ORIGINAL ARTICLE

Validity of temporomandibular disorder

examination procedures for assessment of
temporomandibular joint status
Marc Schmitter,a Bodo Kress,b Michael Leckel,c Volkmar Henschel,d Brigitte Ohlmann,c
and Peter Rammelsberge
Heidelberg, Germany

Introduction: This hypothesis-generating study was performed to determine which items in the Research
Diagnostic Criteria for Temporomandibular Disorders (RDC/TMD) and additional diagnostic tests have the
best predictive accuracy for joint-related diagnoses. Methods: One hundred forty-nine TMD patients and 43
symptom-free subjects were examined in clinical examinations and with magnetic resonance imaging (MRI).
The importance of each variable of the clinical examination for correct joint-related diagnosis was assessed
by using MRI diagnoses. For this purpose, “random forest” statistical software (based on classification trees)
was used. Results: Maximum unassisted jaw opening, maximum assisted jaw opening, history of locked jaw,
joint sound with and without compression, joint pain, facial pain, pain on palpation of the lateral pterygoid
area, and overjet proved suitable for distinguishing between subtypes of joint-related TMD. Measurement of
excursion, protrusion, and midline deviation were less important. Conclusions: The validity of clinical TMD
examination procedures can be enhanced by using the 16 variables of greatest importance identified in this
study. In addition to other variables, maximum unassisted and assisted opening and a history of locked jaw
were important when assessing the status of the TMJ. (Am J Orthod Dentofacial Orthop 2008;133:796-803)

D
ental or orthodontic treatment, including occlusal reliable if any investigator or clinician can apply the
changes, might have a positive or negative effect measurement technique to asymptomatic or symptom-
on the functional status of the temporomandibu- atic subjects and repeatedly obtain approximately the
lar joint (TMJ).1 Especially when activator treatment is same value.3 The validity of a clinical measure is an
planned, the status of the disc-condyle complex and the indicator of whether a diagnostic test gives the correct
posterior attachment of the TMJ should be assessed diagnosis. The Research Diagnostic Criteria for Tem-
because this treatment might have an impact on the TMJ.2 poromandibular Disorders (RDC/TMD) were intro-
This would enable the dentist or orthodontist to assess the duced in 1992 in an attempt to approach an ideal
effect of this treatment on the joint’s status. Because diagnostic test.3 The RDC/TMD was found to have
imaging techniques are cost-intensive (eg, magnetic reso- acceptable-to-good agreement. Evaluation of the valid-
nance imaging [MRI]) or invasive (eg, radiography), ity of the RDC/TMD examination has not been com-
clinical examinations are preferable. pleted, however. It is not known whether parts of the
Several clinical procedures have been used to eval- examination are more or less important for correct
uate the status of the TMJ. There are, however, large joint-related diagnosis. To identify important aspects of
demands on the validity and reliability of a diagnostic the clinical examination procedure for valid diagnosis
system. This seems to be necessary for correct classi- of TMD, a gold standard must be identified with which
fication of the status of the TMJ. A clinical measure is other clinical findings can be compared. In this context,
From the University of Heidelberg, Heidelberg, Germany. the gold standard for a decision about treatment might
a
Associate professor, Department of Prosthodontics. be the clinical examination or the medical history itself.
b
Associate professor, Department of Neurology, Division of Neuroradiology.
c
In this study, however, the status of the TMJ (healthy
Assistant professor, Department of Prosthodontics.
d
Assistant professor, Institute of Medical Biometrics and Informatics. joint or different joint-related diagnoses) should be
e
Director, Department of Prosthodontics. evaluated without a decision on the need for treatment.
Reprint requests to: Marc Schmitter, Poliklinik für Zahnärztliche Prothetik, Im
Neuenheimer Feld 400, 69120 Heidelberg, Germany; e-mail, Marc_Schmitter@
Because magnetic resonance imaging (MRI) is re-
med.uni-heidelberg.de. garded as the gold standard4 in assessment of TMJ disc
Submitted, March 2006; revised and accepted, June 2006. position,5 and MRI can also be used to evaluate
0889-5406/$34.00
Copyright © 2008 by the American Association of Orthodontists. degenerative changes of the condyle,6,7 it was used in
doi:10.1016/j.ajodo.2006.06.022 this study as the reference standard.
796
American Journal of Orthodontics and Dentofacial Orthopedics Schmitter et al 797
Volume 133, Number 6

Table I. Inclusion and exclusion criteria tion procedure has been described elsewhere.3 The
examiners were calibrated beforehand, and the result of
Inclusion criteria Exclusion criteria
the calibration was assessed during a reliability-assess-
Age of consent Acute trauma ment study:10 87% of the examinations achieved inter-
Symptom free subjects: Acute infections class correlation coefficient (ICC) values greater than
No history of TMD
0.4, indicating satisfactory reliability. The change in
No acute TMD problems
No diseases affecting the joints joint sounds (JS) and TMJ pain on horizontal and
No painful muscle sites vertical mandibular movements under cranial-oriented
TMD patients: Acute dental problems compression of the TMJ (CCT) were also assessed.
At least 1 RDC/TMD axis I diagnosis These additional examinations were performed by the
Factors impeding MRI
same examiners, who were then calibrated for these
acquisition
examinations also. Cranial compression was performed
by applying 3 lbs upward pressure oriented at the
angulus mandibulae. The reliability of this examination
The purpose of this study was to investigate has not, however, been assessed in detail.
whether some components of the RDC/TMD clinical As stated in the RDC/TMD, a psychosocial assess-
examination are less important than others for distin- ment (somatization, depression) was also performed.
guishing joint diseases and healthy joints, and whether The MRI acquisition was performed by using a 1.5-
these components can be weighted, for both patients tesla tomograph (Symphony, Siemens, Erlangen, Ger-
and asymptomatic volunteers. This might help the many), combined with a TMJ surface coil. The average
dentist or orthodontist to assess the status of the TMJ time between the clinical examination and MRI acqui-
with sufficient validity before (and after) treatment; this sition was 7.8 days (minimum, 0 days; maximum, 45
is important for both forensic reasons and therapeutic days; SD, ⫾9.52). The image acquired was an axial
decisions. To achieve this objective, sophisticated sta- pilot with the jaw open or closed. Subsequently, 5
tistical analysis based on classification trees was used. T1-weighted, fast, low-angle shot, 2-dimensional sag-
This statistical technique has been used in several ittal-oblique slices were obtained, 1 at the lateral edge
disciplines in medicine (urology8 and dermatology9), of the condyle, 1 at the medial edge of the condyle, and
but rarely in dentistry. 3 between the medial and lateral edges of the condyle
(time of repetition, 208 ms; time of echo, 10.2 ms; field
MATERIAL AND METHODS of view, 120 mm ⫻ 120 mm; matrix, 256 ⫻ 256; slice
One hundred ninety-two subjects were enrolled in thickness, 3 mm; acquisition time, 5.3 min). These
this study, which was approved by the university sagittal-oblique images were used as a localizer for
review board of the University of Heidelberg. The coronal-oblique imaging, as described elsewhere.11 For
patients were recruited from a pool of 402 patients the opened jaw position, a mechanical jaw opener
seeking treatment for TMD in the Department of (Burnett BiDirectional TMJ-Device, Medrad, Pitts-
Prosthodontics. The symptom-free subjects were re- burgh, Pa) was used to stabilize the jaw. The MRIs
cruited from the staff (including retired persons) and were evaluated by 2 calibrated examiners (1 radiologist
students of the university. Exclusion and inclusion and 1 dentist, unaware of the clinical diagnosis), with
criteria are listed in Table I. regard to disc position and shape of the condyle.
Forty-three were symptom-free control subjects Reliability was assessed after calibration, and agree-
with no signs or history of TMD (35% men, 65% ment had been shown to be acceptable (overall inter-
women; mean age, 34.4 years; SD, ⫾15.5; 26% mar- observer ␬ ⬇ 0.7).12 Both raters evaluated the position
ried; education, 39.6% specialized secondary school or of the disc with the jaw open and closed13,14 and also
higher), and 149 were patients (25% men, 75% women; the direction of the displacement, the shape of the
mean age, 38.5 years; SD, ⫾15.3; 40% married; edu- condyle, and the shape of the disc.15,16 Changes in the
cation, 30.9% specialized secondary school or higher). shape of the condyle were evaluated in accordance with
All subjects were examined by 2 calibrated exam- guidelines17 as modified by deLeeuw et al.18 The
iners (M.S., M.L.) using the RDC/TMD procedure. results of the MRIs were regarded as the reference
This examination includes assessment of the presence standard, and 2 diagnostic categories were defined
or absence of joint sounds and pain, palpation of according to the RDC/TMD classification: group IIb/c,
intraoral and extraoral masticatory muscles by using disc displacement with (ADDR) or without (ADDWR)
defined pressure, and measurement of the range of reduction or no disc displacement; and group III:
mandibular motion. The detailed RDC/TMD examina- arthrosis or no arthrosis.
798 Schmitter et al American Journal of Orthodontics and Dentofacial Orthopedics
June 2008

Statistical analysis ity was increased by unity. At the end of the run, the
Random forest Fortran computer software (www. proximities were normalized by dividing by the number
stat.berkeley.edu/users/breiman/RandomForests/cc_ of trees. The percentage of missing data was 5.3%.
home.htm) by Breiman and Cutler with the g77 com- Most of this arose from CCT examinations, which were
piler was used on a Linux system (SUSE Linux system; sometimes not applied, and from answers missing from
Novell Inc, Waltham, Mass).19,20 This statistical the questionnaires. Proximities were used when replac-
method evaluates the importance of different variables ing missing data.20
In each tree, the values of the out-of-bag data of
in classification and also enables effective estimation of
each variable were randomly changed before running
missing data. Random forest analysis is attractive for
down the tree a second time (evaluation set). If this
assessment of diagnostic accuracy, because it effec-
change alters the resulting diagnosis (eg, arthrosis or no
tively covers all components of the clinical examina-
arthrosis), this variable is important and is awarded a
tion, weighting the contribution of each to the diag-
high score (z-score). The median of these z-scores over
nosis. Because the procedure is unsupervised, no
all the numbers of nodes was then calculated for each
pathologic data or traditional classifications are used a
variable. In the following discussion, this median of
priori. The effect of the different variables on the
z-scores is abbreviated “z-score.” This median is used
distinction between the multiple diagnostic subcatego-
exploratively only. No further inference is regarded
ries can also be assessed by using only 1 statistical
as appropriate. Both weighted (z-score-w) and un-
method rather than many tests (eg, regression analysis).
weighted (z-score-u) z-scores were calculated for ar-
The core of the random forest technique is to create a
throsis, because of the different numbers of diagnoses.
set of decision trees with an optimum number of nodes. The weight was selected so that the out-of-bag errors
The data are put down the trees, and the votes cast for for both diagnoses were equal. The unweighted runs led
the correct class are counted. Next, the values of the to small errors for the healthy joints and large errors for
variables are randomly permuted and are put down the the arthrotic joints. Hence, the z-score-u described the
trees again. The number of votes for the correct class in importance of the variable to diagnosis of healthy joints
the permuted data is then subtracted from the number of and the z-score-w the importance of the variable to
votes for the correct class in the untouched data. The diagnosis of arthrotic joints. Differences between these
average of this number over all trees in the forest is values therefore identify variables which are suitable
the raw importance score for a variable. Thus, if the for distinguishing between healthy joints and arthrotic
number of correct diagnoses does not change substan- joints. If z-score-u was much greater than z-score-w, an
tially when a random value is used for a variable, rather important variable is suitable for identification of
than when the correct value is used, this variable is not healthy joints. If z-score-w was much greater than
important to the diagnosis. In this study, random z-score-u, an important variable is suitable for identi-
classification trees with different numbers of nodes fication of arthrotic joints.
(including 42 variables in each tree) were generated in For disc displacement, z-scores were calculated
each group. Each tree was drawn by sampling with weighed for and ADDWR to balance the out-of-bag
replacement but about one-third of each trees (out-of- error. The z-score-r described the importance of the
bag data) of the cases were left out of and were used to variable for diagnosis of ADDR; the z-score-wr de-
get a running estimate of the classification error as trees scribed the importance of the variable for diagnosis
are added to the forest.19,20 In the examination for ADDWR. High values (ⱖ10) for both z-score-r and
arthrosis, 1000 trees were run each time, with 5 to 20 z-score-wr combined with low differences between
nodes. In the examination for disc displacement, 1000 these 2 values indicated variables which could be used
trees were run each time with 10 to 30 nodes. The to identify healthy joints. Greater differences between
number of nodes was determined by the range where z-score-r and z-score-wr identify variables suitable for
the out-of-bag error was lowest. That is the range of distinguishing between ADDR and ADDWR.
nodes where there is an optimum relationship between The benchmarks for identification of important
the strengths and the correlations of the trees, as variables for differentiation between joints with/with-
described by Breiman.21 Although this range is usually out arthrosis and differentiation of joints with ADDR or
quite wide, the large number of nodes used in this study ADDWR were set to | [z-score-u] –[z-score-w] | ⱖ5 or
was remarkable. After the construction of each tree | [z-score-r] –[z-score-wr] | ⱖ5, respectively. Although
(training set), all data were run down the tree, and the z-scores indicate the importance of the variable in
proximities were computed for each pair of cases. If 2 building a decision tree, they do not show whether, for
cases resulted in the same terminal node, their proxim- example, values are higher for healthy joints than for
American Journal of Orthodontics and Dentofacial Orthopedics Schmitter et al 799
Volume 133, Number 6

pathologic joints in a metric measurement. For vari- classification of more than 95% of cases. Tong et al22
ables with different importance, especially, different used the decision forest technique to classify prostate
values could be seen on a univariate level. These cancer and achieved 99.2% sensitivity and 98.2%
variables were analyzed by using descriptive statistics specificity for the high-confidence prediction group.
(version 8.2, SAS Institute, Cary, NC), to support Our findings support the assumption of important
interpretation of the values of the variables. and less-important examination aspects in the assess-
ment of TMJ status. Pain proved to be an important
RESULTS variable in the assessment of TMD, because pain
No age differences were found between the asymp- location helped to distinguish between joint-related
tomatic subjects and the patients (Mann-Whitney U TMD subgroups. Other variables suitable for differen-
test: P ⫽ 0.07). The other socioeconomic variables tiating between the types of joint disease and healthy
(marriage, sex, and education) also were similar in joints were also identified.
symptomatic and asymptomatic subjects. This finding is important in 2 ways. First, it is
The results for both arthrosis and disc displacement important for construction and modification of classi-
are given in Table II. Overbite measurement, somati- fication procedures (eg, RDC/TMD). Second, it is
zation, depression, opening pattern, and joint sounds important to dental practitioners (dentists and orth-
during opening under CCT had high z-scores (z-score-r odontists) for ruling out joint-related diagnosis or for
and z-score-wr ⱖ10) and low differences ( | [z-score-r] classification of joint-related findings (with or without
–[z-score-wr] | ⬍5) and are, consequently, suitable for obvious clinical limitations) before orthodontic treat-
distinguishing between healthy joints and joints with ment. Although it was reported that pretreatment joint-
disc displacement. Facial pain, tenderness of the lateral related findings did not seem to be a contraindication
pterygoid area, overjet, maximum unassisted opening, for bite-jumping (eg, Herbst treatment23), Popowich
maximum assisted opening (stretch), locked jaw, joint et al1 concluded that methodological deficiencies in the
sounds on opening, joint sounds on protrusion, and analyzed studies prevented major conclusions regard-
joint sounds on lateral excursion to the ipsilateral or ing the effect of this therapy on the disc position.
contralateral side under CCT had | [z-score-r] –[z- Additionally, the reviewed studies highlighted the im-
score-wr] | ⱖ5. These variables are therefore suitable portance of further research. Thus, there might be
for distinguishing joints with ADDR from those with effects of bite-jumping therapy on the TMJ. Conse-
ADDWR. Locked jaw, joint sounds on lateral excur- quently, the assessment of TMJ status seems to be
sion to the ipsilateral side, and joint sounds on pro- useful for both clinicians and scientists.
trusion had | [z-score-u] –[z-score-w] ⱖ5. These vari- As in any statistical assessment, however, classifi-
ables might thus be suitable for distinguishing between cation trees are based on a model and should be
arthrosis and no arthrosis. The z-scores showed 16 discussed. In this context, “important” variables might
variables of interest; these were also analyzed by us- not be suitable for distinguishing between different
ing descriptive statistics. The results are also given in joint-related diagnoses if they are used alone, but only
Table II. The question about locked jaw and joint if they are used within classification schemes.
sounds on protrusion, especially, led to high z-scores in There are 4 parts of a TMD examination: linear
both groups. Most of the other z-scores of interest were measurements (lateral excursion and so on), detection
in the disc-displacement group. of joint sounds, palpation, and questionnaires.
The z-scores of the remaining variables did not There have been few studies on the validity of
indicate importance of these variables for diagnosis of linear measurements.24,25 Because these studies used
joint pathology; eg, palpation of the TMJ seems to have predefined cutoff limits for the diagnostic measure,
less importance. they implied that, for example, linear measurement of
mouth opening is essential for diagnosis of the different
DISCUSSION subtypes of TMD. Thus, the predefined cutoff limits
In this study, we used the random forest technique affected the results of those studies. Our random forest
to evaluate the importance of several variables in approach did not use any predefined cutoff limits, so it
assessment of TMJ status. This technique has proved was possible to assess the importance of each variable
useful in optimizing diagnostic tests in medicine-re- to the diagnostic validity of joint-related diseases,
lated studies: Gerger et al9 used this method to assess avoiding bias. This statistical assessment can also
the sensitivity and specificity of confocal laser-scan- handle the 3 classes in the disc-displacement group
ning microscopy for in-vivo diagnosis of malignant without problems (in contrast to conventional regres-
skin tumors and isolated 3 criteria, enabling correct sion models) and makes it possible to identify variables
800 Schmitter et al American Journal of Orthodontics and Dentofacial Orthopedics
June 2008

Table II. Results from random forest analysis

Examination Arthrosis Disc displacement

z-score-u z-score-w z-score-r (ADDR) z-score-wr (ADDWR) No displacement

†
Overbite (mm) 7.83 4.34 13.95* (4.5) 11.86* (4.7) (4.4)
†
Overjet (mm) 11.57 (4.5) 10.67 (4.5) 12.35 (2.9) 17.87* (3.7) (2.9)
Protrusion (mm) 4.32 4.20 5.15 4.31
Midline deviation (mm) 3.50 3.54 8.15 6.77
Direction of the midline deviation 2.06 0.76 4.95 3.34
Maximum unassisted opening (mm) 10.90 (49.8) 10.96 (45.2) 24.15* (46.7) 34.8† (42.8) *(51.4)
Stretch (mm) 8.15 (2.00) 12.71 (1.80) 15.37* (3.1) 9.32† (2.3) †
(1.5)
†
Somatization 11.30 (16.8) 8.08 (18.4) 11.79* (17.6) 8.05* (18.5) (13.8)
“Have you ever had your jaw 11.88† (60%) 18.03* (89%) 14.71† (76%) 27.43* (93%) †
(54%)
locked?”
†
Depression 11.80 (12.5) 10.69 (15.1) 13.22* (13.6) 11.62* (13.7) (11.5)
“Do you feel pain on this side of 3.95 5.68 3.57† (38%) 11.22* (68%) †
(34%)
your face?”
†
“Could you point to the areas 7.91 9.99 6.98 (29% joint pain) 15.78* (46% joint pain) (19% joint pain)
where you feel pain?”
Opening pattern 2.99 2.05 9.00 (33% straight) 10.97 (33% straight) (47% straight)
Joint sounds on opening (crepitus, 7.45 4.70 15.32* (40% clicking) 5.83† (13% clicking) †
(13% clicking)
clicking)
Joint sounds on closing (crepitus, 10.46 6.50 2.05 1.64
clicking)
Clicking on opening and closing 1.28 1.79 9.20 6.29
Excursion (mm) 3.18 2.13 1.66 5.38
Corrected excursion (mm) 6.79 4.70 2.69 3.68
Joint sounds on excursion to 2.47 1.39 0.40 2.09
contralateral side
Joint sounds on excursion to 11.75† (1% 4.74* (13% 5.04 0.94
ipsilateral side crepitus) crepitus)
Joint sounds on protrusion 10.63† (1% 4.84* (12% 11.49* (38% clicking) 6.09† (12% clicking) †
(9% clicking)
crepitus) crepitus)
Palpation of the temporalis posterior 2.38 0.84 1.01 0.80
Temporalis middle 0.67 3.28 3.53 4.31
Temporalis anterior 0.14 2.51 1.54 0.42
Masseter origin 1.92 1.17 1.30 1.49
Masseter body 2.47 0.63 3.62 2.08
Masseter insertion 5.84 3.42 3.05 0.23
Posterior mandibular region 5.20 1.04 0.18 0.53
Submandibular region 0.65 2.11 0.54 3.00
Lateral pole of the TMJ 0.56 1.95 0.99 1.99
Posterior pole of the TMJ 3.06 0.48 3.52 1.99
Lateral pterygoid area 1.57 0.31 8.02† (36.8% pain) 0.52* (54.8% pain) (43.0% pain)
Tendon of the temporalis 5.48 1.89 7.29 3.85
TMJ pain on excursion to 4.13 3.14 0.92 1.84
contralateral side under CCT
†
Joint sounds on lateral excursion to 2.29 0.80 5.46 (20% louder) 11.03* (34% louder) (9% louder)
contralateral side under CCT
TMJ pain on lateral excursion to 2.90 3.62 4.82 3.45
ipsilateral side under CCT
†
Joint sounds on lateral excursion to 0.64 1.81 5.44 (27% louder) 11.00* (42% louder) (11% louder)
ipsilateral side under CCT
TMJ pain on closing under cranial 4.89 4.07 2.55 5.89
compression of TMJ
Joint sounds on closing under cranial 1.51 1.61 1.44 1.58
compression of TMJ
TMJ pain on opening under cranial 2.54 1.96 1.22 3.78
compression of TMJ
American Journal of Orthodontics and Dentofacial Orthopedics Schmitter et al 801
Volume 133, Number 6

Table II. Continued

Examination Arthrosis Disc displacement

z-score-u z-score-w z-score-r (ADDR) z-score-wr (ADDWR) No displacement

†
Joint sounds on opening under 5.83 5.29 12.86* (25.5% louder) 11.66* (30.3% louder) (10.4% louder)
cranial compression of TMJ
TMJ pain on static compression of 6.25 3.27 0.04 2.27
TMJ

Diagnoses are based on the MRI findings.

Examinations with score differences ⱖ5 are in bold; examinations with z-scores ⱖ10 are in italics.
Numbers in parentheses show the frequency (for categorical variables) and mean values (for continuous variables) for arthrosis or no arthrosis and
ADDR or ADDWR or no displacement, respectively.
*For this important variable, values were higher or frequency of occurrence was greater in the displayed group.
†
For this important variable, values were lower or frequency of occurrence was lower in the displayed group.

that are important for distinguishing between ADDWR method are, however, approximately 0.6229 or between
and ADDR on the one hand and between arthrosis and 0.24 and 0.7.10 Joint sounds during opening were
no arthrosis on the other. identified in this study as an important variable for
Measurement of the maximum unassisted opening, distinguishing between ADDR and ADDWR, although
stretch, and overjet seem to enable distinction between clicking was not present for all joints with ADDR. The
ADDR and ADDWR. The clinical relevance of overjet impact of differences between joint sounds during
measurement is questionable, however: distinguishing opening for joints without displacement (13% clicking)
between 2.9 and 3.7 mm might be difficult clinically. In and joints with ADDR (40% clicking) must be regarded
contrast, differences between ADDWR, ADDR, and no in the context of the construction of the classification
displacement with regard to maximum unassisted open- trees. Isolated consideration of this variable might not
ing are clinically relevant (ADDR, 46.7 mm; ADDWR, be clinically significant. Joint sounds on closing are
42.8 mm; no displacement, 51.4 mm), although all more important in differentiation of the arthrosis group,
values were in the normal range.26 Thus, integration of however. One reason for this could be that clicking of
this variable in a classification tree might be valuable the TMJ on closing in ADDR joints is less pronounced.
for distinguishing between ADDWR (42.8 mm) and no The most important variable for distinguishing nonar-
displacement (51.4 mm). Schmitter et al10 showed that throtic joints (1% crepitus) from arthrotic joints (13%
metric measurements are reliable. This knowledge has crepitus) was found to be the frequency of joint sounds
already been used to choose different cutoff limits in on lateral excursion to the ipsilateral side. This variable
the development of the RDC/TMD. was important only in the arthrosis group. The results
These variables are important in the diagnosis of also showed, however, that crepitus was not found
arthrosis also; this complicates the distinction between frequently enough in arthrotic joints for it be useful in
arthrosis and disc displacement. This finding is not isolation. It is, nevertheless, an important variable
surprising when the results of Emshoff et al27 are together with others in a classification tree.
considered: they found that arthrosis was often com- Joint sounds on protrusion also had remarkably
bined with disc displacement. Their results could also different z-scores in the arthrosis group. This variable
explain the result in this study that many of the also had z-score differences ⱖ5 in the disc displace-
variables with a z-score greater than 10 in the group ment group, however.
ADDR also had z-scores greater than 10 in the arthrosis In this study, we used an additional clinical exam-
group.27 One variable, nevertheless, was identified that ination— dynamic and static compression of the TMJ,
is important for arthrosis alone and consequently en- as usually used in orthopedics. Another study evaluated
ables the examiner to distinguish between arthrotic and these additional examinations and found that both
nonarthrotic joints. In this context, MRI is the gold compression and the joint play test were of minor
standard for the detection of the disc position, although importance for distinction among subgroups of arthrog-
assessment of the shape of the condyle is also possible.6 enous patients.30 In contrast with that result, we found
Interobserver agreement in the detection of joint that joint sounds on lateral excursion to the contralat-
sounds was described as moderate.28 The reliability of eral or ipsilateral side under CCT of the TMJ might
this seems to be lower than for other aspects of the be suitable for distinguishing between ADDR and
clinical examination. Kappa values for the palpation ADDWR and joints with no displacement, because the
802 Schmitter et al American Journal of Orthodontics and Dentofacial Orthopedics
June 2008

joint sounds were more often louder for ADDR and variability. Another limitation of the study is the
ADDWR (27% vs 42% vs 11%). Joint sounds on assumption that both joints of each subject are inde-
opening under CCT of the TMJ also seem to be suitable pendent; this is not necessarily acceptable. Correlations
for distinguishing between healthy joints and joints between both joints of each person could not be
with disc displacements. However, the reliability of modeled with random forests, however. Consequently,
these additional examinations has not yet been evalu- further research is necessary to validate the results of
ated. The results for the validity of these examinations this study and to redefine the cutoff limits of some
must be discussed critically. variables in the diagnosis of TMD subgroups. This
In previous studies, the palpation of masticatory could be done by using another patient pool. The effect
muscles and the provocation of pain on palpation were of revising the RDC/TMD guidelines on overall clas-
in acceptable agreement, after calibration.31 This part sification of patients should then be checked. The
of the examination was not used to distinguish between results of this study could, nevertheless, provide impor-
joint diseases in the RDC/TMD procedure or other tant information about useful variables for analyzing
procedures, however. We found that palpation of the the status of the TMJ. The dentist or orthodontist could
pterygoid area was more often painful in joints with use these variables to enhance the validity of results
ADDWR, so this variable might be useful for differen- from clinical examination of the TMJ.
tiation between ADDR and ADDWR. The z-scores
were less than 10, so interpretation of this result is not CONCLUSIONS
unambiguous. The function of the lateral pterygoid The validity of clinical TMD examination proce-
muscle was described.32 According to this review, it dures might be enhanced by using the 16 variables of
could be concluded that the function of this muscle greatest importance identified in this study. In addition,
might be seriously impaired when the disc is perma- maximum unassisted and assisted opening and a history
nently displaced, and palpation might become painful. of a locked jaw are important when assessing TMJ
Consequently, palpation of the lateral pterygoid muscle status.
might be useful for distinguishing between perma-
nently displaced discs and replacing discs.
Questionnaires including psychosocial questions REFERENCES
are frequently used to assess TMD. Several studies 1. Popowich K, Nebbe B, Major PW. Effect of Herbst treatment on
have shown that psychosocial aspects are different in temporomandibular joint morphology: a systematic literature
TMD subgroups.33,34 It therefore seemed appropriate to review. Am J Orthod Dentofacial Orthop 2003;123:388-94.
2. Ruf S, Wusten B, Pancherz H. Temporomandibular joint effects
include psychosocial aspects (somatization, depression) of activator treatment: a prospective longitudinal magnetic res-
to assess the status of the TMJ. The mean scores for onance imaging and clinical study. Angle Orthod 2002;72:
somatization and depression, and the z-scores, proved 527-40.
to be higher in patients suffering from disc displace- 3. Dworkin S, LeResche L. Research diagnostic criteria for temporo-
ment. Consequently, these variables might help differ- mandibular disorders. J Craniomandib Disord 1992;6:301-55.
4. Liedberg J, Panmekiate A, Petersson A, Rohlin M. Evidence-
entiate healthy joints from joints with disc displace- based evaluation of three imaging methods for the temporoman-
ment. There are, however, only slight differences dibular disc. Dentomaxillofac Radiol 1996;25:234-41.
between joints with and without disc displacement, so 5. Tasaki MM, Westesson PL. Temporomandibular joint: diagnos-
these variables cannot be used exclusively but must be tic accuracy with sagittal and coronal imaging. Radiology 1993;
included in a classification tree. 186:723-9.
6. Watson PJ, Hall LD, Malcolm A, Tyler JA. Degenerative joint
This was a hypothesis-generating study (pilot disease in the guinea pig. Use of magnetic resonance imaging to
study) with a limited number of subjects. The random monitor progression of bone pathology. Arthritis Rheum 1996;
forest technique has, however, been used in studies 39:1327-37.
with a limited sample size9 and not solely in studies 7. Crowley C, Wilkinson T, Piehslingher E, Wilson D, Czerny C.
with large samples.35,36 Ward et al36 investigated the Correlations between anatomic and MRI sections of human
cadaver temporomandibular joint in the coronal and sagittal
short-term prediction of mortality in patients with planes. J Orofac Pain 1996;10:199-216.
systemic lupus erythematosus and concluded that ran- 8. Spurgeon SE, Hsieh YC, Rivadinera A, Beer TM, Mori M,
dom forests were a useful technique for identifying the Garzotto M. Classification and regression tree analysis for the
most important predictors from a larger number and for prediction of aggressive prostate cancer on biopsy. J Urol
validating the classification. This indicates that the 2006;175:918-22.
9. Gerger A, Koller S, Weger W, Richtig E, Kerl H, Samonigg H,
significance of large-sample studies might be better. et al. Sensitivity and specificity of confocal laser-scanning
The out-of-bag error of the different runs, which is an microscopy for in vivo diagnosis of malignant skin tumors.
internal estimate of the test error, also indicated much Cancer 2006;13:13.
American Journal of Orthodontics and Dentofacial Orthopedics Schmitter et al 803
Volume 133, Number 6

10. Schmitter M, Ohlmann B, John M, Hirsch C, Rammelsberg P. 24. Dworkin S, Huggins K, LeResche L, Von Korff M, Howards J,
Research diagnostic criteria for temporomandibular disorders: a Truelove E, et al. Epidemiology of signs and symptoms in
calibration and reliability study. Cranio 2005;23:212-8. temporomandibular disorders: clinical signs in case and controls.
11. Hollender L, Barclay P, Maravilla K, Terry V. A new coronal J Am Dent Assoc 1990;120:273-81.
imaging plane for magnetic resonance imaging of the temporo- 25. Goulet J, Clark G. Clinical TMJ examination methods. J Calif
mandibular joint disc. Dentomaxillofac Radiol 1998;27:48-50. Dent Assoc 1990;18:25-33.
12. Schmitter M, Kress B, Hähnel S, Rammelsberg P. The effect of 26. Gallagher C, Gallagher V, Whelton H, Cronin M. The normal
quality of temporomandibular joint MR images on interrater range of mouth opening in an Irish population. J Oral Rehabil
agreement. Dentomaxillofac Radiol 2004;33:253-8. 2004;31:110-6.
13. Tasaki M, Westesson P, Isberg A, Ren Y, Tallents R. Classifi- 27. Emshoff R, Rudisch A, Innerhofer K, Bosch R, Bertram S.
cation and prevalence of temporomandibular joint disk displace- Temporomandibular joint internal derangement type III: relation-
ment in patients and symptom-free volunteers. Am J Orthod ship to magnetic resonance imaging findings of internal derange-
Dentofacial Orthop 1996;109:249-62. ment and osteoarthrosis. An intraindividual approach. Int J Oral
14. Orsini M, Kuboki T, Terada S, Matsuka Y, Yamashita A, Clark Maxillofac Surg 2001;30:390-6.
G. Diagnostic value of 4 criteria to interpret temporomandibular 28. John M, Zwijnenburg A. Interobserver variability in assessment
joint normal disk position on magnetic resonance images. Oral of signs of TMD. Int J Prosthodont 2001;14:265-70.
29. Dworkin S, LeResche L, DeRouen T, VonKorff M. Assessing
Surg Oral Med Oral Pathol Oral Radiol Endod 1998;86:489-97.
clinical signs of temporomandibular disorders: reliability of
15. Murakami A, Takahashi A, Nishiyama M, Fujishita M, Fuchi-
clinical examiners. J Prosthet Dent 1990;63:574-9.
hata H. Magnetic resonance evaluation of the temporomandibu-
30. Lobbezoo-Scholte AM, Steenks MH, Faber JA, Bosman F.
lar joint disc position and configuration. Dentomaxillofac Radiol
Diagnostic value of orthopedic tests in patients with temporo-
1993;22:205-7.
mandibular disorders. J Dent Res 1993;72:1443-53.
16. Heffez L, Jordan S. A classification of temporomandibular joint
31. Goulet J, Clark G, Flack V. Reproducibility of examiner perfor-
disk morphology. Oral Surg Oral Med Oral Pathol 1989;67:11-9.
mance for muscle and joint palpation in the temporomandibular
17. Laskin D. Diagnosis of pathology of the temporomandibular
system following training and calibration. Community Dent Oral
joint. Clinical and imaging perspectives. Radiol Clin North Am
Epidemiol 1993;21:72-7.
1993;31:135-47. 32. Bertilsson O, Ström D. A Literature survey of a hundred years
18. DeLeeuw R, Boering G, Stegenga B, deBont L. Radiographic of anatomic and functional lateral pterygoid muscle research.
signs of temporomandibular joint osteoarthrosis and internal J Orofacial Pain 1995;9:17-23.
derangement 30 years after nonsurgical treatment. Oral Surg Oral 33. McCreary C, Clark GT, Merril PL, Flack V, Oakley ME.
Med Oral Pathol 1995;79:382-92. Psychological distress and diagnostic subgroups of temporoman-
19. Breiman L. Random forests. Machine Learning 2001;45:5-32. dibular disorder patients. Pain 1991;44:29-34.
20. Breiman L, Friedman J, Olshen R, Stone C. Classification and 34. Lindroth J, Schmidt J, Carlson C. A comparison between
regression trees. Belmont, Calif: Wadsworth; 1983. masticatory muscle pain patients and intracapsular pain patients
21. Breiman L. Technical report #567. Statistical Department. on behavioral and psychosocial domains. J Orofac Pain 2002;16:
Berkeley, Calif: University of California; 1999. 277-83.
22. Tong W, Xie Q, Hong H, Shi L, Fang H, Perkins R, et al. Using 35. Diaz-Uriarte R, Alvarez de Andres S. Gene selection and
decision forest to classify prostate cancer samples on the basis of classification of microarray data using random forest. BMC
SELDI-TOF MS data: assessing chance correlation and predic- Bioinformatics 2006;7:3.
tion confidence. Environ Health Perspect 2004;112:1622-7. 36. Ward MM, Pajevic S, Dreyfuss J, Malley JD. Short-term
23. Ruf S, Pancherz H. Does bite-jumping damage the TMJ? A prediction of mortality in patients with systemic lupus erythem-
prospective longitudinal clinical and MRI study of Herbst pa- atosus: classification of outcomes using random forests. Arthritis
tients. Angle Orthod 2000;70:183-99. Rheum 2006;55:74-80.