Efficacy of comprehensive rehabilitation programs and back school for patients with low back pain: a metanalysis.

Various rehabilitation programs have been developed to treat low back pain in addition to the impairments and disabilities that are the sequelae of low back pain.[1-14] The treatment of low back pain frequently involves some type of "back school." Most back schools include a didactic component in which basic anatomy, function of the spine, and proper lifting techniques are taught. Most schools also provide exercise training (see Schlapbach[15] for a review). The use of multiple cointerventions, differences in program format (eg, inpatient versus outpatient), and the use of back school for patients with different levels of chronicity, however, complicate the interpretation of efficacy across clinical trials.

In terms of cointerventions, back schools can be categorized as primary treatments (limited or no cointervention)[16-23] or as part of a comprehensive rehabilitation program that includes work-site visits, general physical conditioning, work hardening, or operant conditioning.[1-11,24-32] A cursory review of prospective randomized controlled trials revealed that inpatients were sometimes studied,[28,30] whereas other studies included only outpatients[5,6,16-23,29,31] and multiple studies addressed both inpatient and outpatient programs.[24-27,32] In addition, back school has been evaluated for patients with acute low back pain (ie, usually pain of less than 4 weeks' duration),[16,22,31] subacute symptoms (ie, usually pain of 4 to 8 weeks' duration),[5,6,29] and chronic low back pain (ie, usually pain of more than 8 weeks' duration).[17,18,20,23-27,30,32] In some cases, a mix of patients with acute and chronic low back pain were studied as a "homogeneous" group.[19,21,28] The conclusions from these studies are equivocal and contradictory. Because of differences in medical care and insurance systems among nations, the studies' country of origin might also have influenced outcome.[33,34] Using traditional reviews of the literature, the variations in study methods and location make it difficult to determine the efficacy of back schools.

Several reviews attempted to evaluate the efficacy of back schools for reducing pain and disability.[15,35,36] Traditional reviews, however, may not be adequate for systematic investigations of the effect of variation in study methods on study results. Quantitative reviews (meta-analyses) have been advocated as an alternative to traditional reviews.[37-40] This technique of secondary analysis integrates quantitative evidence across clinical trials and requires the same level of methodological rigor demanded of primary research.[37,38,41] In addition, because traditional reviews may be affected by reviewer bias, some authors advocate quantitative reviews to obtain more objective assessments of treatment efficacy.[37,38,41,42]

A quantitative review was done in my study to facilitate the interpretation of prospective randomized controlled trials that used back schools to treat low back pain. The purpose was to conduct a meta-analysis to (1) synthesize existing evidence on the efficacy of back schools from a global perspective and (2) describe the efficacy of selected categories of back schools that were identified a priori. The global analysis addressed two questions:

1. Are back schools affective in reducing pain, physical impairment, and disability?

2. Are there long-term effects associated with back schools?

The a priori, category-specific analysis addressed the following questions:

1. Are comprehensive rehabilitation programs that include back school more effective than programs offering back school as the primary intervention?

2. Are inpatient programs more effective than outpatient programs?

3. Does chronicity of the patient population affect outcome?

4. Do outcomes vary with the study's country of origin?

Method

Potentially relevant studies were obtained through a computer-assisted search of the Index Medicus database for 1976 to 1994. The key words used initially in the Index Medicus search were "low back pain," "back school," and "exercise." To focus the search, "back school" and "exercise" were linked with the key word "low back pain." The authors' names from potentially relevant studies were also entered as key words to obtain additional titles. Other references were obtained through examination of the citations listed in the retrieved studies (citation tracking). The entire search yielded 59 nonoverlapping research report titles that I broadly construed to address the use of back schools in the treatment of low back pain.

Selection Criteria

In order for a potentially relevant study to be included in this metaanalysis, the study had to meet all of the following criteria:

1. The study had to address the use of a back school to treat patients with low back pain. The minimum essential elements in determining whether an intervention included a "back school" were the presence of patient education for proper bending and lifting activities (body mechanics instruction) and the implementation of a passive or active back exercise program. The simple distribution of pamphlets that described postural and ergonomic techniques (without instruction from a therapist) was not construed to represent a back school approach. The descriptions of back schools were variable, and acceptable studies used different treatment formats. For example, Harkapaa et al[26] studied both an inpatient program lasting 3 weeks and an outpatient group that received 15 sessions over a 2-month period. Outcomes were compared with those of a control group that received no treatment. Stankovic and Johnell,[16] in contrast, implemented a back school with 1 session and compared outcomes with those of a group that received exercise. Studies such as these[16,26] were accepted for analysis, but were categorized as either a comprehensive rehabilitation program[26] or a primary back school intervention.[16] The studies were also classified by the type of control or comparison intervention.

The criteria used for making these distinctions are described below. Studies dealing entirely with the use of back schools with asymptomatic subjects to address the efficacy of back pain prevention[43-46] or exclusively with the psychological domain[47] were not included.

2. The study had to compare patients who received back school with subjects in a control or comparison group who did not receive back school. In addition, an explicit statement of random assignment of subjects to the treatment group and to a control or comparison group was necessary. Studies that used a control group, but without random assignment in a prospective manner, were not included in the metaanalysis. The rationale for this decision was based on previous work that described the potential bias of patient selection when the randomization process was not used.[42,48,49]

3. It was necessary that the studies judged patient performance on outcome variables that involved the assessment of pain, physical impairment, functional disability, or work or vocational outcomes. Studies were also included if an assessment of a change in the patient's knowledge or behavior related to back care (an education/compliance factor) was reported. For example, Moffett et al[20] measured the patients' knowledge of back care by using a 20-item questionnaire. Harkapaa et al[24] rated the quality of exercise performance to determine the degree of self-care accomplishment as well as the degree of compliance on a graduated numerical scale at each follow-up examination. Lindstrom et al[5] measured the mean duration of sick leave (a work/vocational outcome) during a 3-year period following a comprehensive program that included back school. Examples of the types of measures that were acceptable for each category of outcome are outlined in Table 1.[50-53]

[TABULAR DATA 1 OMITTED]

4. The study had to report statistical results in sufficient detail to compute or estimate effect sizes. The calculation of effect size is described below.

Forty studies did not meet the criteria for inclusion in the meta-analysis and were eliminated from the review. These studies are listed in the Appendix to conform with the guidelines for meta-analysis format established by Sacks et al.[38] The 19 studies that met the criteria for analysis are outlined in Table 2.[5,6,16-32] In most of these studies, there were multiple comparisons among various conditions. Berwick et al,[19] for example, randomly assigned patients to three groups: usual care (UC), back school (BS), and back school with encouragement for compliance (BSC). Outcomes reflecting the level of pain and function were measured at four different follow-up sessions spanning 18 months after back school enrollment. Statistical hypotheses for the meta-analysis involved a reduction of the design to relevant pair-wise comparisons (ie, UC versus BS and UC versus BSC) for each follow-up session. Because most studies contained multiple comparisons between back school groups and control or comparison groups, there were a total of 206 statistical hypothesis tests for all studies included in the analysis.

[TABULAR DATA 2 OMITTED]

Calculating Effect Size

I abstracted all data from each study accepted for analysis. It was not possible to do the abstraction in a blind fashion. The effect size index--the d-index described by Cohen[54]--was used as a measure of the efficacy of back schools. This method has previously been used to evaluate the efficacy of other nonsurgical treatments for low back pain.[42] The d-index gauges the difference between two groups in terms of the control group standard deviation.[37] Consider a hypothetical example where the experimental group had a mean visual analogue pain score of 2 cm and the control group had a mean score of 1.5 cm. The effect size will be determined by the difference between the means of each group as well as the control group standard deviation. If the control group standard deviation was 1 cm, the d-index would be 0.5 [2-1.5]/1). If the control group standard deviation was 2 cm, then the d-index would be 0.25 ([2-1-5]/2). The control group standard deviation was used because it creates equivalent effect sizes for equal means when more than one treatment condition is presented.[37] If d=0.30, then 3/10 of a standard deviation separates the performance of the average score in the two groups. A positive d-index indicates in this study that the back school group performed better than the control or comparison group. A negative d-index, in contrast, shows that the control or comparison group had a superior outcome compared with the group receiving back school. The d-index, therefore, transforms the results of any two-group comparison into a standardized metric so that findings from a number of studies can be analyzed simultaneously.

Effect sizes were computed from means and standard deviations for the experimental groups and control or comparison groups,[54] or from t and F ratios when means and standard deviations were not reported in the article.[37,55] When t and F ratios were not reported, they were estimated from the significance level and sample size.[37] In addition, nonparametric statistics and percentages were converted to effect sizes using the formulas and rationale provided by Friedman[55] and Glass et al.[37] Hedges[56] noted that d-indexes may be biased with sample sizes less than 50, so the correction factors published by Hedges[56] were used to adjust for potentially inflated effect sizes. The magnitude of the correction factor depends on the degrees of freedom of the control group and varies from 0. 5642 (df=2) to 0.9849 (df=50). Five of 19 studies had control or comparison groups containing fewer than 50 subjects (Tab. 2). The Hedge's correction factor that corresponded with the control group degrees of freedom was multiplied by the d-index to arrive at a "corrected," unbiased effect size. When a study reported a nonsignificant result without enough information to determine the effect size, a d-index of 0.0 was assumed. This method provided a conservative estimate of the true effect size. The [U.sub.3] described by Cohen[54] was used as a measure of the distribution overlap. The [U.sub.3] indicates the percentage of scores in the control or comparison group that were exceeded by the mean score in the back school group. For example, a d-index of 0.40 has an associated U3 value of 65.5%. The translation from d to U3 essentially treats the d-index as a value that is normally distributed. In this hypothetical example, the mean score of subjects receiving back school is better than 65.5% of the scores in the control or comparison group (ie, d=0.40; equivalent to z=0.40 with a cumulative probability of 0.655). A d-index of -O.40), in contrast, indicates that the mean score of subjects in the back school group is better than only 34.5% of the scores in the control or comparison group. In this case, [U.sub.3]=34.5%.

Data Analysis

Global assessment of back school efficacy. The d-index was described, in aggregate, for all hypotheses. An analysis of variance (ANOVA) was used to evaluate the difference between effect sizes. A fixed-effects factorial design with one or more grouping factors was selected to evaluate the long-term effects of back school (less than and greater than 1 year following treatment), cointerventions, type of back school program, chronicity, and study country of origin. In addition, the fixed-effects model was used to evaluate the effect size across the types of outcome measures (Tab. 1) using the d-index as the dependent variable. This analysis essentially considered the unweighted average of the results for each type of d-index (ie, pain, disability, and so on). All ANOVAs comparing more than two means were evaluated in a post hoc fashion using Tukey's studentized range test. All analyses were done using BMDP statistical software.(*)

Efficacy of selected categories of back school. The data were stratified according to the extent of cointerventions (comprehensive rehabilitation with multiple cointerventions versus back school as the primary intervention), the type of program (inpatient versus outpatient), chronicity (acute/ subacute, chronic, mixed), and the study's country of origin.

Comprehensive rehabilitation programs were defined as interventions that coupled the minimal essential elements of back school (stated earlier) with a work-site visit, operant conditioning, cognitive-behavioral group therapy, or an intensive physical training regimen that supplemented traditional back exercise programs. When the d-indexes were stratified according to program type, inpatient programs were defined as those that admitted patients to a controlled residential environment at least 5 days per week (eg, hospital or dorm). Outpatient programs were defined as any back school that did not admit patients to a residence (regardless of the number of sessions).

When chronicity was used to stratify the data, patients with symptoms of 8 weeks or less of duration were defined as being in the acute/subacute stage and those with Symptoms of longer than 8 weeks' duration were defined as being in the chronic phase. Some studies included patients with both acute and chronic symptoms in the same group and, therefore, these populations were categorized as having mixed" chronicity (Tab. 2).

The d-indexes were also evaluated by the study's country of origin. This was done because the medical care and insurance systems of different countries could potentially influence how back pain is reported, treated, and compensated in the case of work-related injury.[33,34] The grouping for the study's country of origin consisted of Scandinavia/the Netherlands and "other." The "other" category included studies from the United States, Canada, England, and Israel as a group, because there was only one study from each of these countries.

A "0" effect size was assigned to a nonsignificant result only when the d-index could not be calculated. When study results were not statistically significant, the effect size may still be different from "0." The method of assigning a "0" d-index to some of these hypothesis tests will underestimate the true effect size. The assignment of a "0" d-index in cases where the effect size could not be directly calculated for nonsignificant results was, therefore, evaluated by including and then excluding the hypotheses with a d-index of 0.

Analysis of Heterogeneity

There is considerable variation in the treatment procedures used in the back schools and in the methods used to assess the outcome of back school or comprehensive programs.[15] Differences between studies can be due to experimental error (within-study variability) and to differences in the populations studied (between-study variability).[38] Levene's F test for homogeneity of variance, therefore, was conducted for each stratification variable.[57] This test is an analysis of variance of the absolute deviations of each effect size from its cell mean. All probability values were two-sided to test the inequality of variance in either direction. When the Levene F test is significant (P<.05), the effect sizes from one group are likely to have greater variance compared with those from the other group. If Levene's F test was statistically significant, it was likely that the d-indexes came from different populations of patients with low back pain. in this case, combining effect sizes (ie, ignoring or averaging across the stratification variable) should be done with caution.

Power Analysis

A power analysis was done using Solo Power Analysis software.([dagger]) The ability to detect an effect size difference of 0.10 was evaluated with different sample sizes and standard errors (Tab. 3). When the sample size is 60 (hypothesis tests), for example, an effect size difference of 0.10 can be detected with 84% probability at an alpha level of .05, assuming a standard error of 0.5.

Table 3. Power to Detect an Effect-Size Difference of 0.10 With Standard Errors (SEs) of 0.5, 0.75, and 1.00 at an Alpha Level of .05

      SE     SE      SE
 N    0.5    0.75    1.00

 20   0.33   0.16   0.11
 40   0.65   0.30   0.18
 60   0.84   0.45   0.26
 80   0.94   0.59   0.34
100   0.98   0.71   0.43
200   0.99   0.99   0.99

Analysis of Independent Effect Sizes

The analyses used in this study rest on the assumption that the outcome measures are independent. Multiple within-study d-indexes, however, are not independent. The potential influence of "correlated" data, therefore, was tested by selecting pain and disability outcomes and averaging multiple within-study d-indexes so that each study assessing pain or disability would be analyzed with only one mean d-index for each outcome category.

Results

Global Assessment of Back School Efficacy

In the studies that were included in the analysis, there were 2,373 patients. Back school was part of a comprehensive rehabilitation program or the primary intervention for 1,279 patients, whereas 1,094 patients were assigned to control or comparison groups. The ages of the subjects varied from 18 to 61 years, with a mean age of 41.8 years (SD = 5.5). Gender was reported for 1,913 patients (55% male, 45% female). Fifty-three percent of the patients (n=1,263) had chronic low back pain, whereas 27% (n=630) had acute or subacute symptoms. Twenty percent (n=480) were patients grouped with mixed chronicity. Fifty-seven percent (n=1,357) attended comprehensive rehabilitation programs, and 24% (n=575) were treated as inpatients.

The d-indexes computed for 206 hypothesis tests are shown in Table 4. Each d-index is indicated by a combination "stem" and "leaf". computed by BMDP statistical software[58] from a model developed by Tukey.[59] For example, the numbers 1, 2, 2, and 6 to the right of -2 in Table 4 indicate that one d-index was equal to -2.1, two d-indexes were equal to -2.2, and one d-index was equal to -2.6. The stem and leaf provides all the information of a histogram, but also indicates the actual values of all d-indexes. Negative values indicate that the control or comparison group actually performed better than the group receiving back school. At the foot of Table 4, the minimum and maximum d-indexes are shown along with the first and third quartiles.

[TABULAR DATA 4 OMITTED]

The mean overall d-index was 0.07 (SD=0.73), with a U3 value of 52.8%. This means that the average performance of subjects receiving back school was better than only 52.8% of the subjects in the control or comparison groups. The average effect size did not change as a function of the duration of follow-up (Tab. 5). The difference between back school groups and control or comparison groups within a year of the intervention d=0.01, [U.sub.3]=50.4%) was not different from the average d-index beyond 1 year from entry into the program (d=0.13, [U.sub.3]=55.2%; F=1.54; df=1,204; P=.22). In addition, there was no relationship between the follow-up duration in weeks and the magnitude of the d-index (Spearman r=.07).

These results imply that back school, in general, yields no benefit beyond that found for control or comparison interventions. Stratification of d-indexes by outcome variable, however, provided a more precise profile of the areas where back school may (and may not) be efficacious (Tab. 5). There were improvements in educational/compliance and strength/endurance outcomes compared with pain, spinal mobility, disability, and work/ vocational outcomes (F=2.23; df=5,200; P=.053). The largest mean effect size was found for improvements in muscle strength (ie, force production) and endurance (d=0.40, [U.sub.3]=65.5%). The smallest mean effect sizes were for work/vocational outcomes d=-0.02, [U.sub.3]=49.2%), pain outcomes (d=-0.08, [U.sub.3=46.8%), and spinal mobility (d=0.0, [U.sub.3]=50%).

Efficacy of Selected Categories of Back School

The number of cointerventions and the type of program clearly had an influence on outcome. Program that incorporated back school with a comprehensive rehabilitation program had a mean d-index of 0.28 and U3 of 61% (Fig. 1 and Tab. 6), whereas those that used back school as a primary intervention had a mean effect size of -0.14 ([U.sub.3]=44.4%). The difference between comprehensive rehabilitation programs and those providing back school as a primary intervention was significant and in favor of comprehensive rehabilitation (F=18.46; df=1,204; P=.000). Comprehensive rehabilitation programs that included back school were more effective at reducing pain, increasing spinal motion, and increasing muscle strength and endurance compared with programs that focused on back school as the primary intervention (Fig. 2 and Tab. 7). The comparison of comprehensive rehabilitation programs with primary back school intervention yielded no significant differences in the average effect size for disability, work/vocational, or education/compliance outcomes (Fig. 2 and Tab. 7).

The average d-index for inpatient programs (d=0.32, U3=62%; Tab. 6) was greater than the average effect size for outpatient programs (d=0.01, [U.sub.3]=50.4%; F=6.08; df=1,204; P=.015). All inpatient program (n=40 hypothesis tests) and 38% of the outpatient programs (n=63 hypothesis tests), however, involved comprehensive rehabilitation. When inpatient programs were compared with outpatient programs provided with comprehensive rehabilitation, there was no difference (mean [d.sub.comp rehab-inpatient]=0.32, SD=0.36; mean [d.sub.comp rehab-outpatient=0.26, SD=0.31; F=1.01;] df=1,101; P=.32).

Chronicity did not appear to play a role in the magnitude of the effect size (Tab. 6). When the average d-index of patients with acute/subacute symptoms was compared with that of patients with chronic symptoms and with that of groups of patients with mixed chronicity, there was no difference among the groups (F=1.18; df=2,203; P=.31). The study by Gilbert et al[21] accepted patients who were free of back pain 30 days prior to the current episode, but a reference to the sample in the article suggested that the patients studied had mixed chronicity. The d-indexes from this study[21] were initially coded to indicate mixed chronicity. The analysis was repeated, however, with the d-indexes for Gilbert et al[21] recoded for acute symptoms, and the findings did not change F=0.73; df=2,203; P=.48).

The average d-index for studies conducted in Scandinavian countries and in the Netherlands (Tab. 6) showed better outcomes compared with studies done in the United States, England, Canada, and Israel ( F=11.68; df=1,204; P=.0008). All studies done outside of Scandinavian countries and the Netherlands involved the use of back school only as a primary intervention and only on an outpatient basis. Nearly all studies done in Scandinavian countries and in the Netherlands, in contrast, provided back school as a comprehensive program with multiple interventions (Tab. 2).

A d-index of 0 was calculated for 85 hypotheses. When d-indexes equal to 0 were excluded from the analysis, the overall mean d-index rose to 0.43 (median=0.29, mode=0.21, n=121). These values reflect the influence of back school only in studies that demonstrated differences between treatment and control groups. When differences were found, the mean d-index (0.43) was positive, indicating that back school or comprehensive programs that included back school had better outcomes compared with the control or comparison groups.

Heterogeneity

The test of heterogencity of effect-size variance was significant for at least one outcome measure (Tab. 8) when cointerventions (comprehensive versus primary back school) and the study's country of origin (Scandinavia/the Netherlands versus "other") were evaluated. There were no differences in the variance per effect size for any outcome variable when chronicity, follow-up duration, or patient type was analyzed.

Analysis of Independent Effect Sizes

When multiple within-study d-indexes were averaged to produce one d-index per study for disability and pain outcomes, the results did not differ from the analysis involving nonindependent data. With regard to a reduction of pain, the analysis of averaged within-study d-indexes showed that comprehensive rehabilitation programs had significantly better results than primary back school programs (comprehensive [program.sub.mean] =0.33, SD=0.29; primary [program.sub.mean] =- 2,a.=-o.16, SD=0.39; F=5.89; df=1,11; P=.034). The results from noncorrelated data were similar to the findings based on multiple observations from individual studies illustrated in Table 7. The nonsignificant difference between the effect sizes for acute versus chronic populations (Tab. 6) was also replicated using noncorrelated pain scores ([acute.sub.mean] =0.05, SD=0.52; [chronic.sub.mean]=0.07, SD=0.32; mixed [chronic.sub.mean] =-0.07, SD=0.59; F=0.10; df=2,10; P=.907).

[TABULAR DATA 6-7 OMITTED]

With regard to disability scores, the analysis of averaged within-study d-indexes showed no significant difference between comprehensive rehabilitation programs and primary back school intervention (comprehensive [program.sub.mean]=0-10, SD=0.17; primary [program.sub.mean]=-0.13, SD=1.10; F=0.12; df=1,6; P=.754). These results were similar to the findings based on multiple observations from individual studies illustrated in Table 7. The nonsignificant difference regarding acute versus chronic pain populations (Tab. 6) was also replicated using noncorrelated disability scores ([chronic.sub.mean]=0.16; SD=0.13; mixed [chronicity.sub.mean]=-0.66, SD=1.97; F=1.51; df=1,6; P=.265[double dagger]). Although the results were similar using the two different methods of analysis, the advantage of using noncorrelated data with some measures (ie, disability scores) may be offset by the diminished number of hypothesis tests (and reduced statistical power) that results from averaging within-study outcomes.

Discussion

An advantage of quantitative reviews is that they permit the use of broad dependent variables to assess the efficacy of therapeutic interventions, When the analyses were done across outcome variables, it was found that back schools were most effective for improving muscle strength and endurance and for improving knowledge about back care and compliance with the exercise and self-care programs (Tab. 5). Cohen[54] interprets a d-index of 0.20 to 0.50 as small, 0.50 to 0.80 as medium, and greater than 0.80 as large. The magnitude of effect sizes across outcome variables, therefore, was considered small.

The type of program and the use of cointerventions enhanced the effect size for back schools. Inpatient programs had a larger mean effect size than outpatient programs (Tab. 6). In addition, the inclusion of back school within a comprehensive rehabilitation program--even if that program was provided only for outpatients--resulted in a larger effect size compared with programs that used back school as a primary intervention.

[TABULAR DATA 5 OMITTED]

The negative average effect sizes found when back school was used as the primary intervention (Figs. 1 and 2 and Tabs. 6 and 7) may be explained, in part, by the type of control or comparison groups incorporated into these studies. Light[40] noted that differences in the control or comparison groups across studies could actually dilute the effect of the experimental intervention. A posteriori analysis, therefore, was done to determine whether the type of control or comparison group influenced the magnitude of the effect size in primary and comprehensive back school programs. A 2X2 ANOVA (placebo/comparison versus primary/ comprehensive back school) revealed larger effect sizes with placebo control, but only in the primary back school group (Fig. 3; interaction F=4.06; df=1,202; P=.045).

[TABULAR DATA 7 OMITTED]

When a placebo or no treatment was used as a control in studies of primary back school intervention, the mean effect size was larger compared with designs that incorporated traditional care such as exercise-only, analgesics, or physical therapy as the comparison group. This finding suggests that the effect of primary back school was enhanced" when the control group received placebo rather than some type of care. The effect size difference between traditional care and placebo-controlled studies for comprehensive back school programs was not statistically significant (Fig, 3).

The lack of influence of the type of control/comparison group on the effect size of comprehensive programs might be due to the number of hypothesis tests in each category. Only 33% of the hypothesis tests for comprehensive rehabilitation programs (n=34) compared the back school group with groups receiving traditional care. In contrast, 880/o of the hypothesis tests for primary back school programs (n=91) compared the back school group with a comparison group receiving traditional care. The number of traditional care comparison groups in primary back school studies, therefore, was notably larger than the traditional care comparisons in the comprehensive studies. This could have lowered the average effect (ie, blurred the distinction) between treatment and comparison groups to a greater extent in primary back school interventions compared with comprehensive programs (Fig. 3).

In terms of outcome variables, the enhanced effect size for comprehensive rehabilitation programs was due mostly to improvements in pain, spinal mobility, and strength/endurance outcomes. Work/vocational outcomes, the level of disability, or education/ compliance outcomes, however, were not different when comprehensive programs were compared with primary back school programs (Fig. 2 and Tab. 7), but the statistical power of these tests was low (Tab. 3). Patient understanding of his or her medical condition may be an important variable in predicting return to work. LaCroix et al[60] found that 95% of the patients with a good understanding of their condition returned to work compared with 33% of those with a poor understanding. The finding of a reasonable effect size for education/compliance outcomes (d=0.27-0.28; Tab. 7) in my study, however, was not "matched" with similar effect sizes for work/vocational outcomes (d=0.14 to -0.24; Tab. 7). Whether back school was provided as part of a comprehensive rehabilitation program or as a primary intervention, return to work or duration of sick leave was not improved beyond control levels. Although these results are consistent with recent noncontrolled or nonrandomized clinical trials that showed comprehensive treatment programs did not increase return to work rates,[7,8,11] the findings of my study should be viewed with caution because the number of hypothesis tests markedly reduced the power to detect effect-size differences between groups for this outcome.

Chronicity did not influence the magnitude of the effect size (Tab. 6). There was no difference between the mean effect sizes based on chronicity of the patients treated, and there was adequate power to detect a difference if one existed (Tab. 3). Early intervention is claimed by some to improve outcome,[5,6,34] but that position was not supported in the current analysis. This conclusion should be regarded as tentative, however, because the largest block of hypothesis tests were associated with chronic low back pain. Additional research is needed to address the effects of early intervention in subacute populations, with particular attention to work/vocational outcomes.

There was a difference in the variance per effect size between studies conducted in the Scandinavian countries and studies conducted in the other countries represented (Tab. 8). That is, the variance of effect sizes across "regions" was not homogeneous. Studies originating in Scandinavia and the Netherlands had larger effect sizes compared with the studies conducted in the United States, England, Canada, and Israel (Tab. 6). The highly developed social security systems in Sweden, Finland, and the Netherlands might have contributed to better outcomes. it is possible that the delivery of comprehensive care to patients with low back pain is facilitated by Scandinavian social-medical programs.[34] This approach is in contrast to the outpatient-based, primary back schools offered in non-Scandinavian countries.

[TABULAR DATA 8 OMITTED]

There is a potential for bias in reviews that contain only published material.[39,40,61] The underlying assumption is that a finding is more likely to be published if the results were statistically significant. The overall mean effect size of 0.07 (Tab. 4) and the inspection of individual trials in my review indicated that both positive and negative results have been accepted for publication. Publication bias, therefore, did not appear to be a relevant factor in my analysis.

Effect sizes in my study were derived from multiple results within single studies. In several studies,[5,6,24-27,32] the same population was used, but different outcome measures were evaluated. The effect size data, therefore, cannot be considered independent and may produce inflated or unreliable estimates derived from inferential statistical procedures.[42] Several investigators[37,39,62] have addressed this issue, and there is little consensus on how to manage the problem (see Boissel et al[39]). Glass et al stated that, in most cases, averaging all findings within a study up to the level of the study and proceeding with "studies" as the unit of analysis "is likely to obscure many important questions that can only be addressed at the within-study' level of outcome variables."[37(p229)] The analysis using single (averaged) within-study d-indexes for pain and disability, however, produced similar results compared with the analyses that utilized multiple within-study d-indexes for pain and disability. In addition, the use of nonindependent data versus averaged d-indexes did not alter the essential findings.

Conclusions

Back school can be an effective treatment for patients with low back pain when combined with a work-site visit, operant conditioning, cognitive-behavioral group therapy, or an intensive physical training regimen that supplements a traditional back exercise program. When back schools are not combined with a comprehensive rehabilitation program, the outcome is no better than the effects of control group intervention. Comprehensive programs that offer back school are equally effective when presented as inpatient or outpatient programs. Outcomes related to pain and physical impairment showed the greatest improvement with back school, but work/vocational and disability outcomes did not improve substantially beyond control levels.

This meta-analysis has generated several questions that can only be addressed by additional research. What modifications are needed in back school and comprehensive rehabilitation programs to improve work/vocational outcomes? What is the role of back school in preventing chronicity. Finally, what are the factors that contribute to better outcomes in programs implemented overseas, and what modifications in the health care delivery systems of the United States and other countries might be necessary to improve the outcomes of patients with work-related back injury?

[Figure 1-3 ILLUSTRATIONS OMITTED]

Appendix. Studies That Did Not Meet the Criteria for Inclusion in the Meta-analysis Study Count Author(s) Citation 1 Delitto et al Phys Ther, 1993;73:216-222 2 Mellin et al Spine, 1993;18:825-829 3 Jarvikoski et al J Spinal Disord, 1993;6:93-98 4 Brown et al Spine, 1992;17:1224-1228 5 Estlander et al Scand J Rehabil Med, 1991;23:97-102 6 Oland and Tviten Spine, 1991;16:457-459 7 Sirles et al AAOHN J, 1991;39:7-12 8 McCauley Am J Occup Ther, 1990;44:402-407 9 Deyo et al Nengl J Med, 1990;322:1627-1634 10 Kohles et al Spine, 1990;15:1321-1324 11 Mitchell et al Spine, 1990;15:514-521 12 Walsh et al Am J Phys Med Rehabil, 1990;69:245-250 13 Sachs et al Spine, 1990; 1 5:1325-1332 14 Hazard et al Spine, 1989;14:157-161 15 Saal and Saal Spine, 1989;14:431-437 16 Frederickson et al Spine, 1988;13:351-353 17 Julkunen et al Psychother Psychosom, 1988;50:173-181 18 Manniche et al Lancet, 1988;ii:8620-8627 19 Wiesel et al Spine, 1988;13:679-680 20 Carlton Am J Occup Ther, 1987;41:16-20 21 Mayer et al JAMA, 1987;258:1763-1767 22 Heinrich et al J Behav Med, 1985;8:61-78 23 Mayer et al Spine, 1985;10:482-493 24 Nwuga and Nwuga Physiotherapy Practice, 1985;1:99-105 25 Sikorski Spine, 1985;10:571-579 26 Hultman et al Applied Ergonomics, 1984; 1 5:127-133 27 Mellin et al Scand J Rehabil Med, 1984; 1 6:77-84 28 Simmons et al Orthopedics, 1984;7:1453-1456 29 Cohen et al J Clin Physiol, 1983;39:326-333 30 Buswell New Zealand Journal of Physiotherapy,

1982;10:13-17 31 Catchlove and Cohen Pain, 1982;14:181-191 32 Dehlin et al Scand J Rehabil Med, 1981;13:1-9 33 Kvien et al Scand J Rheumatol, 1981;10:318-320 34 Mantle et al Rheumatol Rehabil, 1981;20:227-232 35 Mattmiller Physiotherapy, 1980;66:118-122 36 Zachrisson-Forsell Physiotherapy, 1980;66:112-114 37 Newman et al Pain, 1978;4:283-292 38 Cairns and Pasino Behavior Therapy, 1977;8:621-630 39 Kendall and Jenkins Physiotherapy, 1968;54:154-157 40 Hall and Iceton Clin Orthop, 1983;179:10-17

References

[1] Mayer TG, Gatchel RJ, Kishino N, et al. Objective assessment of spine function following industrial injury: a prospective study with comparison group and one-year follow-up. Spine. 1985;10:482-493. [2] Mayer TG, Gatchel RJ, Mayer H, et al, A prospective two-year study of functional restoration in industrial low back injury: an objective assessment procedure. JAMA. 1987;258:1763-1767. [3] Hazard RG, Fenwick JW, Kalish SM, et al. Functional restoration with behavioral support: a one-year prospective study of patients with chronic low back pain. Spine. 1989;14:157-161. [4] Mitchell RI, Carmen GM. Results of a multicenter trial using an intensive active exercise program for the treatment of acute soft tissue and back injuries. Spine. 1990;15:514-521. [5] Lindstrom I, Ohlund C, Eek C, et al. The effect of graded activity on patients with subacute low back pain: a randomized prospective clinical study with an operant-condition behavioral approach. Phys Ther. 1992;72: 279-290. [6] Lindstrom I, Ohlund C, Eek C, et al. Mobility, strength, and fitness after a graded activity program for patients with subacute low back pain: a randomized prospective clinical study with a behavioral therapy approach. Spine. 1992;17:641-652. [7] Mellin G, Harkapaa K, Vanharanta H, et al. outcome of a multimodal treatment including intensive physical training of patients with chronic low back pain. Spine. 1993;18: 825-829. [8] Jarvikoski A, Mellin G, Estlander AM, et al. outcome of two multimodal back treatment programs with and without intensive physical training. J Spinal Disord. 1993;6:93-98. [9] Sachs BL, David JAF, Olimpio D, et al. Spinal rehabilitation by work tolerance based on objective physical capacity assessment of dysfunction: a prospective study with control subjects and twelve-month review. Spine. 1990;15:1325-1332. [10] Oland G, Tveiten G. A trial of modern rehabilitation for chronic low-back pain and disability: vocational outcome and effect of pain modulation. Spine, 1991;16:457-459. [11] Estlander AM, Mellin G, Vanharanta H, Hupli M. Effects and follow-up of a multimodal treatment program including intensive physical training for low back pain patients. Scand J Rehabil Med. 1991;23:97-102. [12] Saal JA, Saal JS. Nonoperative treatment of herniated intervertebral disc with radiculopathy: an outcome study. Spine. 1989;14:431-437. [13] Cairns D, Pasino JA. Comparison of verbal reinforcement and feedback in the operant treatment of disability due to chronic low back pain. Behavior Therapy. 1977;8:621-630. [14] Frederickson BE, Trief PM, Vanbeveren P, et al. Rehabilitation of the patient with chronic back pain: a search for outcome predictors. Spine. 1988;13:351-353. [15] Schlapbach P. Back school. in: Schlapbach P, Gerber NJ, eds. Physiotherapy: Controlled Trials and Facts--Rheumatology. Basel, Switzerland: S Karger AG, Medical and Scientific Publishers; 1991;14:25-33. [16] Stankovic R, Johnell O. Conservative treatment of acute low-back pain: a prospective randomized trial--McKenzie method of treatment versus patient education in "mini back school." Spine. 1990;15:120-123. [17] Donchin M, Woolf O, Kaplan L, Floman Y. Secondary prevention of low-back pain: a clinical trial. Spine. 1990;15:1317-1320. [18] Hurri H. The Swedish back school in chronic low back pain, part I: benefits. Scand J Rehabil Med. 1989;21:33-40. [19] Berwick DM, Budman S, Feldstein M. No clinical effect of back schools in an HMO: a randomized prospective trial. Spine. 1989;14:338-344. [20] Moffett JAK, Chase SM, Portek I, Ennis JR. A controlled prospective study to evaluate the effectiveness of a back school in the relief of chronic low back pain. Spine. 1986;11: 120-122. [21] Gilbert JR, Taylor DW, Hildebrund A, Evans C. Clinical trial of common treatments for low back pain in family practice. BMJ. 1985;291:791-794. [22] Lindequist S, Lundberg B, Wikmark R, et al. Information and regime at low back pain. Scand J Rebabil Med. 1984;16:113-116. [23] Lankhorst GJ, Van de Stadt RJ, Vogelarr TW, et al. The effect of the Swedish back school in chronic idiopathic low back pain: a prospective controlled study. Scand J Rebabil Med. 1983; 15:141-145. [24] Harkapaa K, Mellin G, Jarvikoski A, Hurri H. A controlled study on the outcome of inpatient and outpatient treatment of low back pain, part III: long-term follow-up of pain, disability, and compliance. Scand J Rehabil Med. 1990;22:181-188. [25] Mellin G, Harkapaa K, Hurri H, Jarvikoski A. A controlled study on the outcome of inpatient and outpatient treatment of low back pain, part IV: long-term effects on physical measurements. Scand J Rehabil Med. 1990;22:189-194. [26] Harkapaa K, Jarvikoski A, Mellin G, Hurri H. A controlled study on the outcome of inpatient and outpatient treatment of low back pain, part I: pain, disability, compliance, and reported treatment benefits three months after treatment. Scand J Rebabil Med. 1989;21: 81-89. [27] Mellin G, Huffi H, Harkapaa K, Jarvikoski A. A controlled study on the outcome of inpatient and outpatient treatment of low back pain, part II: effects on physical measurements three months after treatment. Scand J Rehabil Med. 1989;21:91-95. [28] Linton SJ, Bradley LA, Jensen I, et al. The secondary prevention of low back pain: a controlled study with follow-up. Pain. 1989;36:197-207. [29] Choler U, Larsson R, Nachemson A, Peterson LE. Back pain: attempt at a structural treatment program for patients with low back pain. SPRI Report 188. Social Planerings-och Rationaliseringsinstitut Rapport, Stockholm. Cited in: Nordin M, Cedraschi C, Balague F, Roux EB. Back schools in the prevention of chronicity. Baillieres Clin Rheumatol. 1992;6:685-703. [30] Aberg J. Evaluation of an advanced back pain rehabilitation program. Spine. 1984;9: 317-318. [31] Bergquist-Ullman M, Larsson U. Acute low back pain in industry: a controlled prospective study with special reference to therapy and confounding factors. Acta Orthop Scand Suppl. 1977; 170:1-117. [32] Hurri H, Mellin G, Korhonen O, et al. Aerobic capacity among chronic low-back-pain patients. J Spinal Disord. 1991;4:34-38. [33] Skovron ML. Epidemiology of low back pain. Baillieres Clin Rheumatol. 1992;6: 559-573. [34] Nordin M, Cedraschi C, Balague F, Roux EB. Back schools in the prevention of chronicity. Baillieres Clin Rheumatol. 1992;6: 685-703. [35] Linton SJ, Kamwendo K. Low back schools: a critical review. Phys Ther. 1987;67:1375-1383. [36] Klingenstiema U. Back schools/education programs: a review. Critical Reviews in Physical and Rehabilitation Medicine. 1991;3: 155-171. [37] Glass GV, McGaw B, Smth ML. Meta-Analysis in Social Research. London, England: Sage Publications Ltd; 1987. [38] Sacks HS, Berrier J, Reitman D, et al. Meta-analysis of randomized controlled trials. N Engl J Med. 1987;316:450-455. [39] Boissel JP, Blanchard J, Panak E, et al. Considerations for the meta-analysis of randomized clinical trials. Control Clin Trials. 1989;10:254-281. [40] Light RJ. Accumulating evidence from independent studies: what we can win and what we can lose. Stat Med. 1987;6:221-228. [41] Cooper HM. Scientific guidelines for conducting integrative research reviews. Rev Educ Res. 1982;52:291-302. [42] Ottenbacher K, Di Fabio RP. Efficacy of spinal manipulation/mobilization therapy: a meta-analysis. Spine. 1985;10:833-837. [43] Mantle MJ, Holmes J, Currey HLF. Backache in pregnancy, II: prophylactic influence of back care classes. Rheumatol Rebabil. 1981;20:227-232. [44] Walsh NE Schwartz RK. The influence of prophylactic orthoses on abdominal strength and low back injury in the workplace. Am J Phys Med Rebabil. 1990;69:245-250. [45] McCauley M. The effect of body mechanics instruction on work performance among young workers. Am J Occup Ther. 1990;44:402-407. [46] Carlton RS. The effects of body mechanics instruction on work performance. Am J Occup Ther. 1987;41:16-20. [47] Julkunen J, Hurri H, Kankainen J. Psychological factors in the treatment of chronic low back pain. Psychother Psychosom. 1988;50:173-181. [48] Deyo RA. Conservative therapy for low back pain: distinguishing useful from useless therapy. JAMA. 1983;250:1057-1062. [49] Di Fabio RP. Efficacy of manual therapy. Phys Ther. 1992;72:853-864. [50] Huskisson EC. Measurement of pain. Lancet. 1974;2:1127-1131. [51] Meizack R. The McGill pain questionnaire: major properties and scoring methods. Pain. 1977;1:277-299. [52] Follick MJ, Smith TV, Ahem DK. The sickness impact profile: a global measure of disability in chronic low back pain. Pain. 1985;21:67-76. [53] Fairbank JCT, Couper J, Davies JB, O'Brien JP. The Oswestry low back pain disability questionnaire. Physiotherapy. 1980;66: 271-273. [54] Cohen J. Statistical Power Analysis for the Behavioral Sciences. Rev ed. London, England: Lawrence Erlbaum Accociates; 1987. [55] Friedman H. Magnitude of experimental effect and a table for its rapid estimation. Phsychol Bull. 1968;70:245-251. [56] Hedges LV. Unbiased estimation of effect size. Eval Educ. 1980;4:25-27. [57] Brown MB, Forsythe AB. Robust tests for the equality of variances. J Amer Statist Assoc. 1974;69:364-367. [58] BMDP Statistical Software Manual. Berkeley, Calif: University of California Press, Berkeley; 1992. [59] Tukey JW. Exploratory Data Analysis. Boston, Mass: Addison-Wesley; 1977. [60] Lacroix JM, Powell J, Lloyd GJ, et al. Lowback pain: factors of value in predicting outcome. Spine. 1990;15:495-499. [61] Rosenthal R. Combining results of independent studies. Psychol Bull, 1978;85: 185-193. [62] Wittes RE. Problems in the medical interpretation of overviews. Stat Med. 1987;6:269-276.

(*) Version PC90, BMPD Statistical Software Inc, Los Angeles, CA 90086, ([dagger]) Power Analysis is a "stand-alone" software product distributed by BMDP Statistical Software Inc, Los Angeles, CA.

RP Di Fabio, PhD, PT, is Professor and Director of Graduate Studies, Program in Physical Therapy, Department of Physical Medicine and Rehabilitation, UMHC Box 388, University of Minnesota, Minneapolis, MN 55455 (USA) ([e-mail]).

This report was modified from a presentation at the Second Joint Congress of the American Physical Therapy Association and the Canadian Physiotherapy Association; Toronto, Ontario, Canada; June 8, 1994.

This article was submitted December 14, 1994, and was accepted May 5, 1995.