, the
Golower ( ) –.348 0.728 NS DA resulted in a mean score for all subjects in the rele-
NL-NSL ( ) –1.473 0.141 NS
vant group. The dividing line halfway between the scores
ML-NSL ( ) –1.124 0.261 NS
1-NSL ( ) –2.799 0.005 was used to delineate the group to which a subject be-
1-ML ( ) –3.342 0.001 longs. This critical score is the mean value of the group
1/1 ( ) –1.082 0.279 NS centroids of the 2 groups. Finally, the classification value
S-N (mm) –1.399 0.162 NS of the selected variables was tested.
M/M ratio (%) –5.084 0.000
Mand MLD (mm) –3.961 0.000

NS, not significant.

P \0.01; P \0.001.
orthognathic surgery for correcting the malocclusion
from those who did not. Therefore, a forward stepwise
variable selection was performed to obtain a model
with the smallest set of significant parameters to avoid
redundancy among the variables. The independent
RESULTS
For the univariate data analysis in Table II, the de-
scriptive statistics are listed for all variables from both
patient groups. The levels of significance are given in
Table III. Significant intergroup differences were found
for parameters representing the sagittal maxillomandib-
ular relationship as indicated by the Wits appraisal, ANB,
and ANB-ANBind. Furthermore, there were significant
differences for the ratio between the lengths of the

February 2011 Vol 139 Issue 2 American Journal of Orthodontics and Dentofacial Orthopedics
 Table IV. Stepwise discriminant analysis
Unstandardized canonical
Predictive variable discriminant function coef cient
Wits 0.243
NSAr 0.069
M/M ratio 0.055
Mand MLD –0.589
(Constant) –10.988
Individual score 5 –10.988 1 0.243Wits 1 0.069NSAr 1
0.055M/M ratio – 0.589mand MLD). Discriminant scores for
maxilla and the mandible, the SNB, the saddle angle, the
axis of the maxillary and mandibular central incisors,
and the difference between the axis of the maxillary
incisor and the palatal plane. The added transversal
parameter mand MLD was also highly significantly
different between the 2 patient groups.
In contrast, the position and inclination of the maxilla,
the cranial deflection, the inclination of the mandible, the
gonial angles (ArGoMe, Goupper, and Golower), the length of
the cranial base, and the interincisal angle showed no
significant differences.
For the multivariate data analysis, stepwise variable
selection resulted in a highly significant model of 4 vari-
ables that best discriminated between patients with and
without an indication for surgical correction.
The variables selected were the Wits appraisal (F like-
lihood to remove 5 .000) and M/M ratio (F likelihood to
remove 5 .010), both sagittal parameters. Furthermore,
the saddle angle (F likelihood to remove 5 .011) and the
mand MLD (F likelihood to remove 5 .001) were also
selected (Table IV).
The unstandardized discriminant function coeffi-
cients of the selected variables, along with a calculated
constant (Table IV), led to the following equation that
provides a score for the assignment of a new patient
to the nonsurgery or surgery group:
M
Score 5 10 988 1 0 243 Wits 1 0 055 ratio
M relevance for treatment decisions. Based on the overlaps
1 0 068 NSAr 0 589 mand MLD
The critical score was 0.251, which is the mean value of
the group centroids of the 2 groups (Table IV). Each new
patient with a Class III malocclusion with a score higher
than the critical score will probably be treated
successfully by orthodontics alone. In contrast, each
new Class III patient with a score lower than the critical
score should be treated by combined orthodontic-
orthognathic therapy.
The percentage of patients correctly classified by the
equation was 91.3% (Table V). Only 3 patients from
each group (nonsurgery group, n 5 28; surgery group,
n 5 41) were misclassified. The sensitivity was 0.92,
and the specificity was 0.89 (Table V).
DISCUSSION
Class III malocclusions are a relatively rare skeletal
anomaly with familial influences. The prevalence varies
among ethnic and racial groups.39-41 The decision as
to which type of treatment is indicated is usually based
on the degree of the anteroposterior and vertical
skeletal discrepancy, the inclination and position of the
incisors, and the dentofacial appearance. Several
lateral cephalometric studies have been performed to
illuminate the growth pattern in Class III subjects
compared with eugnathic subjects and to show the
effectiveness of orthopedic therapy and the stability
of the treatment outcome.1-13,30,31,42 Profitt and
Ackerman43 suggested, in their concept of the “3 enve-
lopes of discrepancies,” the degree of maxillary incisor
protrusion relative to mandibular incisor retrusion as
a critical limitation for differentiating between ortho-
dontic and combined orthodontic-surgical treatment.
Kerr et al16 tried to establish cephalometric yardsticks
to objectify treatment decisions. The most important
factors that differentiated the surgery and orthodontic
patients in their study were size of the anteroposterior
discrepancy, inclination of the mandibular incisors,
and appearance of the soft-tissue profile. The vertical
dimensions—gonial angle and y-axis—were of limited
of box-and-whisker plots, the following critical values
were set: ANB, –4 , M/M-ratio, 0.84; mandibular incisor
inclination, 83 ; and Holdaway angle, 3.5 .
However, univariate statistics are regarded as insuffi-
cient to reflect complex craniofacial relationships.17,25,43
For these reasons, Stellzig-Eisenhauer et al1 and Schus-
ter et al2 used multivariate statistics to categorize
patients into nonsurgery and surgery groups. Also,
Ghiz et al44 presented a logistic equation with 4 variables
to predict the future success of early orthopedic Class III