Copyright 1997 by The Journal of Bone and Joint Surgery, incorporated

A Meta-Analysis of the Efficacy of Non-Operative Treatments for Idiopathic Scoliosis*1

BY DALE E. ROWE, M.D.t, SAUL M. BERNSTEIN, M.D.§, MAX F. RIDDICK, M.D.I, FEDERICO ADLER, M.D.#,

JOHN B. EMANS, M.D.**, AND DARYLE GARDNER-BONNEAU, PH.D4, KALAMAZOO, MICHIGAN

Investigation performed at the Kalamazoo Center for Medical Studies, Kalamazoo

ABSTRACT: With use of data culled from twenty studies, members of the Prevalence and Natural His­tory Committee of the Scoliosis Research Society con­ducted a meta-analysis of 1910 patients who had been managed with bracing (1459 patients), lateral electrical surface stimulation (322 patients), or observation (129 patients) because of idiopathic scoliosis. Three vari­ables — the type of treatment, the level of maturity, and the criterion for failure — were analyzed to deter­mine which had the greatest impact on the outcome. We also examined the effect of the type of brace that was used and the duration of bracing on the success of treatment.

The number of failures of treatment in each study was determined by calculating the total number of patients who had unacceptable progression of the curve (as defined in the study), who could not comply with or tolerate treatment, or who had an operation. The percentage of patients who completed a given course of treatment without failure, adjusted for the sample sizes of the studies in which that treatment was used, yielded the weighted mean proportion of success for that treatment.

The weighted mean proportion of success was 0.39 for lateral electrical surface stimulation, 0.49 for obser­vation only, 0.60 for bracing for eight hours per day, 0.62 for bracing for sixteen hours per day, and 0.93 for bracing for twenty-three hours per day. The twenty-three-hour regimens were significantly more successful than any other treatment (p < 0.0001). The difference

*No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article. No funds were received in support of this study.

tThis study was conducted by the Prevalence and Natural His­tory Committee of the Scoliosis Research Society.

^Michigan State University/Kalamazoo Center for Medical Stud­ies, Linda Richards Building, 1000 Oakland Drive, Kalamazoo, Michi­gan 49008. Please address requests for reprints to Dr. Rowe. E-mail address for Dr. Rowe: rowe@kcms.msu.edu.

§Department of Orthopaedic Surgery, Southern California Or­thopaedic Institute, University of Southern California, 6815 Noble Avenue, Van Nuys, California 91405.

IJewett Orthopaedic Clinic, 515 West Highway 434, Suite 210, Longwood, Florida 32750.

#Section of Orthopaedic Surgery, Department of Surgery, Uni­versity of Kansas Medical Center, 39th and Rainbow Boulevard, Kansas City, Missouri 66103.

"Department of Orthopaedic Surgery, Children's Hospital, 300 Longwood Avenue, Boston, Massachusetts 02115.

between the eight and sixteen-hour regimens was not significant, with the numbers available. Although lat­eral electrical surface stimulation was associated with a lower weighted mean proportion of success than observation only, the difference was not significant, with the numbers available.

This meta-analysis demonstrates the effectiveness of bracing for the treatment of idiopathic scoliosis. The weighted mean proportion of success for the six types of braces included in this review was 0.92, with the highest proportion (0.99) achieved with the Mil­waukee brace. We found that use of the Milwau­kee brace or another thoracolumbosacral orthosis for twenty-three hours per day effectively halted progres­sion of the curve. Bracing for eight or sixteen hours per day was found to be significantly less effective than bracing for twenty-three hours per day (p < 0.0001).

Adolescent idiopathic scoliosis, as defined by the Scoliosis Research Society49, is diagnosed when a lat­eral spinal curve of at least 11 degrees is observed in a patient who is between ten years old and skeletal matu­rity. In the United States, this condition affects approx­imately 1 to 3 per cent of children between the ages of ten and sixteen years58.

Natural History

Without intervention, the curve is likely to progress between the time of detection and the time of skeletal maturity; the risk of progression increases as the degree of curvature increases59. Nachemson et al.41, in a study of untreated female patients who had thoracic scoliosis, suggested that the risk of progression increases with the magnitude of the curve at the time of detection and decreases with increased age at the time of detection (Table I). Younger girls (ten, eleven, or twelve years old) who had a curve of at least 30 degrees at the time of detection had the highest likelihood of progression, ranging from 90 to 100 per cent.

Bunnell performed a retrospective study of the nat­ural history of idiopathic scoliosis before skeletal matu­rity in 326 untreated female and male patients6. Of the 123 patients who had had a curve of less than 50 de­grees at the time of diagnosis, 68 per cent had more than 5 degrees of progression and 34 per cent, more than 10 degrees of progression.

664 THE JOURNAL OF BONE AND JOINT SURGERY

A META-ANALYSIS OF THE EFFICACY OF NON-OPERATIVE TREATMENTS FOR IDIOPATHIC SCOLIOSIS 665

TABLE I

RISK OF PROGRESSION OF THORACIC SCOLIOSIS IN GIRLS41

Magnitude of Curve at

Detection (Degrees)

<19

20-29

30-59

>60

Risk of 10,11, or 12 Yrs.*

25

60

90

100

Progression (Per cent)

'• 13, 14, or 15 Yrs.*

10

40

70

90

>16Yrs.*

0

10

30

70

'Skeletal age at detection.

Weinstein et al. examined the natural history of id­iopathic scoliosis to determine prognostic factors59. In 120 patients who were followed for a mean of approx­imately forty years, forty-four curves progressed 5 de­grees or more after skeletal maturity. Progression was most rapid between the time of detection and skele­tal maturity, with the next most rapid progression oc­curring between skeletal maturity and the time of the thirty-year follow-up examination; the least progression occurred between the thirty and forty-year follow-up examinations. A Cobb angle10 of more than 30 degrees at the time of skeletal maturity in patients who had a lumbar or thoracolumbar curve and an angle of more than 50 degrees in those who had a thoracic curve or combined curves indicated a high likelihood of pro­gression, as did apical vertebral rotation of more than 30 degrees in patients who had a thoracic, lumbar, or thoracolumbar curve. Most patients in that series had severe curvature at the time of diagnosis, and many of the untreated curves continued to progress throughout the lifetime of the patients.

Current Options for .Treatment

Curves that are 20 degrees or less before the time of skeletal maturity are considered mild and gener­ally are re-evaluated every six months. Curves that pro­gress 5 to 10 degrees and those that are more than 30 degrees at the time of diagnosis (considered moderate) usually are treated with a brace, as early and intensive bracing is believed to preclude the need for an opera­tion in most instances. Curves of less than 30 degrees rarely progress after maturity, but larger curves may continue to increase throughout the life of the pa­tient58. Arthrodesis with spinal instrumentation is the treatment of choice for curves of more than 45 degrees in children who are still growing, curves of more than 60 degrees in patients who have reached skeletal matu­rity, and curves that have continued to progress even after treatment with bracing.

Although bracing has long been the mainstay of conservative treatment of scoliosis, its efficacy has not been demonstrated definitively in prospective or ran­domized clinical studies in which it has been compared with other forms of non-operative treatment. Miller et al.38, in 1984, retrospectively compared bracing with ob­

servation for the treatment of mild idiopathic scoliosis; they noted a systematic but non-significant trend in fa­vor of bracing, but the curve failed to progress more than 5 degrees in 80 per cent (203) of the 255 patients in both groups.

Focarile et al.17, in 1991, performed a quantitative analysis of the available evidence regarding the efficacy of non-operative treatment of scoliosis. They found a fivefold increase in the rate of failure, defined as pro­gression of a curve to 45 degrees or the need for an operation, among patients who had a curve of more than 30 degrees at the time of diagnosis compared with those who had a curve of less than 30 degrees. How­ever, they found no difference in progression, defined as worsening of at least 5 degrees, between treated and untreated patients. They concluded that the difference in the rate of failure strongly favored early treatment.

More recently, the lack of scientific evidence regard­ing the efficacy of non-operative treatment has fueled debate regarding the necessity and cost-effectiveness of routine screening for scoliosis. The United States Preventive Services Task Force56, in 1993, stated: "There is inadequate evidence to determine whether brace therapy limits the natural progression of the disease in a significant proportion of cases." Goldberg et al.19, in a 1994 study of the long-term results of scoliosis screen­ing in Dublin, noted: "Since the incidence of signifi­cant scoliosis and of surgery is independent of changes in bracing policy, the efficacy of bracing in causing significant change in natural history must be chal­lenged." Of course, the efficacy of various forms of treatment is just one factor that influences a recom­mendation for or against screening. However, questions remain as to whether non-operative treatment alters the natural history of idiopathic scoliosis; whether it prevents or moderates complications such as substan­tial deformity of the trunk and pulmonary compromise; and, more specifically, whether bracing substantially re­duces the number of curves that will progress suffi­ciently to necessitate operative intervention.

To the best of our knowledge, no previous investi­gators have examined the full range of non-operative treatment options available for large numbers of pa­tients. Accordingly, in 1993, the Prevalence and Natural History Committee of the Scoliosis Research Society decided to compare, with use of meta-analysis, the re­sults of non-operative treatment of idiopathic scoliosis. We chose to include lateral electrical surface stimula­tion, a method that is no longer used, so that the results of this meta-analysis could be compared with those ob­tained by means of clinical observation and conven­tional research methods.

Materials and Methods

Search of the Literature We began our search with the bibliography from

the most recent edition of Campbell's Operative Pedi-

VOL. 79-A, NO. 5, MAY 1997

666 D. E. ROWE ET AL.

5 U -

45-CO

5 40-o 2 35H o •g 30J

CO

2: 25^ o CD 2 0 " O)

i 151 CO

S 10J CO

CL i-5"

Accept

o

.-̂ "̂ o

0

.^ > "

L oo

i i i i i i i i i i i i i i i i i i i i i i i i

1970 1975 1980 1985 1990 Year

FIG.

Reject

1970

l

Q Q i I i i i i I i i i i I

1980 1985 1990 1995 Year

Graph showing the quality scores as a percentage of the perfect score by year. The quality of both the accepted and the rejected studies, as indicated by their quality scores, increased over time. However, many studies lost one-third of the possible points because they were not randomized or controlled, which explains why the over-all percentages (even for some of the accepted studies) were relatively low.

atric Orthopedics^, prepared in part from an on-line search of the literature. We identified thirty-seven peer-reviewed articles on scoliosis1-3'5'7'9-12-16'20'22'24'25'27-29'3032-35'3739'40'43-45,47.53-55.57,60.61.63 t h a t w e r e p ubii shed between the years 1975

and 1993. In addition, we included two unpublished studies": the first, which was awaiting publication at the time of our analysis, was a report on bracing that was prepared for the Scoliosis Research Society in 198242, and the other was a doctoral thesis detailing a retrospec­tive study of 290 consecutive patients who had been managed with bracing because of idiopathic scoliosis52.

Criteria for Selecting Studies

Each study was reviewed by two Committee mem­bers and the Chairman. The reviewers used a data-collection sheet to obtain information on the type of treatment; the numbers of male and female patients; the description of the curve; the level of the apex; the spinal balance (or imbalance); the progression factor33; the duration for which the brace was worn each day; the age of the patient and the magnitude of the curve at the times of diagnosis, bracing, weaning from the brace, and latest follow-up; the initial Risser sign46; and the need for an operation. The reviewers also assessed the quality of each study by evaluating the criteria that were used to select and reject participants, the description of the braces that were used, the design of the study (with regard to randomization, blinding, the inclusion of a control group, the participation of an independent ob­server, and so on), the level of significance, the criteria for weaning, the duration of follow-up, progression fac­tors33, the testing of compliance, and the duplication of end-point variables. The highest possible score was 42 points. The reviewers were permitted to see the names of the authors of each study but were asked to confine their evaluation to the Materials and Methods and Re­

sults sections in an effort to avoid being biased by the authors' interpretations.

The quality of both the accepted and the rejected studies, as reflected by their quality scores, increased over time (Fig. 1). Nevertheless, many studies lost 15 points because they were not randomized or controlled; this explains why the raw scores, even for some of the accepted studies, were relatively low. Evaluation of the accepted studies with use of Pearson correlation coeffi­cients demonstrated a significant relationship between the quality score and the year of publication (r = 0.45, p < 0.05); however, there was no significant relationship between the year of publication and the size of the treatment effect (the proportions of successes and fail­ures) (r = -0.08, p = 0.74) or between the quality score and the size of the treatment effect (r = -0.22, p = 0.33). These Pearson correlation coefficients also were calcu­lated separately for the accepted studies of bracing and the accepted studies of lateral electrical surface stimu­lation; none were found to be significant, with the num­bers available.

Nineteen studies were rejected because they ad­dressed only psychological or social issues related to bracing1-91236'40 or because they contained insufficient data regarding treatment and follow-up or they lacked data regarding the completion of treatment3-52025-27'2930323537'40'47'55-57. The remaining twenty studies (thirteen studies of brac­ing; six studies of lateral electrical surface stimulation; and one study of bracing, lateral electrical surface stim­ulation, and no treatment) contained sufficient data for meta-analysis (Table II).

From those twenty studies, we collected data on 1910 patients who had completed a course of non-operative treatment for idiopathic scoliosis. Over-all, 1459 patients had been managed with bracing; 322, with lateral electrical surface stimulation; and 129 (all

THE JOURNAL OF BONE AND JOINT SURGERY

A META-ANALYSIS OF THE EFFICACY OF NON-OPERATIVE TREATMENTS FOR IDIOPATHIC SCOLIOSIS 667

TABLE II

RAW DATA FOR STUDIES INCLUDED IN META-ANALYSIS

Study, Year of Publication

Bracing (n = 1459) Nachemson et al.42,1995 Hanks et al.24,1988

Federico and Renshaw16,1990 Price et al.45,1990 Willers et al.61,1993 Piazza and Bassett44,1990 Ylikoski et al.63,1989 Park et al.43,1977 Kahanovitz et al.28,1982 Edmonson and Morris14,1977 Green22,1986

Montgomery and Willner39,1989

Styblo52,1991

Emans et al.15,1986

Lateral electrical surface stimulation (n = 322)

Nachemson et al.42,1995 Swank et al.54,1989 Goldberg et al.21,1988 Axelgaard and Brown2,1983 Sullivan et al.53,1986 Durham et al.13,1990 Bylund et al.7,1987

Control group (n = 129) Nachemson et al.42,1995

Rate of Success

(Per cent)

64 81

85 45 88 80

100 72 60

100 57

72

76

81

37 64 35 44 32 50 22

49

No. of Patients

111 100

13 44 25 76

107 43 15 52 44

244

290

295

46 39 23 48

118 30 18

129

Criterion for Failure

(Degrees of Progression)

6 10

6 5

10 5

Unspecified Unspecified Unspecified Unspecified

5

Unspecified

10

5

6 10 10 10 10 5 3

6

Level of Maturity

Immature Immature

Mixed Immature Mixed Immature Mature Mixed Juvenile Mixed Immature

Mixed

Mixed

Mixed

Immature Immature Immature Immature Immature Immature Mixed

Immature

Daily Duration of Brace Wear*

(Hrs.)

16 23

8 8

23 23 23 23 23 23 16

23

23

23

NA NA NA NA ' NA NA NA

NA

Type of Brace*t

Others (TLSO) Others (Wilmington,

TLSO) Charleston Charleston Others (Boston) Others (Wilmington) Others (Boston) Others (Prenyl) Milwaukee Milwaukee Others (Boston and

Milwaukee) Others (Boston and

Milwaukee) Others (Boston and

Milwaukee) Others (Boston)

NA NA NA NA NA NA NA

NA

*NA = not applicable. tTLSO = thoracolumbosacral orthosis.

of whom were included in the only prospective study42), with observation only (Table II).

Study Variables

Inconsistencies among the studies were common. Sometimes, it was impossible to identify the number of patients who had discontinued treatment, as we could not determine how many patients had actually com­pleted treatment or whether all of the patients who had discontinued treatment were considered to have had a failure of treatment. Only the study that was prospec­tive included a control group42. Many studies included insufficient data for meta-analysis of at least one of the relevant variables.

Therefore, we selected three important variables for which sufficient information was available across stud­ies: the type of treatment, the level of maturity, and the criterion used to determine progression of the curve (or failure of treatment). The type of treatment was the most straightforward variable, as all of the patients had been managed with bracing, lateral electrical surface stimulation, or observation. Still, there were numerous variables and missing information among the studies of each type of treatment. For example, five reports on bracing did not specify the criterion that was used to

determine the failure of treatment. For the purpose of analysis, the braces were subdivided into Milwaukee braces, Charleston braces, and all other types of braces (primarily thoracolumbosacral orthoses), and the brac­ing regimens were classified on the basis of whether the brace was worn for eight, sixteen, or twenty-three hours per day.

The chronological age of the patients at the begin­ning and end of treatment was reported in virtually all studies. However, some investigators also used the Risser classifications46 that reflect skeletal maturity, skeletal age, and menarche. On the basis of the vari­ables just cited, we grouped the studies into four cate­gories according to the predominant level of maturity of the patients: juvenile (composed of children who were nine years old or less), immature adolescent (com­posed mostly of children who were ten to thirteen years old and had a Risser sign of 2 or less), mature adoles­cent (composed mostly of children who were more than thirteen years old and had a Risser sign of 3 or 4), and mixed (composed of a mixture of immature and ma­ture adolescent patients, with no clear majority). The reported mean age at the time of menarche (12.5 years6) conveniently coincides with the upper chronological age of patients in the second category, and skeletal ma-

VOL. 79-A, NO. 5, MAY 1997

668 D. E. ROWE ET AL.

turity is unlikely to be reached until a patient is a mature adolescent.

The criterion for failure ranged from 3 to 10 degrees of progression; the five studies in which no criterion was specified were classified as unspecified in the analysis. Progression of the curve was measured with use of the Cobb method10 in all studies. Unfortunately, the range of measurement error associated with this method is +5 degrees4'23'31-34,62, which means that the probability of mis-classifying a successful treatment as a failure (or a failed treatment as a success) may be quite high when only a few degrees of progression is used as the criterion for failure.

Information regarding other variables (for example, the type of curve) was insufficient for analysis.

Statistical Methods

Meta-analysis is a statistical method for pooling and analyzing data from a number of studies. According to Sacks et al.48, the purposes of meta-analysis are to in­crease statistical power, to resolve uncertainty when re­ported results disagree, to improve estimates of effect size, and to answer questions not posed at the start of individual studies.

The number of failures of treatment was determined by calculating the total number of patients in each study who had unacceptable progression of the curve (as de­fined by the criterion in that study), who failed to com­

plete the course of treatment, or who had an operation. The percentage of patients who completed the course of treatment without failure yielded the mean propor­tion of success for a given study. Because the number of failures of treatment included some patients who did not complete treatment but in whom the curve had not progressed, the success rates were conservative.

The weighted means that are calculated with meta­analysis are adjusted for sample size, so that the results from studies of a large sample are given relatively more weight than those from studies of a small sample. The analysis uses inverse variance weightings (W = 1/V), where V is the estimated asymptotic variance of the sample proportion (p [1 - p]/n). In meta-analysis, p is the proportion of successes for a given study, (1 - p) is the proportion of failures in the study, and n is the number of patients in the study50.

We performed several sets of categorical and regres­sion analyses to explore the relationships between out­come (that is, the proportion of successes to failures) and the type of treatment (bracing, lateral electrical surface stimulation, or observation), the level of matu­rity (juvenile, immature adolescent, mature adolescent, or mixed), and the criterion for failure (3, 5, 6, or 10 degrees of progression, or unspecified). Meta-analysis begins with an over-all test of homogeneity, called the Q test26. This test, which is performed by calculating the weighted total sum of the squares for the outcomes,

•BRACING

LESS

CONTROL

Nachemson et al.42

Hanks et al.24

Federico and Renshaw16

Price et al.45

Willers et al.61

Piazza and Bassett44

Ylikoski et al.63

Park et al.43

Kahanovitz et al.28

Edmonson and Morris,4

Green22

Montgomery and Willner39

Styblo52

Emans et al.15

Nachemson et al.42

Swank et al.54

Goldberg etal.21

Axelgaard and Brown2

Sullivan etal.53

Durham etal.13

Bylundetal.7

Nachemson etal.42 "

1 1.2

95% Confidence Intervals for Proportion of Successes

XWB - Weighted Mean Success Proportion for Bracing Group

XWL - Weighted Mean Success Proportion for LESS Group

Xwc - Weighted Mean Success Proportion for Control Group

FIG. 2

Graph showing the unadjusted mean proportions of success, with 95 per cent confidence intervals, for the twenty studies. LESS = lateral electrical surface stimulation.

THE JOURNAL OF BONE AND JOINT SURGERY

A META-ANALYSIS OF THE EFFICACY OF NON-OPERATIVE TREATMENTS FOR IDIOPATHIC SCOLIOSIS 669

indicates whether all of the studies had a similar propor­tion of successes. Typically, the Q test is useful when most of the studies involved in the meta-analysis are not randomized, controlled trials. A significant result on the Q test indicates that the study variables (in this case, the type of treatment, the level of maturity, and the criterion for failure) may cause the proportion of successful out­comes to vary among the studies. Additional analyses are performed to determine the individual contribution of each variable. These analyses produce two test statis­tics, QB and QW, which indicate the predictive power of the variable being analyzed (that is, the degree of variance that it explains) and the degree of variance that is unexplained, respectively50.

When a variable is shown to be a significant contrib­utor, z tests are performed to determine which specific categorical groupings within the variable (for example, which treatment groups) differ from each other. The Bonferroni method26 was used to protect the level of alpha when multiple pairwise comparisons were per­formed on the same set of data.

Results

The weighted mean proportion of success in the studies of bracing was 0.92 (Fig. 2).

Bracing Compared with Lateral Electrical Surface Stimulation and No Treatment

The first set of categorical analyses compared the studies of bracing, as a group, with the studies of lat­eral electrical surface stimulation and the large control group in the study by Nachemson et al.42. The over-all Q value for the categorical analysis was significant (Q = 791, degrees of freedom = 22, p < 0.0001), so individual analyses were performed for each variable.

Significant differences were found among the three types of treatment with regard to progression of the curve (QB = 452, p < 0.0001; QW = 339, p < 0.0001). The weighted mean proportions of success were 0.92 for bracing, 0.39 for lateral electrical surface stimulation, and 0.49 for observation only (Fig. 3). Bracing was sig­nificantly more successful than both lateral electrical surface stimulation and observation only (z = -19.30 and -9.59, respectively; p < 0.0001 for both comparisons). With the numbers available, lateral electrical surface stimulation was not significantly more effective than treatment with observation only.

The level of maturity (juvenile, immature adoles­cent, mature adolescent, or mixed) also had a significant effect on progression of the curve (QB = 491, p < 0.0001; QW = 300, p < 0.0001). Curves generally were less likely to progress as the level of maturity increased (Fig. 4). All pairwise comparisons were significant (p < 0.01) according to the z test, except for the comparisons be­tween the juvenile and immature adolescent groups and between the juvenile and mixed groups, which did not demonstrate significance with the numbers available.

. 1 "

8 0.9H o w 0.8^

o •E 0.6-o Q.

e o.5-Q.

£ °-4-2 0.3-•o SZ 0 . 2 -

1 0.1-* 0

; U

0.49

fljjH ^^^H ^^^H

m m Control LESS

Control or Treatment Group

FIG. 3

0.92

Bracing

Graph showing the weighted mean proportions of success for the control group and for the groups treated with lateral electrical surface stimulation (LESS) and bracing.

The criterion for failure also had a significant effect on the outcome (QB = 505, p < 0.0001; QW = 285, p < 0.0001) (Fig. 5). All pairwise comparisons were signifi­cant (p < 0.01). As it is well known that the precision of estimates of progression made with the Cobb method10

is no better than ±5 degrees4'23,3'3462, we conducted an additional analysis after subdividing the studies into three less specific groups. The first group included stud­ies in which the criterion for failure was 3, 5, or 6 degrees; the second group, those in which it was 10 de­grees; and the third group, those in which it was not specified. There was no difference between the first two groups with regard to the weighted mean propor­tion of success (0.68 and 0.67, respectively). However, the weighted mean proportion of success in the third group (0.97) was significantly higher than those in the

0.99

Juvenile Immature Mixed FIG. 4

Mature

Graph showing the weighted mean proportions of success accord­ing to the level of maturity. Curves generally were less likely to progress as the level of maturity increased.

VOL. 79-A, NO. 5, MAY 1997

670 D. E. ROWE ET AL.

0.97

3 5 6 10 Unspecified Curve Progression (degrees)

FIG. 5

Graph showing the weighted mean proportions of success accord­ing to the criterion for failure.

first two groups (z = -16.61 and -15.82, respectively; p < 0.0001 for both comparisons).

A regression model that included the type of treat­ment, the level of maturity, and the criterion for failure accounted for the highest proportion of variability (F = 16.37, p < 0.0001, r2 = 0.79).

Bracing Only

Because the results of initial analyses supported the contention that lateral electrical surface stimulation is ineffective for the treatment of scoliosis, the studies of that method were eliminated from all subsequent analyses.

For the first set of categorical analyses, the braces were subdivided into Milwaukee braces (two stud­ies1428), Charleston braces (two studies1645), and all other types of braces (ten studies15222439424452"163). In addition to the type of brace, we considered the duration of bracing (eight, sixteen, or twenty-three hours per day), the level of maturity (juvenile, immature adolescent, mature ad­olescent, or mixed), and the criterion for failure (5,6, or 10 degrees of progression, or unspecified). The type of brace and the duration for which it was worn were con­founded in these analyses; the Charleston brace was the only brace worn for eight hours per day, and the Mil­waukee brace was worn for twenty-three hours per day in both of the studies in which it was the only brace used. All other braces were worn for either sixteen or twenty-three hours per day.

The over-all Q value was significant (Q = 320, p < 0.0001), so individual analyses were performed for each variable.

The type of brace had a significant effect on the outcome (QB = 58, p < 0.0001; QW = 262, p < 0.0001), although this effect was small compared with the effects of other variables. The weighted mean proportions of success were 0.99 for the Milwaukee brace, 0.60 for the

Charleston brace, and 0.90 for all other types of braces. The Charleston brace was significantly less successful than the Milwaukee brace and all other types of braces (z = 5.07 and 6.35, respectively; p < 0.0001 for both comparisons), and_ the Milwaukee brace was signifi­cantly more successful than all other types of braces (z = 5.37, p < 0.0001).

The daily duration for which the brace was worn also had a significant effect on the outcome (QB - 90, p < 0.0001; QW = 230, p < 0.0001). The weighted mean proportions of success were 0.93, 0.62, and 0.60 for the twenty-three, sixteen, and eight-hour regimens, respec­tively. The braces that were worn for twenty-three hours per day were significantly more successful than those that were worn for eight or sixteen hours per day (z = -5.55 and -7.82, respectively; p < 0.0001 for both com­parisons). The difference between the success rates of the eight and sixteen-hour regimens was not significant, with the numbers available.

The outcome was significantly influenced by the level of maturity as well (QB = 160, p < 0.0001; QW = 161, p < 0.0001). The weighted mean proportions of success were 0.99, 0.88, 0.71, and 0.60 for the studies of mature adolescent, mixed, immature adolescent, and ju­venile groups, respectively. The studies of mature ado­lescents had a significantly better outcome than those of mixed (z = -7.87, p < 0.0001), immature adolescent (z = -11.36, p < 0.0001), and juvenile groups (z = -3.07, p < 0.01). The studies of mixed groups had a significantly better outcome than the studies of immature adoles­cents (z = -6.95, p < 0.0001). No other pairwise compar­isons were significant, with the numbers available.

The criterion for failure also significantly affected the outcome (QB = 180, p < 0.0001; QW = 140, p < 0.0001). The lowest weighted mean proportion of suc­cess (0.68) was associated with studies in which the criterion for failure was 6 degrees of progression; stud­ies in which the criterion was 5 or 10 degrees had higher proportions of success (0.77 and 0.78, respectively). The highest weighted mean proportion of success (0.97) was associated with the studies in which the criterion was unspecified. We conducted an additional analysis in which the studies that employed the 5 and 6-degree criteria were combined; the weighted mean proportion of success for the combined group was 0.75. The out­come in the combined group was not significantly dif­ferent from that in the group in which the criterion was 10 degrees, but both of these groups had lower mean proportions of success than the group in which the cri­terion was unspecified (z = -11.31 and -8.48, respec­tively; p < 0.0001 for both comparisons).

The best regression model, in terms of r2, was one that included all four variables; this model accounted for 66.7 per cent of the variance (F = 3.20, p < 0.10).

In the next set of categorical analyses, we used a four-way classification in which the braces were subdi­vided into Charleston braces, Milwaukee braces, other

THE JOURNAL OF BONE AND JOINT SURGERY

A META-ANALYSIS OF THE EFFICACY OF NON-OPERATIVE TREATMENTS FOR IDIOPATHIC SCOLIOSIS 671

braces that were worn for twenty-three hours per day, and braces that were worn for sixteen hours per day. The over-all Q value was 320 (degrees of freedom = 13, p < 0.0001). The results for the other three variables that were considered — the daily duration for which the brace was worn, the level of maturity, and the crite­rion for failure — remained as reported in the previous analysis.

This additional analysis also demonstrated that the type of brace had a significant effect on the outcome (QB = 112, p < 0.0001; QW = 208, p < 0.0001). The weighted mean proportions of success were 0.99 for the Milwaukee brace, 0.91 for the other braces that were worn for twenty-three hours per day, 0.62 for the braces that were worn for sixteen hours per day, and 0.60 for the Charleston brace. All pairwise comparisons demon­strated a significant difference (p < 0.0001), except for that between the Charleston brace and the braces that were worn for sixteen hours per day.

The regression analysis was problematic because the type of brace and the daily duration for which the brace was worn were confounded. Unbiased sum of the squares estimates could not be calculated for models that included both of these variables. The best remain­ing models were those that combined three of the four variables: the daily duration for which the brace was worn, the level of maturity, and the criterion for failure (F = 5.50, p < 0.025, r2 = 0.62) and the type of brace, the level of maturity, and the criterion for failure (F = 3.20, p < 0.10, r2 = 0.67).

Bracing Compared with No Treatment

Two final sets of categorical analyses were per­formed to compare bracing with no treatment. In the first set of analyses, the braces were subdivided into Milwaukee braces, Charleston braces, and all other braces. Only the type of brace and the daily duration for which it was worn were considered as variables. The over-all Q value was 412 (degrees of freedom = 14, p < 0.0001). The type of brace had a significant effect on the outcome (QB = 150, degrees of freedom = 3, p < 0.0001; QW = 262, degrees of freedom = 11, p < 0.0001), as did the duration of brace wear (QB = 182, degrees of freedom = 3, p < 0.0001; QW = 230, degrees of free­dom = 11, p < 0.0001).

The outcome associated with the" Milwaukee brace was significantly better than that associated with all other types of braces and that of treatment with obser­vation only (p < 0.0001 for all comparisons). The only non-significant pairwise comparison, with the numbers available, was the one between the Charleston brace and observation only.

The braces that were worn for twenty-three hours per day were significantly more successful than those that were worn for eight or sixteen hours per day and observation only (p < 0.0001 for all comparisons). No other pairwise comparisons were significant, with the

0.99

8 I 0.91 M

§ °9 H •

itllll Control Charleston 16-Hour 23-Hour 23-Hour

Brace TLSO Braces Milwaukee TLSO Brace

FIG. 6 Graph showing the weighted mean proportions of success for the

control condition and various bracing regimens. Braces that were worn for twenty-three hours per day were significantly more effec­tive than all other treatments (p < 0.0001). TLSO = thoracolumbosa-cral orthosis. numbers available. The weighted mean proportions of success for the control, eight-hour, sixteen-hour, and twenty-three-hour regimens were 0.49, 0.60, 0.62, and 0.93, respectively.

In the final set of categorical analyses, the braces were subdivided into Charleston braces, Milwaukee braces, other braces that were worn for twenty-three hours per day, and braces that were worn for sixteen hours per day; these groups then were compared with the control group. Again, the relevant variables were the type of brace and the daily duration for which the brace was worn. The effect of the bracing regimen did not change from the previous analysis. The effect of the type of brace also was significant (QB = 204.100, degrees of freedom = 4, p < 0.0001; QW = 208.275, degrees of freedom = 10, p < 0.0001). Seven often possible pairwise comparisons were significant (p < 0.0001). The weighted mean proportion of success was highest for the Mil­waukee brace (0.99), followed by the other braces that were worn for twenty-three hours per day (0.91) and then by those that were worn for sixteen hours per day (0.62) (Fig. 6). The results associated with the Charles­ton brace (which was worn for eight hours per day) and the braces that were worn for sixteen hours per day did not differ from the results associated with observa­tion only.

Discussion The results of this meta-analysis support the efficacy

of bracing compared with lateral electrical surface stim­ulation and observation only. Bracing for twenty-three hours per day was associated with the highest rates of success. The age of the patient at the start of treatment, the criterion for failure, and the bracing regimen all had effects on the statistical model. We were unable to pair

VOL. 79-A, NO. 5, MAY 1997

672 D. E. ROWE ET AL.

the data so that the type of brace and the daily duration for which the brace was worn were not confounded. In practice, the type of brace and the daily duration for which it is worn cannot be separated completely because it is unreasonable to ask a patient to wear a Charleston brace when not lying down.

Bracing for twenty-three hours per day was signifi­cantly more effective than any other treatment (p < 0.0001). Meta-analysis did not demonstrate a difference between bracing for eight or sixteen hours per day and no treatment at all. The difference between the eight and sixteen-hour regimens was not significant with the numbers available, perhaps because there were insuffi­cient data.

In theory, the rate of success is expected to increase as the number of degrees of progression defining a fail­ure of treatment increases. Our data do not support this theory because the studies in which the criterion for failure was 5 degrees demonstrated a higher rate of success than those in which it was 10 degrees. However, one must consider that, in this analysis, we pooled the data from studies in which different types of treatment were used. As it turns out, four of the seven studies in which the criterion for failure was 10 degrees were stud­ies of lateral electrical surface stimulation. Thus, the lower rate of success associated with the criterion of 10 degrees probably reflects the low rate of success of elec­trical surface stimulation.

It should be noted that we did not conduct a power analysis. Power analysis is extremely complex in meta­analysis. To our knowledge, no power analyses have been conducted for meta-analyses of dichotomous treatment effects measured as proportions of successes or fail­ures. Nevertheless, one power analysis for treatment effects expressed as means8 and another power analysis for treatment effects expressed as correlations51 indi­

cated that the homogeneity (Q) test achieves reason­able levels of power for even a small number of samples.

These data were not adjusted to account for compli­ance of the patient with the prescribed period of brace wear. We can only state that when patients are told to wear the brace longer each day, they have a better chance of preventing progression of the curve. Previous studies have demonstrated rates of non-compliance of 12 to 20 per cent or higher as well as a decrease in partial compliance of 10 to 15 per cent annually during adoles­cence7. It is likely that patients wear the brace for fewer hours than actually prescribed.

The results of the present study run counter to the current trend toward more restrictive bracing for fewer hours each day. The demonstrated relationship between greater compliance and a higher rate of success should be emphasized to the patient during counseling.

Lack of agreement with regard to certain common elements (chronological and skeletal age, the descrip­tion of the curve, and the criterion for failure) unneces­sarily complicated the task of evaluating the efficacy of non-operative treatment of idiopathic scoliosis. There­fore, we strongly recommend the adoption of a standard protocol that involves an assessment of skeletal age; a description of the curve according to the terminology defined by the Scoliosis Research Society49; and the use of progression of more than 10 degrees from the start of treatment, withdrawal from treatment, or the need for an operation as the only criteria for failure. Only patients who have completed the course of treatment, dropped out, or had a failure of treatment should be included in the analysis of the results. Patients still re­ceiving treatment should be excluded42.

NOTE: The authors thank Betsy Jane Becker, Ph.D., Professor of Counseling, Educational Psychology, and Special Education at Michigan State University, East Lansing, Michigan, for guidance in applying the principles of meta-analysis to these data.

References 1. Apter, A.; Morein, G.; Munitz, H.; Tyano, S.; Maoz, B.; and Wijsenbeek, H.: The psychosocial sequelae of the Milwaukee brace in

adolescent girls. Clin. Orthop., 131:156-159,1978. 2. Axelgaard, J., and Brown, J. C: Lateral electrical surface stimulation for the treatment of progressive idiopathic scoliosis. Spine, 8:

242-260,1983. 3. Bassett, G. S., and Bunnell, W. P.: Effect of a thoracolumbosacral orthosis on lateral trunk shift in idiopathic scoliosis. J. Pediat. Orthop.,

6:182-185,1986. 4. Beekman, C. E., and Hall, V.: Variability of scoliosis measurement from spinal roentgenograms. Phys. Ther., 59:764-765,1979. 5. Bradford, D. S.; Tanguy, A.; and Vanselow, J.: Surface electrical stimulation in the treatment of idiopathic scoliosis: preliminary results in

30 patients. Spine, 8:757-764,1983. 6. Bunnell, W. P.: The natural history of idiopathic scoliosis before skeletal maturity. Spine, 11:773-776,1986. 7. Bylund, P.; Aaro, S.; Gottfries, B.; and Jansson, E.: Is lateral electric surface stimulation an effective treatment for scoliosis? J. Pediat.

Orthop., 7:298-300,1987. 8. Chang, L.: A power analysis of the test of homogeneity in effect-size meta-analysis. Doctoral dissertation, Department of Counseling,

Educatidnal Psychology, and Special Education, Michigan State University, East Lansing, Michigan, 1992. 9. Clayson, D.; Luz-Alterman, S.; Cataletto, M. M.; and Levine, D. B.: Long-term psychological sequelae of surgically versus nonsurgically

treated scoliosis. Spine, 12: 983-986,1987. 10. Cobb, J. R.: Outline for the study of scoliosis. In Instructional Course Lectures, The American Academy of Orthopaedic Surgeons. Vol. 5,

pp. 261-275. Ann Arbor, J. W. Edwards, 1948. 11. Cook, D. J.; Guyatt, G. H.; Ryan, G.; Clifton, J.; Buckingham, L.; Willan, A.; Mcllroy, W.; and Oxman, A. D.: Should unpublished data be

included in meta-analyses? Current convictions and controversies. /. Am. Med. Assn., 269: 2749-2753,1993. 12. DiRaimondo, C. V., and Green, N. E.: Brace-wear compliance in patients with adolescent idiopathic scoliosis. / Pediat. Orthop., 8:

143-146,1988.

THE JOURNAL OF BONE AND JOINT SURGERY

A META-ANALYSIS OF THE EFFICACY OF NON-OPERATIVE TREATMENTS FOR IDIOPATHIC SCOLIOSIS 673

13. Durham, J. W.; Moskowitz, A.; and Whitney, J.: Surface electrical stimulation versus brace in treatment of idiopathic scoliosis. Spine, 15: 888-892,1990.

14. Edmonson, A. S., and Morris, J. T.: Follow-up study of Milwaukee brace treatment in patients with idiopathic scoliosis. Clin. Orthop., 126:58-61,1977.

15. Emans, J. B.; Kaelin, A.; Bancel, P.; Hall, J. E.; and Miller, M. E.: The Boston bracing system for idiopathic scoliosis. Follow-up results in 295 patients. Spine, 11: 792-801,1986.

16. Federico, D. J., and Renshaw, T. S.: Results of treatment of idiopathic scoliosis with the Charleston bending orthosis. Spine, 15: 886-887,1990.

17. Focarile, F. A.; Bonaldi, A.; Giarolo, M. A.; Ferrari, U.; Zilioli, E.; and Ottaviani, C: Effectiveness of nonsurgical treatment for idiopathic scoliosis. Overview of available evidence. Spine, 16: 395-401,1991.

18. Freeman, B. L.: Non-operative management of idiopathic scoliosis. In Campbell's Operative Pediatric Orthopedics, edited by S. T. Canale and J. H. Beaty. Ed. 2, pp. 640-642. St. Louis, Mosby-Year Book, 1995.

19. Goldberg, C. J.; Dowling, F. E.; Fogarty, E. E.; and Moore, D.: School scoliosis screening and the U.S. Preventive Services Task Force. An examination of long-term results. Orthop. Trans., 19: 590-591,1995-1996.

20. Goldberg, C. J.; Dowling, F. E.; Hall, J. E.; and Emans, J. B.: A statistical comparison between natural history of idiopathic scoliosis and brace treatment in skeletally immature adolescent girls. Spine, 18: 902-908,1993.

21. Goldberg, C; Dowling, F. E.; Fogarty, E. E.; Regan, B. F.; and Blake, N. S.: Electro-spinal stimulation in children with adolescent and juvenile scoliosis. Spine, 13: 482-484,1988.

22. Green, N. E.: Part-time bracing of adolescent idiopathic scoliosis. /. Bone and Joint Surg., 68-A: 738-742, June 1986. 23. Gross, C; Gross, M.; and Kuschner, S.: Error analysis of scoliosis curvature measurement. Bull. Hosp. Joint Dis., 43:171-177,1983. 24. Hanks, G. A.; Zimmer, B.; and Nogi, J.: TLSO treatment of idiopathic scoliosis. An analysis of the Wilmington jacket. Spine, 13:

626-629,1988. 25. Hassan, I., and Bjerkreim, I.: Progression in idiopathic scoliosis after conservative treatment. Acta Orthop. Scandinavica, 54: 88-90,1983. 26. Hedges, L.V.: Fixed effects models. In The Handbook of Research Synthesis, pp. 285-299. Edited by H. Cooper and L. V. Hedges. New

York, Russell Sage Foundation, 1994. 27. Jonasson-Rajala, E.; Josefsson, E.; Lundberg, B.; and Nilsson, H.: Boston thoracic brace in the treatment of idiopathic scoliosis. Initial

correction. Clin. Orthop., 183:37-41,1984. 28. Kahanovitz, N.; Levine, D. B.; and Lardone, J.: The part-time Milwaukee brace treatment of juvenile idiopathic scoliosis. Long-term

follow-up. Clin. Orthop., 167:145-151,1982. 29. Karol, L. A.; Johnston, C. E., II; Browne, R. H.; and Madison, M.: Progression of the curve in boys who have idiopathic scoliosis. / Bone

and Joint Surg., 75-A: 1804-1810, Dec. 1993. 30. Keiser, R. P., and Shufflebarger, H. L.: The Milwaukee brace in idiopathic scoliosis. Evaluation of 123 completed cases. Clin. Orthop.,

118:19-24,1976. 31. Kitteson, A. C, and Lim, L. W.: Measurement of scoliosis. Am. J. Roentgenol, 108: 775-777,1970. 32. Laurnen, E. L.; Tupper, J. W.; and Mullen, M. P.: The Boston brace in thoracic scoliosis. A preliminary report. Spine, 8: 388-395,1983. 33. Lonstein, J. E., and Carlson, J. M.: The prediction of curve progression in untreated idiopathic scoliosis during growth. J. Bone and Joint

Surg., 66-A: 1061-1071, Sept. 1984. 34. McAlister, W. H., and Shackelford, G. D.: Measurement of spinal curvatures. Radiol. Clin. North America, 13:113-121,1975. 35. McCollough, N. C, III; Schultz, M.; Javech, N.; and Latta, L.: Miami TLSO in the management of scoliosis: preliminary results in 100

cases./. Pediat. Orthop., 1:141-152,1981. 36. MacLean, W. E., Jr.; Green, N. E.; Pierre, C. B.; and Ray, D. C: Stress and coping with scoliosis: psychological effects on adolescents and

their families./ Pediat. Orthop., 9: 257-261,1989. 37. Mellencamp, D. D.; Blount, W. P.; and Anderson, A. J.: Milwaukee brace treatment of idiopathic scoliosis. Late results. Clin. Orthop.,

126:47-57,1977. 38. Miller, J. A.; Nachemson, A. L.; and Schultz, A. B.: Effectiveness of braces in mild idiopathic scoliosis. Spine, 9:632-635,1984. 39. Montgomery, F., and Willner, S.: Prognosis of brace-treated scoliosis. Comparison of the Boston and Milwaukee methods in 244 girls.

Acta Orthop. Scandinavica, 60: 383-385,1989. 40. Montgomery, F.; Willner, S.; and Appelgren, G.: Long-term follow-up of patients with adolescent idiopathic scoliosis treated conserva­

tively: an analysis of the clinical value of progression. /. Pediat. Orthop., 10: 48-52,1990. 41. Nachemson, A.; Lonstein, J. E.; and Weinstein, S. L.: Prevalence and Natural History Committee report. Read at the Annual Meeting of

the Scoliosis Research Society, Denver, Colorado, Sept. 25,1982. 42. Nachemson, A. L.; Peterson; L.-E.; and members of the Brace Study Group of the Scoliosis Research Society: Effectiveness of

treatment with a brace in girls who have adolescent idiopathic scoliosis. A prospective, controlled study based on data from the Brace Study of the Scoliosis Research Society. /. Bone and Joint Surg., 11-A: 815-822, June 1995.

43. Park, J.; Houtkin, S.; Grossman, J.; and Levine, D. B.: A modified brace (Prenyl) for scoliosis. Clin. Orthop., 126:67-73,1977. 44. Piazza, M. R., and Bassett, G. S.: Curve progression after treatment with the Wilmington brace for idiopathic scoliosis. /. Pediat. Orthop.,

10:39-43,1990. 45. Price, C. T.; Scott, D. S.; Reed, F. E., Jr.; and Riddick, M. E: Nighttime bracing for adolescent idiopathic-scoliosis with the Charleston

bending brace. Preliminary report. Spine, 15:1294-1299,1990. 46. Risser, J. C: The iliac apophysis: an invaluable sign in the management of scoliosis. Clin. Orthop., 11:111-119,1958. 47. Rudicel, S., and Renshaw, T. S.: The effect of the Milwaukee brace on spinal decompensation in idiopathic scoliosis. Spine, 8: 385-387,1983. 48. Sacks, H. S.; Berrier, J.; Reitman, D.; Ancona-Berk, V. A.; and Chalmers, T. C: Meta-analyses of randomized controlled trials. New

England J. Med., 316: 450-455,1987. 49. Scoliosis Research Society Terminology Committee: A glossary of scoliosis terms. Spine, 1: 57-58,1976. 50. Shadish, W. R., and Haddock, C. K.: Combining estimates of effect size. The Handbook of Research Synthesis, pp. 261-281. Edited by

H. Cooper and L. V. Hedges. New York, Russell Sage Foundation, 1994. 51. Spectors, P. E., and Levine, E. L.: Meta-analysis for integrating study outcomes: a Monte Carlo study of its susceptibility to type I and

type II errors. / Appl. Psychol, 72:3-9,1987.

VOL. 79-A, NO. 5, MAY 1997.

674 D. E. ROWE ET AL.

52. Styblo, K.: Conservative treatment of juvenile and adolescent idiopathic scoliosis. Unpublished doctoral thesis, Rijksuniversiteit te Leiden, Leiden, The Netherlands, 1991.

53. Sullivan, J. A.; Davidson, R.; Renshaw, T. S.; Emans, J. B.; Johnston, C; and Sussman, M.: Further evaluation of the Scolitron treatment of idiopathic adolescent scoliosis. Spine, 11: 903-906,1986.

54. Swank, S. M.; Brown, J. C; Jennings, M. V.; and Conradi, C: Lateral electrical surface stimulation in idiopathic scoliosis. Experience in two private practices. Spine, 14:1293-1295,1989.

55. Uden, A., and Willner, S.: The effect of lumbar flexion and Boston thoracic brace on the curves in idiopathic scoliosis. Spine, 8: 846-850,1983.

56. United States Preventive Services Task Force: Screening for adolescent idiopathic scoliosis. Review article. J. Am. Med. Assn., 269: 2667-2672,1993.

57. Watts, H. G.; Hall, J. E.; and Stanish, W.: The Boston brace system for the treatment of low thoracic and lumbar scoliosis by the use of a girdle without superstructure. Clin. Orthop., 126: 87-92,1977.

58. Weinstein, S. L.: Editorial. Advances in the diagnosis and management of adolescent idiopathic scoliosis. J. Pediat. Orthop., 14: 561-563,1994.

59. Weinstein, S. L.; Zavala, D. C; and Ponseti, I. V.: Idiopathic scoliosis. Long-term follow-up and prognosis in untreated patients. /. Bone and Joint Surg., 63-A: 702-712, June 1981.

60. Wickers, F. C; Bunch, W. H.; and Barnett, P. M.: Psychological factors in failure to wear the Milwaukee brace for treatment of idiopathic scoliosis. Clin. Orthop., 126:62-66,1977.

61. Willers, U.; Normelli, H.; Aaro, S.; Svensson, O.; and Hedlund, R.: Long-term results of Boston brace treatment on vertebral rotation in idiopathic scoliosis. Spine, 18:432-435,1993.

62. Wilson, M. S.; Stockwell, J.; and Leedy, M. G.: Measurement of scoliosis by orthopedic surgeons and radiologists. Aviat. Space and Environ. Med., 54:69-71,1983.

63. Ylikoski, M.; Peltonen, J.; and Poussa, M.: Biological factors and predictability of bracing in adolescent idiopathic scoliosis. J. Pediat. Orthop, 9:680-683,1989.

THE JOURNAL OF BONE AND JOINT SURGERY