Ideally, classification systems are used to assess a clinical entity,
enable a surgeon to recommend specific treatment, and allow comparison
of different treatment methods1.
In 1983, King et al. measured scoliotic deformities on coronal radiographs,
described five thoracic curve types, and recommended specific vertebral
levels to be included in a spinal arthrodesis2.
All of the patients in their study underwent a spinal arthrodesis
with Harrington rod instrumentation to correct the coronal plane
deformity. King et al. did not include thoracolumbar, lumbar, or
double or triple major curves in their classification.
The King classification has continued to be utilized despite the
increasing acceptance of the need to consider scoliosis as a three-dimensional
deformity when considering operative intervention and the use of
segmental spinal fixation3-8.
Recently, the interobserver and intraobserver reliability of the
King classification was found, by two groups of surgeons working
independently, to have only poor-to-fair validity, reliability,
and reproducibility1,9. The poor
reliability of the King classification indicated the need for a
new classification that would (1) be comprehensive and include all
types of curves, (2) emphasize consideration of sagittal alignment,
(3) help to define treatment that could be standardized, (4) be
based on objective criteria for each curve type, (5) have good-to-excellent
interobserver and intraobserver reliability, and (6) be easily understood
and of practical value in the clinical setting.
We developed a new system for classification of adolescent idiopathic
scoliosis, on the basis of radiographs made in the coronal and sagittal
planes, that could be used to determine the appropriate vertebral
levels to be included in an arthrodesis. Two groups of surgeons
tested this new classification, and the results were used to calculate
interobserver and intraobserver reliability.
Four radiographs of the spine (standing long-cassette coronal and
lateral as well as right and left supine side-bending views) for
each of twenty-seven patients were reviewed by two groups of surgeons.
One of the groups consisted of five surgeons, members of the Scoliosis
Research Society, who had developed the new system and who had previously
reviewed the same radiographs to test the reliability of the King
classification9. The other group
consisted of seven randomly selected members of the Scoliosis Research
Society who had not been involved in the development of this new
system. The scoliosis was classified according to curve type (1
through 6) combined with a lumbar spine modifier (A, B, or C) and
a sagittal thoracic modifier (-, N, or +). The definitions
established by the Scoliosis Research Society were used to determine
the type of curve.
Thoracic curves, the apex of which is located between the second
thoracic vertebral body and the eleventh and twelfth thoracic intervertebral
disc, include proximal thoracic curves with the apex at the third,
fourth, or fifth thoracic level and main thoracic curves with the
apex between the sixth thoracic body and the eleventh and twelfth
thoracic disc. The apex of thoracolumbar curves is located between
the cephalad border of the twelfth thoracic vertebra and the caudad
border of the first lumbar vertebra. The apex of lumbar curves is
located between the first and second lumbar disc and the caudad
border of the fourth lumbar vertebra. Structural curves, described by
their location, lack normal flexibility and are termed as major
(if they have the largest Cobb measurement) or minor. Minor curves
can be structural or nonstructural.
In order to simplify the classification, there was some overlap between
the structural characteristics of the minor curves. A structural
proximal thoracic curve has a minimum residual coronal curve on
side-bending radiographs of at least 25° (with or without a positive
first thoracic tilt) and/or kyphosis (from the second to
the fifth thoracic level) of at least +20°. A structural
main thoracic curve has a minimum residual coronal curve of at least
25° and/or thoracolumbar kyphosis (from the tenth thoracic
to the second lumbar level) of at least +20°. A structural
thoracolumbar/lumbar curve also has a minimum residual
coronal curve of at least 25° and/or thoracolumbar kyphosis
(from the tenth thoracic to the second lumbar level) of at least +20°
even though sagittal malalignment may be due to a rotational deformity
instead of a true kyphosis. A minor curve is structural if these
criteria are present. On the basis of this classification, we propose
that spinal arthrodesis include only the major curve and structural
minor curves.
Curve Types (1 through 6)
Curve types (Table I) are based on the identification
of the major curve and the structural characteristics of the minor
curves.
Type 1—main thoracic: The main thoracic
curve is the major curve, and the proximal thoracic and thoracolumbar/lumbar
curves are minor nonstructural curves.
Type 2—double thoracic: The main thoracic
curve is the major curve, while the proximal thoracic curve is minor
and structural and the thoracolumbar/lumbar curve is minor
and nonstructural.
Type 3—double major: The main thoracic
and thoracolumbar/lumbar curves are structural, while the
proximal thoracic curve is nonstructural. The main thoracic curve
is the major curve and is greater than, equal to, or no more than
5° less than the Cobb measurement of the thoracolumbar/lumbar
curve.
Type 4—triple major: The proximal thoracic,
main thoracic, and thoracolumbar/lumbar curves are all
structural; either of the two latter curves may be the major curve.
Type 5—thoracolumbar/lumbar: The
thoracolumbar/lumbar curve is the major curve and is structural.
The proximal thoracic and main thoracic curves are nonstructural.
Type 6—thoracolumbar/lumbar-main
thoracic: The thoracolumbar/lumbar curve is the
major curve and measures at least 5° more than the main thoracic
curve, which is structural. The proximal thoracic curve is nonstructural.
If the difference between the lumbar and thoracic curves is <5°,
the scoliosis can be categorized as type 3, 4, or 5 on the basis
of the structural characteristics of the main thoracic and thoracolumbar/lumbar
regions. For the sake of clarity, the major curve (the curve with
the largest Cobb measurement) always distinguishes between type
3 (main thoracic curve is major) and type 6 (thoracolumbar/lumbar
curve is major). If the Cobb measurements of the main thoracic and
thoracolumbar/lumbar curves are equal, then the thoracic
curve is considered the major curve. Thus in Figures 4-A, 4-B, 4-C, 4-D, 4-E, and 4-F, the curve classification
is type 6.
Lumbar Spine Modifiers (A, B, or C)
When operative intervention is being considered, the degree of
lumbar deformity must be assessed because it alters spinal balance
and affects proximal curves. Three types of lumbar deformity were
defined on the basis of the relationship of the center sacral vertical
line to the lumbar curve as noted on the coronal radiograph (Figs. 1-A and 1-B). The center sacral
vertical line should bisect the cephalad aspect of the sacrum and
be perpendicular to the true horizontal. Pelvic obliquity secondary
to limb-length inequality of <2 cm is ignored unless the
surgeon believes that the pelvic obliquity increases the degree
of spinal deformity. In those cases and when the discrepancy is >2
cm, the coronal radiograph is made with the appropriately sized
lift under the short limb. The center sacral vertical line is extended
in a cephalad direction, and the most cephalad lumbar or thoracic
vertebra most closely bisected by the line is considered the stable
vertebra. If a disc is most closely bisected by the center sacral vertical
line, then the vertebra caudad to it is deemed to be the stable
vertebra. The apex of a thoracolumbar or lumbar curve is the most
horizontal and laterally placed vertebral body or intervertebral
disc.
Modifier A: Modifier A is used when the center sacral vertical line
runs between the lumbar pedicles to the level of the stable vertebra.
The curve must have a thoracic apex at or cephalad to the eleventh
and twelfth thoracic disc level. Therefore, modifier A can be used
only for a main thoracic curve (types 1 through 4) and cannot be
used to define thoracolumbar/lumbar curves (types 5 and
6). It also should not be used when the center sacral vertical line
falls directly against the medial aspect of the lumbar apical pedicle.
Modifier B: Modifier B is used when, because of deviation of the
lumbar spine from the midline, the center sacral vertical line touches
the apex of the lumbar curve, between the medial border of the lumbar
concave pedicle and the concave lateral margin of the apical vertebral
body or bodies (if the apex is a disc). These curves all have an
apex in the main thoracic region, so thoracolumbar/lumbar
curves are excluded. This modifier is also used when there is any
doubt about whether the center sacral vertical line is, in fact,
apposed to the lateral margin of the apical vertebral body or bodies.
Modifier C: Modifier C is used when the center sacral vertical line
falls completely medial to the entire concave lateral aspect of
the thoracolumbar or lumbar apical vertebral body or bodies (if
the apex is a disc). These deformities of the lumbar spine may have
a major curve with the apex at the thoracic, thoracolumbar, or lumbar
level. When the center sacral vertical line does not fully lie off
the lateral aspect of the apical thoracolumbar or lumbar vertebra
or the apex is not clearly lateral to the center sacral vertical
line, then modifier B should be used. Modifier C may include all
major main thoracic curves (types 1 through 4) and must include
all thoracolumbar/lumbar curves (types 5 and 6).
The three lumbar modifiers can be used to define the alignment
of the lumbar spine in relation to the six curve types, and they
can be used to assess the position of the lumbar spine after operative
intervention10.
Sagittal Thoracic Modifiers (-, N, or +)
The sagittal alignment of the thoracic spine is another critical factor
when operative intervention is being considered for patients who
have adolescent idiopathic scoliosis4,11.
In an attempt to address this concern, we devised a sagittal thoracic
modifier to further define the six curve types. The mean normal
sagittal thoracic alignment from the fifth to the twelfth thoracic
vertebra is +30° with a range of +10° to +40°3. Patients who have adolescent idiopathic
scoliosis tend to have decreased thoracic kyphosis or even thoracic
lordosis in comparison with normal controls3,4.
The sagittal thoracic modifiers were determined by measurements
from the superior end-plate of the fifth thoracic vertebra to the
inferior end-plate of the twelfth thoracic vertebra on a standing
lateral radiograph. A minus (-) sign (hypokyphosis) identified a
curve of less than +10°, N (normal kyphosis) identified
a curve of +10° to +40°, and a plus (+)
sign (hyperkyphosis) identified a curve of more than +40°.
Classification of Curve Types
First the specific curve type (1 through 6) should be identified and
then the lumbar spine modifier (A, B, or C) and sagittal thoracic
modifier (-, N, or +) should be defined in order to determine
the exact, complete classification of the curve (for example, 1A-,
1AN, 6CN, and so on) (Fig. 2).
Curve Reliability
Coronal and sagittal Cobb measurements and the center sacral vertical
line were drawn on each radiograph. All of the radiographs were
interpreted on one day and then reinterpreted a day later, in a
different sequence. The two groups of reviewers were asked to choose
the appropriate curve type (1 through 6), lumbar spine modifier
(A, B, or C), and sagittal thoracic modifier (-, N, or +).
Interobserver and intraobserver reliability were estimated by calculating
the kappa coefficient values of simple and weighted components at
95% confidence intervals established with SAS software
(SAS Institute, Cary, North Carolina). The kappa value was the balance
of the part of agreement that would occur by random chance subtracted from
the actual agreement. Thus, kappa coefficients ranged from +1
(perfect agreement), to 0 (chance agreement), to -1 (less agreement
than expected by chance). Svanholm et al. suggested that kappa values
of >0.75 represent good or excellent reliability; 0.5 to
0.75, fair reliability; and <0.5, poor reliability12.
Finally, one of us (L.G.L.) retrospectively reviewed radiographs
of 315 consecutive patients with operatively treated scoliosis to
assess the prevalence of curve types in a typical surgical practice.
Curve Classification
Among the five surgeons who had developed the new system, the
mean interobserver reliability for determining the curve type was
93% (range, 85% to 100%), with a mean
kappa value of 0.92 (range, 0.83 to 1.00), indicating good-to-excellent
reliability (Table II).
When the same five reviewers had applied the King classification,
the mean interobserver reliability had been 64%, with a
mean kappa value of 0.49, indicating poor reliability9. Using the new system, all five reviewers
agreed about the curve type in twenty-two patients and four reviewers
agreed about the curve type in the remaining five patients. Using
the King classification, all five had agreed about the curve type
in only one of the twenty-seven patients. The intraobserver reliability
of these surgeons with the new classification was 85% (range,
72% to 100%), with a mean kappa value of 0.83 (range,
0.79 to 1.0), indicating good-to-excellent reliability. The mean
intraobserver reliability when these five surgeons had used the
King classification had been 69%, with a mean kappa of
0.62, indicating fair reliability9.
In the group of seven reviewers who had been randomly selected
from the Scoliosis Research Society, the kappa values for interobserver
reliability with the new classification were 0.740 (range, 0.384
to 1.000) for the curve type, 0.800 (0.738, 0.763, and 0.880) for
the lumbar modifier, and 0.938 (0.901, 0.930, and 1.000) for the
sagittal thoracic modifier (Table III). The respective values for intraobserver
reliability were 0.893 (range, 0.75 to 1.00), 0.840 (range, 0.66
to 1.00), and 0.970 (range, 0.93 to 1.00) (Table IV). These values
represent good-to-excellent reliability except for the interobserver
reliability for curve type, which was 0.01 below the 0.75 level
for good-to-excellent reliability.
Clinical Testing of the Classification System
In the consecutive series of 315 patients who had been treated operatively
by one of us (L.G.L.), the structural regions were included in the
arthrodesis, commensurate with the predictions derived from the
classification system, in 284 patients (Table V). The main thoracic curve (type
1) was the most prevalent type of curve, being present in 126 (40%)
of the 315 patients. The double thoracic curve (type 2) (fifty-six
patients), the double major curve (type 3) (fifty-eight patients),
and the thoracolumbar/lumbar curve (type 5) (fifty-six
patients) were the next most common curves, with each being identified
in 18% of the patients. The triple major curve (type 4)
(eight patients) and the thoracolumbar/lumbar-main thoracic
curve (type 6) (eleven patients) each had a prevalence of only 3%.
The lumbar spine modifier A defined the curve in ninety-four (30%) of
the patients; the modifier B, in sixty-seven (21%); and
the modifier C, in 154 (49%). The sagittal thoracic modifier revealed
hypokyphosis (-) in fifty-six (18%) of the patients, normal
kyphosis (N) in 224 (71%), and hyperkyphosis (+)
in thirty-five (11%).
The structural characteristics in the sagittal plane were critical factors
in the development of this classification system, since sagittal
alignment determines the regions of the spine to be included in
an arthrodesis. Hyperkyphosis in the proximal thoracic and thoracolumbar/lumbar
regions and lack of flexibility of the curve on side-bending are
important components of the minor curves. The extent of arthrodesis
and instrumentation of the main thoracic curve, the major curve
in types 1 through 4, will be influenced by an increased kyphosis
in the proximal thoracic and thoracolumbar/lumbar regions.
In types 5 and 6, the thoracolumbar/lumbar curve is the
major curve, and the main thoracic curve is nonstructural in type
5 and structural in type 6.
The center sacral vertical line, used to define the lumbar modifiers,
does not account for any pelvic obliquity, unlike the center sacral
line used in the King classification2.
Lumbar modifier A indicates that there is minimal or no scoliosis,
and modifier B indicates mild-to-moderate scoliosis. Curves assigned
lumbar modifier B were difficult to classify with the King system;
hence they were termed type II or III on the basis of the appearance
of the lumbar curve on standing and side-bending radiographs2,9. We propose that, when curves are
assigned lumbar modifier A or B, the lumbar spine should not be
included in the arthrodesis unless there is a kyphosis of at least +20°
in the thoracolumbar region. The curves that are assigned lumbar modifier
C had been previously classified as King type I or II, or occasionally
as type V, and also include all double major, triple major, and
thoracolumbar and lumbar curves. In many cases, when a curve is
assigned lumbar modifier C, the lumbar spine probably should be
included in the arthrodesis. However, patients who have a 1C or
2C curve may have a selective thoracic arthrodesis as long as an
acceptable balance of the lumbar curve is maintained10.
The deviation of the lumbar curve from the center sacral vertical
line increases from modifier A, to modifier B, to modifier C, with
a corresponding increase in malalignment. The need for and method
of correction of malalignment of the lumbar spine can be assessed
more accurately with use of these lumbar spine modifers10,13. We believe that increased consistency
in the assessment of various curve types will, in the future, permit
comparative analysis of different treatments9.
A sagittal thoracic modifier was developed because of the importance
of assessing this area of the spine when determining the need for
operative treatment4,11, cosmesis,
operative approach6,14-16, type
of instrumentation5,8,17,18, and
potential for decreased pulmonary function in patients who have
loss of normal kyphosis or who have true thoracic lordosis19. Dickson suggested that lateral
translation of the spine is preceded by apical thoracic lordosis11. A frequent criticism of the King
classification is the lack of analysis of sagittal-plane alignment9.
A selective thoracic arthrodesis of a type-1 curve with any lumbar
modifier (A, B, or C), previously classified as King type II or
III, has often led to coronal decompensation with segmental spinal
instrumentation6,15,18,20-22.
The thoracic and lumbar curves, considered to be false double major
curves, both cross the midline; however, a selective thoracic arthrodesis
with anterior or posterior instrumentation can often still be performed10,14. The lumbar curve should correct
to <25° on side-bending, and thoracolumbar kyphosis should
not be present6,10,15,18,20-22.
In addition, the thoracic rotation should be more prominent than
the lumbar rotation23. Of the
126 patients with a type-1 curve in our clinical series, 114 had
only the main thoracic region included in the arthrodesis.
Type-2A curves (-, N, or +) include structural proximal
thoracic and major main thoracic curves and a nonstructural thoracolumbar/lumbar
curve2,16,17. Any structural thoracic
or lumbar curve may also be associated with a structural proximal
thoracic curve that has a residual curve of 25° on side-bending
and/or increased kyphosis in the proximal thoracic region.
The proximal structural curve in a type-2B thoracic curve has similar
characteristics9. Unlike the King
classification, the creation of a 2C group permits separation of
the proximal thoracic and thoracolumbar/lumbar components.
Of our fifty-six patients with a type-2 curve, fifty had both the
proximal thoracic and the main thoracic components included in the
arthrodesis.
Type-3A and 3B (-, N, or +) double major curves are
rare and include structural main thoracic and thoracolumbar/lumbar curves.
The residual lumbar component is of large magnitude and is structural
in the coronal or sagittal plane even though the lumbar spine does
not completely deviate from the midline. In a type-3C (-, N, or +)
double major curve, the lumbar curve is structural and deviates
completely from the midline. Thus, the main thoracic and thoracolumbar/lumbar
components should be considered for the arthrodesis. In our clinical series,
fifty-six of the fifty-eight patients with a type-3 curve had the
main thoracic and thoracolumbar/lumbar components included
in the arthrodesis.
All three regions (proximal thoracic, main thoracic, and thoracolumbar/lumbar)
in type-4A and 4B (-, N, or +) triple major curves are
structural, and the main thoracic or thoracolumbar/lumbar
component is the major curve. The lumbar spine does not deviate
completely from the midline, and when there is a large thoracic
curve the residual lumbar component is sufficiently large to be
inflexible on side-bending or to have a thoracolumbar kyphosis.
The lumbar curve in type 4C deviates completely from the midline,
as expected with a large structural thoracolumbar/lumbar
curve. Of the eight patients with a type-4 curve, six had all three
regions (proximal thoracic, main thoracic, and thoracolumbar/lumbar)
included in the arthrodesis.
Thoracolumbar/lumbar major curves include type 5C, which has
a nonstructural main thoracic component, and type 6C, which has
a structural main thoracic component. When a patient has a type-5
curve, the arthrodesis should include only the thoracolumbar/lumbar
region, whereas when a patient has a type-6 curve, the major thoracolumbar/lumbar
and minor main thoracic structural curves both should be included
in the arthrodesis (Figs. 4-A, 4-B, 4-C, 4-D, 4-E, and 4-F). In our series, forty-nine of
the fifty-six patients who had a type-5 curve had only the thoracolumbar/lumbar
region included in the arthrodesis; nine of the eleven with a type-6 curve
had the main thoracic and thoracolumbar/lumbar components
included in the arthrodesis.
The interobserver reliability of the five reviewers who had designed
the new system was 93% for the new system and 64% for
the King classification; the intraobserver reliability was 85% for
the new system and 69% for the King classification. These
five reviewers found the new system to be more reliable than the
King classification. The interobserver and intraobserver reliability
of the new classification when used by the seven surgeons who had
not been involved with its development was good-to-excellent12 for all of the components except
for interobserver reliability for curve type. Reports1,9 have shown that the King classification
is not reliable; therefore, because the criteria for each curve
type are imprecise, the results of different treatment methods cannot
be compared with use of that system.
Ideally, a classification system should reflect the current concept
of a three-dimensional analysis of scoliotic deformities. An initial
effort to include a grade for lumbar alignment in the axial plane
as one of the structural criteria was found to be difficult to reproduce
and thus was abandoned. This difficulty was related to problems
with accurately assessing an axial plane deformity on biplanar radiographs.
In the absence of a reliable, simple, and universally accepted method
of three-dimensional modeling of scoliotic deformities, two-dimensional
radiographs (coronal and sagittal) remain the standard. Though the
new classification is based on two-dimensional radiographs, the
inclusion of sagittal thoracic and coronal lumbar modifiers suggests
that axial thoracic and lumbar modifiers can be included in the
analysis when and if methods for three-dimensional analysis of scoliotic
deformities become universally available.
One criticism of this classification concerns the definition
of structural minor curves. There has been no universally accepted
and reproducible definition of a structural minor curve. Previously
published data, simplified to facilitate their use and to avoid
variations, were employed to define the characteristics of structural
minor curves4,8,10,16,18. The
new classification permits minor modifications and inclusion of
clinical findings to determine whether a curve is structural. Grading
the curve as structural does not suggest that all structural minor
curves, regardless of magnitude, are to be included in the arthrodesis.
This is highlighted by the fact that, in the consecutive series
of 315 operatively treated patients, thirty-one did not have structural
regions included in the arthrodesis or had nonstructural regions
included in the arthrodesis. Often the specific characteristics
and ratios of structural curves, such as the degree of curvature,
degree of apical vertebral translation and rotation, and degree
of flexibility on side-bending, are more important than the absolute values.
These criteria, assessed when an investigator is attempting to differentiate
a true King type-II curve (new type 1C) from a double major curve
(new type 3C)15, need to be evaluated
in relation to the new classification. The clinical examination
of the patient is also a critical component in the surgical decision-making
process and may override radiographic information for certain curve
patterns23.
A valid criticism of this classification is that forty-two different
curve patterns can be derived. This complexity may deter the busy
orthopaedic surgeon from using the classification in clinical practice.
In defense of the new system, the six specific curve types should
be well known to surgeons who treat scoliosis. The characteristics
of the curve types should be clearly understood, and then the lumbar
and sagittal thoracic modifiers can be added to the curve type,
thereby providing additional information to assist in determining
the appropriate treatment (Fig. 3).
At present, it is unknown whether this new classification is easy
to use, universally acceptable, or useful in clinical practice.
The classification will no doubt undergo changes as advances are
made in imaging techniques. We hope that it will prove to be useful
as a thorough two-dimensional radiographic evaluation predictive
of spinal regions to be included in the arthrodesis and that it
is a prelude to three-dimensional classification.
Note: The authors gratefully acknowledge the careful review and
comments provided by George Bassett, MD, Steven Glassman, MD, David
Godfried, MD, Thomas Haher, MD, James Hardacker, MD, Serena Hu,
MD, John Lubicky, MD, Andrew Merola, MD, Peter Newton, MD, John
Sarwark, MD, Paul Sponseller, MD, and Dennis Wenger, MD. They also thank
Biedermann-Motech for research support; Jack D. Baty, PhD, and Paul
Thompson, PhD, for statistical analysis; and Gail Huss, RN, and
Sally McGlothen, RN, for data collection. The classification system
was developed in association with the Harms Scoliosis Study Group,
1995 through 1999. Finally, they acknowledge the participation of
the independent reviewers: John Dimar, MD, John Flynn, MD, Robert Gaines,
MD, Andrew King, MD, William Lauerman, MD, Richard McCarthy, MD,
and Alan Moskowitz, MD.