Assessment of the validity of HLD (CalMod) in identifying orthodontic treatment need
-
Upload
mary-cooke -
Category
Documents
-
view
214 -
download
0
Transcript of Assessment of the validity of HLD (CalMod) in identifying orthodontic treatment need
Introduction
When a third party is responsible for the costs of
orthodontic treatment, funds are limited, which
correspondingly limits treatment eligibility. An
index is typically used to identify those patients
who are in greatest need of orthodontic treatment.
Occlusal indexes can identify patients in need of
orthodontic treatment and prioritize their treat-
ment requirements, both for quality assurance and
for use in research (1, 2). Occlusal indexes are
common in Northern Europe and are accepted in
clinical orthodontics (3). In the United States,
however, the use of occlusal indexes in every day
practice is extremely limited(4) due primarily to the
fact that the American Association of Orthodontics
(AAO) does not recognize any index rating classi-
fication or coding system as a scientifically valid
measure of the need for orthodontic treatment (5).
Nevertheless, public health planners in 15 states
have adopted several occlusal indexes with arbi-
trary cut-off scores to determine eligibility for
orthodontic care with state funds (6). The Handi-
capping Labiolingual Deviation (HLD) index is one
of the indexes used in the United States, developed
originally to identify those with handicapping
malocclusions (7). Some states have modified the
HLD index in order to determine and prioritize
eligibility for state-funded orthodontic treatment.
For example, Maryland’s index, HLD (Md), has
Community Dent Oral Epidemiol 2010; 38: 50–57All rights reserved
� 2009 John Wiley & Sons A/S
Assessment of the validity ofHLD (CalMod) in identifyingorthodontic treatment needCooke M, Gerbert B, Gansky S, Miller A, Nelson G, Orellana M. Assessment ofthe validity of HLD (CalMod) in identifying orthodontic treatment need.Community Dent Oral Epidemiol 2010; 38: 50–57. � 2009 John Wiley & Sons A ⁄ S
Abstract – Objective: The purpose of this study was to assess the validity of theHandicapped Labio-Lingual Deviation index with California modifications,HLD (CalMod), in identifying handicapping malocclusions. Methods: A set of153 study casts representing all types of malocclusion was utilized in this study.Models were randomly chosen the UCSF Division of Orthodontics clinic.Treatment need was determined by the HLD (CalMod) index and by a panel of13 orthodontists, conventionally established as the ‘gold standard’. SpearmanRank correlation analysis was used to evaluate the correlation between HLD(CalMod) and the gold standard. The Classification and Regression Tree(CART) modeling was used to determine the HLD (CalMod) cut-off point oforthodontic treatment need according to the gold standard. Results: ASpearman Rank correlation Coefficient of 0.71 demonstrated a moderatelyhigh correlation between HLD (CalMod) and the gold standard. The CARTmodeling determined a value of 18.5 as the cut-off point of HLD (CalMod) fororthodontic treatment need, considerably lower than the cut-off point of 26currently used by Medi-Cal. At a value of 26 points as the cut-off HLD(CalMod) displayed a low sensitivity (25.9%) and high specificity (96.8%).Witha cut-off point of 18.5, specificity decreased to 55.6% while sensitivity increaseddramatically to 92.9%. Conclusion: Our results show that the HLD (CalMod)with a cut-off point of 26 fails to indentify a considerable percentage ofhandicapping malocclusions. More studies should be done assessing theefficacy of the HLD (CalMod) in identifying handicapping malocclusion.
Mary Cooke1, Barbara Gerbert2, Stuart
Gansky2, Arthur Miller3, Gerald Nelson3
and Maria Orellana3
1Private Practice, Napa, CA, 2Department of
PRDS, School of Dentistry, University of
California San Francisco, San Francisco, CA,3Department of Orofacial Sciences, School of
Dentistry, University of California San
Francisco, San Francisco, CA
Key words: malocclusion; orthodontics;public health policy
Maria Orellana, Orofacial Sciences,University of California, San Francisco, CA,USA.Tel.: +1 415 476 4730Fax: +1 415 502 1013e-mail: [email protected]
Submitted 8 January 2009;accepted 26 August 2009
50 doi: 10.1111/j.1600-0528.2009.00506.x
raised the cut-off from 13 to 15 points and modified
the HLD’s original scoring formula for overjet and
overbite (8). The State of Washington modification
includes five qualifying conditions and a cut-off
point of 30 (9).
California uses the HLD index with modifica-
tion, hence the designation HLD (CalMod).
Twelve factors are weighted and summed to
provide a score: overjet, overbite, open bite, cleft
lip-palate, anterior crowding, mandibular protru-
sion, labiolingual spread, deep impinging overbite,
severe traumatic deviations, crossbite of individ-
ual anterior teeth, ectopic eruption of anterior
teeth, and posterior unilateral crossbite. California
modified the index as a result of settlements
stemming from two lawsuits. As a result of the
first lawsuit, two qualifying exceptions, recog-
nized as causing injury to the buccal tissue, were
added: deep impinging bites and crossbites of
individual teeth, with tissue damage. In addition,
unilateral posterior crossbite was inserted as a
weighted factor. As an outcome of the second
lawsuit, a reversed overjet greater than 3.5 mm
was added as a qualifying exception. At that time,
an overjet greater than 9 mm was also inserted as
a qualifying exception. These modifications
encompass the current HLD (CalMod) index with
a cutoff point of 26.
Many interest groups involved in government
policy contribute to a public treatment benefit,
especially when funds are limited (10). Three issues
are paramount, the index validity, the given cut-off
point and the qualifying exceptions. A valid index
must accurately measure what it intended to
measure; in the case of the HLD (CalMod) index,
to measure the level of a handicapping malocclu-
sion. Within a specific population, it is essential to
establish a relationship between the index values
and the providers’ opinions of treatment need. One
should specifically ask: Does this index accurately
reflect providers’ opinions?
An occlusal index is validated on evidence that
an occlusal anomaly will have a short or long term
effect on the dental ⁄ oral structures and by
comparing its scores with a conventional ‘gold
standard’, commonly the subjective consensus
opinion of a group of experienced specialists (11).
Yet, the locale in which an orthodontic specialist
practices has an effect on his or her evaluation of
the treatment need (12). In a large country such as
the United States, one can expect to find regional
differences that may influence decisions regarding
the need for orthodontic treatment.
There have been few studies utilizing large
panels of experts to assess the validity and reli-
ability of specific occlusal indexes (9, 13) and only
one involving the HLD (CalMod) (14). There has
been no study assessing the validity of the HLD
(CalMod) with a panel of orthodontic specialists
practicing in California.
The objective of this study was to determine the
validity of the HLD (CalMod) in identifying hand-
icapping malocclusions. Specifically, we wanted to
assess if the HLD (CalMod) reflects the providers’
opinion, the ‘gold standard’, of what constitutes
handicapping malocclusion.
Material and methods
One hundred and fifty-three sets of models were
randomly chosen from the UCSF Division of
Orthodontics clinic. The models represented a
spectrum of malocclusions and included pretreat-
ment models and posttreatment models. Models
were not included if any appliances were visible.
Models were trimmed to Centric Occlusion (CO)
and were evaluated for bubbles or imperfections
that could influence or interfere with evaluation of
the malocclusion. Each model was evaluated and
any information that could not be determined from
the model alone was listed on the ‘information
sheet’ that was kept with each model. This sheet
included information such as impacted teeth,
missing teeth, and CR-CO shifts. The models were
then assigned a random number from 1001 to 1153.
This study was approved by the University of
California San Francisco Committee on Human
Research (CHR#H2582-24865-01).
Gold standardOriginally, 15 orthodontists from the San Francisco
Bay area were recruited to participate in this study.
One subject never began the study and a second
completed the evaluation on only 23 models.
Hence, that participant’s observations were not
included in the study results (Table 1). Inclusion
criteria for participants were five or more years of
clinical orthodontic experience and being active in
clinical practice on a half-time basis or greater. All
participants had to be members of the AAO. Any
orthodontist that had had previous experience with
orthodontic indexes, with the exception of the
California Medi-Cal ⁄ California Children Services
(CCS) index, was excluded from the study. The
orthodontists in this study were selected on the
51
Assessment of the validity of HLD
basis of their availability and commitment to
this study and do not reflect the distribution of
orthodontists in the state of California.
The orthodontic professionals were each pro-
vided one on-site session for evaluating the
models. During the session, casts were displayed
in numerical order on countertops in a large room.
At the beginning of the session, the following
verbal and written instructions were given to each
orthodontist: ‘You are the orthodontic consultant for a
private corporation for which a limited fund has been
established to provide orthodontic treatment for person-
nel. You are to evaluate these study casts of personnel
and answer the following question: In your opinion, to
what extent does this occlusion need orthodontic treat-
ment?’
The orthodontists scored the 153 pairs of cast and
recorded the need for treatment of each pair as a
score of 1–12 on an adjectival scale where:
• 1–3 = No treatment needed
• 4–6 = Treatment optional or elective
• 7–9 = Treatment advisable
• 10–12 = Treatment essential
All models with a score of ‘7’ or above were
considered to be in the treatment category. In
models with a gold standard score of greater than
9.5, orthodontic treatment was deemed medically
necessary. This would correspond to a HLD score
of 26 or above indicating a handicapping maloc-
clusion with medical justification for orthodontic
treatment. For the orthodontic specialists’ evalua-
tions, interrater reliability was evaluated with
analysis of variance and Scheffe’s multiple
comparison tests. Linear regression was used to
compare the gold standard and HLD(CalMod);
since this showed a clear quadratic relationship, a
square root transformation of HLD(CalMod) was
used to meet linear regression assumptions.
HLD (CalMod) indexEach model was evaluated and scored with the
instructions given by Medi-Cal for evaluation with
the HLD (CalMod) index by a qualified orthodontist
(7, 15). It was first determined if it possessed one or
more automatically-qualifying-exception traits
such as cleft lip and palate deformity; a severe
traumatic deviation; deep impinging bites affecting
the soft tissue of the palate; crossbite on individual
anterior teeth with destruction of soft tissue; overjet
greater than 9 mm; and reverse overjet greater than
3.5 mm (16). In addition, each model was evalu-
ated by the scoring protocol, whether or not an
automatically qualifying exception existed. Some
assumptions had to be made given that photos and
radiographs were not available. For example, if an
overjet of greater than 9 mm was determined, lip
incompetence was assumed or if a negative overjet
of greater than 3.5 mm was determined, difficulty
masticating was assumed.
Intra-rater reliability for HLD (CalMod) was
evaluated using Lin’s concordance correlation,
which assesses equivalence of 2 measurements.
The reproducibility of measurements was assessed
by statistically analyzing the difference between
double measurements made 4 months apart on a
subset of 40 casts randomly chosen.
Validity of HLD (CalMod) was evaluated by two
methods: (i) Spearman Rank correlation was used
to evaluate the correlation between HLD (CalMod)
and the gold standard. (ii) the Classification and
Regression Tree (CART) modeling (Salford
Systems Inc, San Diego, CA, USA) was used to
determine the cut-off of orthodontic treatment need
(17, 18). Specifically, CART interactively evaluated
the predictor to determine the optimal cut-off: as
this is highly dependent, 10-fold cross-validation
was used to make the results robust and avoid
overfitting (19).
Results
Gold standard scoresEach model was given 13 scores (one from each
orthodontic rater). The mean of the 13 scores
from each model was compared to the mode and
the median. The average mean score for all casts
was 7.2, the average mode was 6.9 and the
average median was 7.1. The mean was chosen as
the gold standard score for each model. Table 2
shows the number of models that fell into the
four different orthodontic needs categories. The
Scores breakdown differed from the orthodontist
evaluation sheet because the gold standard score
was an average of 13 scores rounded to the
closest tenth. The cut-off between categories was
Table 1. Orthodontist subject characteristics
Mean age range 35–70 50Average years certified range 6–41 20.1Number of different programs 6Private practice 8Group practice 2University practice 3Male 10Female 3
52
Cooke et al.
determined by choosing a number that split the
difference of the highest number of the first cat-
egory and the lowest number of the next category.
Interrater reliability analysis showed rater 4 to be
significantly different from every other rater apart
from rater 5 while rater 5 was significantly
different from five of the other 12 raters. Rater
9 was significantly different from 3 other raters.
Apart from raters 4, 5 and 9, the raters had a
maximum of 2 significantly different findings
(Table 3). There was no relationship between
rater’s orthodontic experience and mean model
score and between rater’s age and mean model
score.
HLD (CalMod) scoresLin’s concordance correlation showed high levels
of intraexaminer reliability: 0.894. A Spearman
Rank correlation Coefficient of 0.71 demonstrated
a moderately high correlation between HLD (Cal-
Mod) and the gold standard (P < 0.01). Fig. 1
shows the quadratic relationship between the gold
standard and HLD (CalMod). The relationship
between the gold standard and the square root of
HLD (CalMod) is essentially linear.
Sensitivity and specificity of HLD (CalMod)
were determined using the HLD (CalMod) cut-off
of 26 points. The results displayed a low sensitivity
(25.9%) and high specificity (96.8%) for this mea-
sure as it is used currently (Table 4). When we
applied CART modeling to determine the cut-off of
HLD (CalMod) for orthodontic treatment need, we
arrived at a value of 18.5 points, considerably lower
than the cut-off of 26 currently used by Medi-Cal.
Table 2. Distribution of cases according to the goldstandard
Category
Goldstandardrange
Numberof cases
No treatment needed 1)3.5 21Treatment optional >3.5)6.5 28Treatment advisable >6.5)9.5 77Treatment essential >9.5 27
Table 3. Descriptive statistics of orthodontist rater scores
Mean SD SE Count Minimum Maximum # Missing
Rater 1 6.869 2.662 0.215 153 1 11 0Rater 3 6.647 2.575 0.210 150 1 12 3Rater 4 9.955 2.558 0.214 143 2 12 10Rater 5 8.348 2.734 0.255 115 1 12 38Rater 6 7.678 2.696 0.223 146 1 12 7Rater 7 6.553 2.681 0.217 152 1 11 1Rater 8 6.203 2.165 0.175 153 1 11 0Rater 9 8.054 2.962 0.243 148 1 12 5Rater 10 6.926 3.203 0.262 149 1 12 4Rater 12 6.270 3.333 0.274 148 1 12 5Rater 13 7.020 3.284 0.265 153 1 12 0Rater 14 6.894 3.529 0.287 151 1 12 2Rater 15 6.033 2.753 0.224 151 1 10 2
Table 4. Sensitivity and specificity of HLD (CalMod)with a cut-off of 26
HLD (CalMod)with cut off of 26
Gold Standard
Handicappingmalocclusion
Nohandicappingmalocclusion
Handicappingmalocclusion
7 4
NO handicappingmalocclusion
20 122
Sensitivity (%) 25.9Specificity (%) 96.8Prevalence (%) 17.7(+) Predictivevalue (%)
63.6
()) Predictivevalue (%)
85.9
0
1
2
3
4
5
6
7
0 2 4 6 8 10 12Gold standard
Sqr
root
of
HL
D(C
alM
od)
Fig. 1. Gold standard scores versus the square root of theHLD (CalMod) index scores.
53
Assessment of the validity of HLD
The sensitivity and specificity were then recalcu-
lated using the new cut-off of 18.5 for HLD
(CalMod). Specificity decreased to 55.6% while
sensitivity increased dramatically to 92.9%. Posi-
tive predictive value was still low at 62.5%
(Table 5).
Discussion
We found the HLD (CalMod) index to be fairly
easy to utilize in terms of the measurement tech-
niques employed. However, some of the instruc-
tions designated by the Medi-Cal statutes were
unclear, specifically section 531–11 (c) (2), which
states that only teeth that are visible in the
study model should be considered. This is later
contradicted in section 531–11(c) (6) where exam-
ples of ectopic eruption are described. Our obser-
vations are in agreement with those of Parker
(1998) (16) who recognized the confusion sur-
rounding the definition of ectopic. In his paper, he
provided some clarification for this definition as it
is applied to the HLD (CalMod). However, he
failed to address how ‘teeth in the maxillary sinus
and ascending ramus of the mandible’ (clearly
unerupted, ectopic teeth) will be visible in the
study model.
Our analysis showed that the HLD (CalMod)
was well correlated with the gold standard. How-
ever, the cut-off for HLD (CalMod) did not corre-
spond with the gold standard threshold level for
models that displayed handicapping malocclu-
sions. A good example of this discrepancy is
shown in Fig. 1. This specific case was given a
gold standard score of 11.0, which falls in the
treatment essential category. However, the HLD
(CalMod) index only gave this case a score of 12,
well below the cut-off of 26.
Our results are in agreement with a previous
study comparing the HLD (Md) with the HLD
(CalMod) index (4) that identified several
weaknesses in the latter. For example, it fails to
identify localized crowding and missing teeth or
spacing in the anterior segment (Fig. 2). It does not
weight open bites heavily enough and fails to
identify bilateral crossbites. Further, those maloc-
clusions with fairly aligned arches and large overjet
(Figs 3 and 4) that fall short of the 9 mm qualifying
exception would not qualify under the HLD (Cal-
Mod) index.
In the late 90s, approximately 1 700 000 people
were eligible for MediCal-mandated orthodontic
Table 5. Sensitivity and specificity of HLD (CalMod)with a cut-off of 18.5
HLD (CalMod)with cut off of 18.5
Gold standard
Handicappingmalocclusion
Nohandicappingmalocclusion
Handicappingmalocclusion
15 9
NO handicappingmalocclusion
12 117
Sensitivity (%) 55.6Specificity (%) 92.9Prevalence (%) 17.7(+) Predictivevalue (%)
62.5
()) Predictivevalue (%)
90.7
Fig. 2. This model had a gold standard score of 11.0 (out of a maximum of 12) and an HLD (CalMod) score of 12.
54
Cooke et al.
benefits in the state of California. Of this group,
160 745 potential patients were assessed for eligi-
bility by the HLD (CalMod) index. The orthodon-
tists involved determined that 58 193 (36%) of this
number qualified for treatment. Study models of
these patients were then measured and scored
again by a group of orthodontists at the fiscal
intermediary. Only 27 637 were approved for
treatment, less than half (47%) of the patients
already approved by the orthodontist providers
and only 17% of all the referred patients (20).
The present study showed that the expert opin-
ion of a panel of orthodontist considered two and
one-half times more patients to have handicapped
malocclusion (27) than the HLD (CalMod) index
(11). When the cut-off is reduced to 18.5, the HLD
(CalMod) index increases the number of patients
with handicapped malocclusions to 24.
The expert opinion of a group of orthodontists is
considered the ‘gold standard’. It is against this
‘gold standard’ that occlusal indexes, such as the
HLD, are validated. Why is it then possible to find
such a discrepancy between the HLD (CalMod)
index and the gold standard in terms of treatment
that is deemed medically necessary? Why is there a
difference between the opinions of the orthodon-
tists reviewing the cases as the fiscal intermediaries
and the orthodontists submitting the cases to be
approved for treatment? Are both sets of opinions
not considered to be the ‘gold standard’? As a
result of the differences in opinion between
these parties, 53% of the cases submitted by
orthodontists are refused Medi-Cal funded treat-
ment. It has been suggested that better monitoring
and better training in the use of the index are
needed for better ‘budget control’ (20). We believe,
Fig. 3. This model had a gold standard score of 10.8 (out of a maximum of 12) and an HLD (CalMod) score of 24.
Fig. 4. This model had a gold standard score of 10.6 (out of a maximum of 12) and an HLD (CalMod) score of 18.
55
Assessment of the validity of HLD
given our findings, that a comprehensive revision
of HLD (CalMod) use in the State of California
should be undertaken not solely for ‘budget con-
trol’ but to achieve what this index was designed to
accomplish: to indentify the patients with handi-
capping malocclusions.
Another issue that arises from our results is the
sensitivity and specificity of this index. Sensitivity
refers to the proportion of people with disease who
have a positive test result whereas specificity refers
to the proportion of people without disease who
have a negative test result. When an index has a high
sensitivity, a negative result rules out the diagnosis.
On the other hand, when an index has a high
specificity, a positive result rules in the diagnosis.
The HLD (CalMod) index has a low sensitivity
(25.9%), meaning that a score below the cut-off
score of 26 does not rule out the possibility that the
case has handicapping malocclusion and that
orthodontic treatment would be needed. However,
a score of 26 or higher strongly indicates a patient
with orthodontic treatment need. This conclusion is
corroborated by the large number of false negatives
we found (Table 4) and also by several examples
such as the one shown in Fig. 1.
When the sensitivity and specificity were deter-
mined using the new cut-off of 18.5 for HLD (Cal
Mod), specificity decreased (92.9%) and sensitivity
increased dramatically (55.6). Beglin et al. (14)
demonstrated that, with a more lenient cut-off of
12, sensitivity increased to (91.7%) and specificity
decreased to 82.3%.
Poor sensitivity may be a result of cut-off based
on the index’s author’s personal opinion, or a result
of a policy decision that was made without a
validation process to determine a cut-off that
reflects the professional experts’ opinion (gold
standard) (6). If the latter is true, it is then a
question of policy as to where the cut-off will be
placed.
The positive predictive value of a test is the
probability that the patient has the condition being
studied when the test for the condition gives a
positive result. The positive predictive value for
HLD (CalMod) was 63.6 meaning that 63.6% of the
patients with scores of 26 or greater actually have
orthodontic treatment needs or handicapping mal-
occlusions. Conversely, the negative predictive
value is the probability that the patient does not
have the condition being studied (in this case,
orthodontic treatment need) when the test for the
condition is negative. The negative predictive value
for HLD (CalMod) was 85.9% signifying that
almost 86% of the models that score 26 or below
do not need orthodontic treatment.
This is somehow a high value. However, unlike
specificity and sensitivity, predictive values are
heavily influenced by the prevalence and less
influenced by the cut-off level. For instance even
when the cut-off was lowered to 18.5, the positive
predictive value was still low at 62.5% and
the negative predictive values was still high at
90.7%.
In the State of California, the HLD (CalMod)
index is used to identify those with handicapping
malocclusions with the premise that funding
should be provided based on a physical handicap
caused by the malocclusion. Our results show that
the HLD (CalMod) with a cut-off of 26 fails to
indentify a considerable percentage of handicap-
ping malocclusions. We recommend further stud-
ies examining patient characteristics that might be
associated with misclassification.
References1. Carlos JP. Evaluation of indices of malocclusion. Int
Dent J 1970;20:606–17.2. Tang EL, Wei SH. Recording and measuring maloc-
clusion: a review of the literature. Am J OrthodDentofacial Orthop 1993;103:344–51.
3. Shaw WC, Richmond S, O’Brien KD. The use ofocclusal indices: a European perspective. Am JOrthod Dentofacial Orthop 1995;107:1–10.
4. Han H, Davidson WM. A useful insight into 2occlusal indexes: HLD(Md) and HLD(CalMod). Am JOrthod Dentofacial Orthop 2001;120:247–53.
5. American Association of Orthodontists. AmericanAssociation of Orthodontists Bulletin 2001(Fall 2001).
6. Younis JW, Vig KW, Rinchuse DJ, Weyant RJ. Avalidation study of three indexes of orthodontictreatment need in the United States. CommunityDent Oral Epidemiol 1997;25:358–62.
7. Draker HL. Handicapping labio-lingual deviations:A proposed index for public health purposes. Am JOrthod 1960;46:295–305.
8. Code of Maryland Regulations. Attachment No 1,Baltimore, MD: EPSDT Services Regulation; 1992.
9. DeGuzman L, Bahiraei D, Vig KW, Vig PS, WeyantRJ, O’Brien K. The validation of the Peer AssessmentRating index for malocclusion severity and treatmentdifficulty. Am J Orthod Dentofacial Orthop1995;107:172–6.
10. Parker WS. A study of 1000 malocclusions selectedby the HLD (CalMod) Index. Am J Orthod Dento-facial Orthop 1999;115:343–51.
11. Brook PH, Shaw WC. The development of an indexof orthodontic treatment priority. Eur J Orthod1989;11:309–20.
12. Richmond S, Daniels CP. International comparisonsof professional assessments in orthodontics: Part
56
Cooke et al.
2–treatment outcome. Am J Orthod DentofacialOrthop 1998;113:324–8.
13. Richmond S, Shaw WC, O’Brien KD, Buchanan IB,Jones R, Stephens CD et al. The developmentof the PAR Index (Peer Assessment Rating):reliability and validity. Eur J Orthod 1992;14:125–39.
14. Beglin FM, Firestone AR, Vig KW, Beck FM, KuthyRA, Wade D. A comparison of the reliability andvalidity of 3 occlusal indexes of orthodontic treat-ment need. Am J Orthod Dentofacial Orthop2001;120:240–6.
15. Parker WS. Useful data from application of the HLD(CalMod) Index. Am J Orthod Dentofacial Orthop2000;117:435–7.
16. Parker WS. The HLD (CalMod) index and the indexquestion. Am J Orthod Dentofacial Orthop1998;114:134–41.
17. Breiman L. Classification and regression trees. Bel-mont, Calif.: Wadsworth International Group; 1984.
18. Salford Systems; available at: http://salford-sys-tems.com/cart.php [last accessed 20 September2009], 2009.
19. Gansky SA. Dental data mining: potential pitfalls andpractical issues. Adv Dent Res 2003;17:109–14.
20. Theis JE, Huang GJ, King GJ, Omnell ML. Eligibilityfor publicly funded orthodontic treatment deter-mined by the handicapping labiolingual deviationindex. Am J Orthod Dentofacial Orthop 2005;128:708–15.
57
Assessment of the validity of HLD