Clinical examination variables discriminate among treatment-based classification groups: a study of construct validity in patients with acute low back pain.

Background and Purpose. Treatment-based classification (TBC) provides matched interventions for patients with acute low back pain (LBP) through key history and clinical findings. This study investigated the discriminant validity of TBC by determining whether commonly used clinical examination variables discriminated among TBC groups. Subjects. The mean age of the 131 participants was 37.7 years (SD = 10.1), 66 participants (50.4%) were female, mean duration of LBP was 16.5 days (SD = 16.1), and 60 participants (45.8%) had a prior history of LBP. Fifty-one study participants (38.9%) were classified for specific exercise, 42 (32.1%) for mobilization, 28 (21.4%) for immobilization, and 10 (7.6%) for traction. Methods. One-way analyses of variance and Kruskal-Wallis tests were used to investigate differences in clinical variables by TBC group. Then, discriminant function analysis (DFA) predicted TBC group membership. Results. The TBC groups differed on present pain intensity, duration of LBP, and history of LBP. Present pain intensity, duration of LBP, total lumbar flexion, presence of leg pain, and history of LBP produced 2 statistically significant discriminant functions that predicted TBC group membership. These functions correctly classified (cross-validation value in parentheses) 65% (65%) for specific exercise, 45% (40%) for mobilization, and 32% (32%) for immobilization. Discussion and Conclusion. This study provided evidence supporting the discriminant validity of TBC. Additional diagnostic information related to TBC groups was generated. [George SZ, Delitto A. Clinical examination variables discriminate among treatment-based classification groups: a study of construct validity in patients with acute low back pain. Phys Ther. 2005;85:306-314.]

Key Words: Classification, Diagnosis, Discriminant validity, Examination, Physical therapy.

Identification of subgroups of patients with low back pain (LBP) and interventions that relate to these subgroups has been highlighted as a research priority. (1) In 1995, Delitto et al (2) described a treatment-based classification (TBC) system that allowed physical therapists to systematically classify patients for physical therapy intervention. The underlying premise of TBC is that subgroups of patients with acute LBP can be identified from key history and clinical examination findings. Furthermore, the authors of TBC hypothesized that each subgroup would respond favorably to a specific intervention, but only when the applied intervention matched the subgroup's clinical presentation. In these ways, TBC is distinguished from another proposed classification system that relies on pathological anatomy and temporal differences (ie, Quebec Task Force Classification [QTFC] (3)) and is similar to one that leads to conservative intervention (ie, McKenzie approach (4)).

The clinician uses TBC to test classification-related hypotheses (ie, which classification group is believed to be appropriate for a particular patient) by identifying history and clinical examination findings described for each classification (Tab. 1). (2) Seven different classification groups were originally described in TBC; however, recent investigations have collapsed the 7 classification groups to 4 (Tab. 1). (5) The classification groups were collapsed because each group received conceptually similar physical therapy intervention, although specific application of the intervention differed within the group. For example, the mobilization classification consists exclusively of patients receiving manual physical therapy. Within the mobilization classification, however, the manual intervention will be directed at either the lumbar or sacroiliac joint region, depending on additional clinical findings.

Since the TBC was originally proposed, reports in the literature have described its decision-making process (6) and short-term treatment outcomes have been reported for groups of patients. (5) Randomized clinical trials have demonstrated that TBC effectively reduces self-reported disability from acute LBP. Specifically, TBC-guided physical therapy was found to be more effective when compared with generic exercise (7,8) or guideline-based intervention. (9) Although these reports document consistent support for the prescriptive validity of TBC, parallel investigations of its discriminant validity have been under-reported in the literature. It appears that only one study has provided data supporting the discriminant validity of TBC. Fritz and George (5) used TBC to classify 120 consecutive patients seeking outpatient physical therapy, and statistically significant initial differences were observed among the specific exercise, immobilization, and mobilization classification groups. Specifically, patients in the specific exercise group were older and more likely to have leg pain, and patients in the immobilization group were more likely to have a previous history of LBP. (5) Although that study provided preliminary evidence supporting the discriminant validity of TBC, only univariate differences among the classification groups were reported. Studies investigating multivariate differences in classification groups have not been reported, but these studies are necessary because TBC theory emphasizes that groups of variables are used to determine classification membership.

Additional studies supporting the discriminant validity of TBC are needed for 2 other reasons. First, there is limited evidence confirming the premise in TBC that distinct subgroups of patients with acute LBP can be identified. For example, the QTFC has demonstrated discriminant validity for patient subgroups with different amounts of radiating pain and neurologic findings. (10-12) The QTFC, however, has not demonstrated discriminant validity for patient subgroups based on differences in acute and chronic LBP. Furthermore, in previous studies of the QTFC, researchers either did not consider physical impairment measures (10-12) or did not find impairment measures to be discriminative. (13) Demonstrating that measures of symptom severity, disability or functional limitation, or physical impairment can accurately identify subgroups of patients with acute LBP is necessary to support underlying TBC theory.

Second, examination findings used in the original TBC classification guidelines have recently been challenged, especially for the mobilization classification. The mobilization classification is based on clinical tests involving palpation of static and dynamic pelvic landmark asymmetries. (2) However, problems with the reliability (14-16) and validity (17,18) of data obtained with these tests have been well documented in the literature. Using a cluster of these tests was believed to address these issues, (19,20) but a multicenter study by Riddle and Freburger (16) has demonstrated that a cluster of tests also yields data with question able reliability. The question of TBC reliability is not exclusive to the mobilization classification, because the interrater reliability for the entire TBC system has been reported as "low" ([kappa] = .14-.41) by some investigators. (21) Further investigation into what examination findings distinguish TBC groups could provide information on how to modify the original classification guidelines and perhaps enhance the repeatability of TBC.

These issues highlight that, with regard to the discriminant validity of TBC, "further research is required to validate, refute, or refine the classifications." (5(p113)) Therefore, we believe that additional studies are warranted, especially those that include measures of symptom severity, disability or functional limitation, and physical impairment and use multivariate analytical techniques. These studies will provide converging or diverging evidence on the prescriptive validity information that has already been presented in support of TBC. The purpose of our study, therefore, was to determine whether commonly used history and clinical examination variables differed and discriminated among TBC groups. We hypothesized that the history and clinical examination variables studied would be able to discriminate among TBC groups at rates that are better than chance. This study used methods similar to those of the original report by Fritz and George, (5) but included symptom severity, disability or functional limitation, and physical impairment measures and used a multivariate technique to predict TBC group membership.

Methods

Subjects All participants provided informed consent before enrolling in the study, and their human rights were protected throughout the duration of this study. Subjects were participants in 2 separate clinical trials that investigated the effects of different physical therapy interventions on acute LBP. The first clinical trial (9) focused solely on patients with work-related LBP (n = 65), whereas the second clinical trial (22) included patients with and without work-related LBP (n = 66). Subjects underwent a standard initial examination, completed self-report measures, and were classified for treatment by licensed physical therapists who had received university-level training with the TBC system.

Measures

Measures were selected for use in this study based on the following inclusionary criteria: (1) consistent use between clinical trials and (2) relevance to TBC guidelines from previously published TBC reports (2,5,6) or other related reports (14-18,23)

Measures were excluded if they met any 1 of the 3 potential exclusionary criteria:

1. They were not consistently used between clinical trials (ie, muscle testing, sensory and reflex testing, generalized ligament laxity and identification of patterns, such as opening and closing tests, during lumbar movement testing).

2. They were not relevant to previously published TBC guidelines for identifying classification group (ie, nonorganic signs and symptoms).

3. They were reported to yield data with poor reliability (14-16) and validity (17,18,23) (ie, measures of static and dynamic pelvic landmark assessment).

The following measures were included in this study.

Demographic and history information. Using a standard form, demographic and history information was obtained. Data collected included: age, sex, the duration of LBP, the previous history of LBP, and the presence of leg pain.

Physical impairment. Physical impairment data were collected in a standardized manner described by Waddell et al. (24) The measures used for this study were total lumbar flexion range of motion (in degrees) and average straight-leg-raise range of motion (in degrees). These 2 measures demonstrated excellent interrater reliability (intraclass correlation coefficients [ICCs] = .94, .94, and .96 for total lumbar flexion and right and left straight leg raise, respectively) in a previous study. (24)

Disability. Low back-related disability was assessed with the Oswestry Disability Questionnaire (ODQ), a 10-item scale originally described by Fairbank et al. (25) Each item is scored from 0 to 5, and the final score is expressed as a percentage, with higher numbers indicating greater disability (range = 0-100). The ODQ used in our study was modified from the original questionnaire by substituting a section regarding employment or homemaking ability for the section related to sex life. (26,27) Previous studies showed the modified version of the ODQ to have high levels of reliability (ICC = .90). (26,27)

Pain intensity. Patients rated their present pain intensity by marking their pain level on a 0-to-10 numerical rating scale. (28) For this scale, a rating of 0 corresponded to "no pain intensity" and a rating of 10 corresponded to "maximum pain intensity." A previous study (29) demonstrated an acceptable range of reliability (ICC = .66-.93) when using a similar method of assessing pain intensity in patients with LBP.

Data Analysis

All statistical analyses were performed with SPSS for Windows, Version 11.0.1,* with an alpha level of .05. Descriptive statistics were generated for the measures. Baseline TBC differences in variables were investigated by one-way analysis of variance (ANOVA) and Kruskal-Wallis tests, as appropriate. A discriminant function analysis (DFA) then was performed to investigate how clinical examination variables discriminated among the classification groups. To avoid excluding any potentially discriminating variables, a liberal criterion (P<.20) was used to determine variables that would be entered into the DFA. Discriminant function analysis is a multivariate technique that classifies individuals into separate groups, and classification is based on how sets of variables differ among the groups. This technique classifies k groups by identifying linear combinations of variables that produce differences and then maximizing the predictive nature of functions that discriminate among the groups. (30,31) When DFA classification is used to predict group membership, the number of functions generated is equal to k-1. (30,31) As in regression analysis, DFA can be used for descriptive and predictive purposes, and both were utilized in this study.

Results

The mean age of the 131 participants was 37.7 years (SD = 10.1), and 66 participants (50.4%) were female. Participants had a mean duration of LBP of 16.5 days (SD = 16.1), 60 (45.8%) had a prior history of LBP, 69 (52.7%) had work-related low back injury, and 53 (40.5%) had leg pain in addition to LBP. Participants had a mean pain intensity of 5.6 (SD = 2.3), a mean self-report of disability of 40.3 (SD = 14.3), mean lumbar flexion range of motion of 63.9 degrees (SD = 30.1), and a mean average straight-leg-raise range of motion of 66.3 degrees (SD = 14.7). Fifty-one study participants (38.9%) were classified for specific exercise, 42 (32.1%) for mobilization, 28 (21.4%) for immobilization, and 10 (7.6%) for traction intervention. Statistical assumptions regarding equal variances were violated when the planned data analyses were attempted, because of the small number of patients in the traction classification (n = 10). Therefore, this classification group was not included in subsequent data analyses.

The specific exercise, mobilization, and immobilization TBC groups (n = 121) differed on 3 clinical examination variables, and these differences are summarized in Tables 2 and 3. The specific exercise group had higher present pain intensity (mean difference = l.4, P<.039) compared with the immobilization group. The immobilization group had a longer duration of LBP (mean difference = 9.6 days, P<.029) and was more likely to have a history of LBP (79% versus 51%, P<.017) compared with the specific exercise group. The immobilization group also was more likely to have a history of LBP (79% versus 38%, P<.007) compared with the mobilization group.

Five variables (pain intensity, duration of LBP, total lumbar flexion, presence of leg pain, and history of LBP) met the criterion (P<.20) for entrance into the DFA that predicted TBC group membership. For the prediction to be robust, an assumption of concern--the homogeneity of the variance-covariance matrix--was not violated in this case (Box's M = 42.76, P<.106), The DFA produced 2 statistically significant discriminant functions, indicating that both predicted TBC membership at better-than-chance probability. The first discriminant function accounted for 62.7% of the between-groups variability, with a canonical correlation of 0.376 (Wilks [lambda] = 0.782, df = 10, [chi square] = 28.460, P<.002). The second discriminant function accounted for 37.3% of the remaining between-groups variability, with a canonical correlation of 0.298 (Wilks [lambda] = 0.911, df = 4, [chi square] = 10.809, P<.029).

The structure coefficients for the variables were used to describe the corresponding discriminant functions (Tab. 4). Inferential testing of DFA structure coefficients is not commonly performed, so we used an a priori determined cutoff of 0.3 for determining coefficients that meaningfully contributed to the function. (30) The first function was described by a negative loading of present pain intensity and positive loadings of previous history and duration of LBP (Tab. 4). The second function was described by a negative loading on lumbar flexion and positive loadings on pain intensity and presence of leg pain (Tab. 4).

Prediction was determined by 2 different methods for our study: one method used all available data points to determine the functions and classify individuals, and the other method used all available data points except one to determine the functions. The excluded data point then was classified using the functions generated by the other data points (cross-validation technique). The overall accuracy for classification using the discriminant functions was 50% when all data points were used to determine classification and 49% when the cross-validation technique was used. The percentages classified correctly (value of cross-validation technique in parentheses) were 65% (65%) for specific exercise, 45% (40%) for mobilization, and 32% (32%) for immobilization. Based on sample distribution, the prior probabilities for chance assignment were 42% for specific exercise, 35% for mobilization, and 23% for immobilization. Therefore, classification by the discriminant functions exceeded chance classification for all 3 TBC groups, even when the cross-validated values were compared.

Discussion

This study provided additional evidence supporting the discriminant validity of TBC, and the results largely agreed with what has been previously reported on this topic. (5) In a previous study, (5) subjects in the specific exercise classification were older than those in the other classification groups, a finding that was not replicated in the present study. We believe the most likely explanation for this discrepancy is that, in the present study, we limited the enrollment of older patients by exclusively recruiting working-age adults (9) or limiting the maximum age to 55 years. (22) Older patients are more likely to have lumbar spinal stenosis and are believed to be more likely to be appropriate for the specific exercise classification. (5) Therefore, an expected outcome of limited recruitment of older patients would be a decrease in the mean age of the specific exercise classification. When sample distribution was considered, a better-than-chance classification rate was observed when history of LBP, duration of LBP, presence of leg pain with LBP, pain intensity, and total lumbar flexion simultaneously predicted TBC membership. Previous reports (32,33) have primarily focused on these examination variables as prognostic factors, but our results suggest that combinations of these variables (ie, discriminant functions) also are useful for classification purposes. Furthermore, the generated functions appeared to be stable estimates, because there was minimal change in the prediction accuracy when a cross-validation technique was used.

These results offer similarity and contrasts compared with studies of discriminant validity for another commonly reported classification system for LBP--the QTFC. Unfortunately, we did not identify any studies that addressed the discriminant validity of the McKenzie approach for comparison purposes. Measures of symptom severity discriminated among subgroups in the TBC and QTFC systems, (10-12) suggesting that these measures may be globally useful for classification purposes. Measures of disability or functional limitation discriminated among subgroups for the QTFC, (11,12) but not for TBC. We believe the reason for this discrepancy is that TBC uses measures of disability and functional limitation to stage patients for intervention. (2) In TBC, patients classified for a specific intervention (ie, mobilization, immobilization, specific exercise, or traction) will, by design, have similar levels of disability or functional limitation. In contrast, QTFC stages patients based on differences in anatomical, temporal, and work-related parameters, which appears to be more likely to result in subgroups with differences in disability or functional limitation.

Our study also demonstrated that a physical impairment measure (total lumbar flexion) discriminated among TBC groups. This finding differs from what has been reported in the literature, because previous studies either have not included impairment measures or have shown that these measures did not discriminate among LBP subgroups. For example, isometric trunk muscle force could not distinguish between patients classified with chronic LBP using the QTFC and patients with general pain syndromes. (13) Our study gives an initial indication that certain physical impairment measures may be useful for classification purposes, at least those related to TBC.

The validity of data for a classification system, such as TBC, is something that cannot be established through one study or with one type of design. Instead, evidence supporting or refuting the system is gathered from different sources and from the use of different methods. In the best-case scenario, these sources converge and indicate similar meanings of the underlying constructs being studied. Our study and the previously reported studies of discriminant (5) and predictive validity (7-9) demonstrate such a convergence, providing broad support for the TBC approach put forth in our report and other reports.

This study also addressed some diagnostic issues related to TBC, which is pertinent because a recent review of 4 prominent rehabilitation journals documented the lack of clinically relevant research, especially in the areas of diagnosis. (34) The original TBC guidelines published in 1995 were proposed primarily from clinical experiences. The validity of data for these classification guidelines has been the focus of recent investigations, and some results suggest the original guidelines should be modified. (16,23) Our study confirmed the importance of leg pain, history of LBP, and duration of LBP from the original guidelines (2) and suggested that total lumbar flexion and pain intensity also should be considered for addition to TBC decision making.

When univariate differences were considered, classifications differed on specific examination variables (history of LBP, duration of pain, and pain intensity). Unfortunately, univariate group differences are of little help when trying to make complex, TBC-related clinical decisions, unless they are associated with mutual exclusiveness. For example, history of LBP significantly differs among classification groups and is a key examination finding in patients classified for immobilization (Tab. 1). However, our study demonstrated that this finding is not mutually exclusive to the immobilization classification, because 38% of the participants in the mobilization classification and 51% of those in the specific exercise classification also exhibited that finding (Tab. 3).

Instead of considering only univariate group differences, we believe that physical therapists should integrate multiple examination findings when using TBC. We believe multivariate statistical approaches, such as DFA, better approximate this process by simultaneously considering examination findings and generating functions that maximally differ on these findings. The resultant functions could help physical therapists determine what factors predict a certain TBC group. In this study, the first discriminant function (Tab. 4) appeared to predict immobilization classification because it was described by a previous history of LBP, a longer duration of LBP, and lower pain intensity ratings. These clinical variables are consistent with lumbar instability, which is believed to be the most likely pathology associated with the immobilization classification. (35) The second discriminant function was described by presence of leg pain, higher pain intensity ratings, and lower amounts of lumbar flexion (Tab. 4). The second function appeared to predict the specific exercise classification because of its hypothesized association with limited lumbar motion in the sagittal plane, (2) and the results of a previous study (5) suggest that patients in this classification are more likely to have leg pain.

Study Limitations

The clinical utility of these findings is limited because some of the examination findings were treated as continuous variables. For example, lower pain intensity ratings, longer pain duration, and a history of LBP predicted the immobilization classification. Although a patient can be categorized into a positive or negative finding for a variable such as history of LBP, our study did not provide threshold values for what constitutes "lower pain duration" and "longer pain duration." Future studies will have to clarify this issue before the functions described in our study can be implemented into routine clinical practice.

A major limitation of this study was that a discriminant function to predict mobilization classification was not generated. This was a product of the statistical analysis used. Discriminant function analysis produces one less function than number of total groups, and the researcher does not control the group eliminated. As chance would have it, the examination variables used in this study maximally predicted classification into the immobilization and specific exercise groups. This finding can be attributed to pure chance, or perhaps this finding corroborates reliability studies indicating that the mobilization classification is difficult to identify above error rates. Flynn et al (23) have approached the mobilization classification from a different perspective and, in the process, have identified 5 examination findings that predict successful outcome after manipulation. Interestingly, 2 findings that predict manipulation success (<16-day duration of LBP and no leg pain with LBP) also were included in our study. They were not associated with immobilization or specific exercise classifications, and we interpret this as a preliminary suggestion that these factors could be associated with the mobilization classification. However, future studies are necessary to confirm whether these factors, or any of the others described by Flynn et al (23) as predicting treatment success, also predict classification in the mobilization category.

Another limitation of this study is that it included only patients with acute or subacute LBP, when a temporal definition is used. Therefore, these results cannot be generalized to patients with chronic LBP when the most common definition of "chronic" is used (ie, duration of symptoms 3 months or greater). It also should be noted that only specific exercise, mobilization, and immobilization TBC groups were included in the analysis because of the small sample size of the traction classification. Therefore, this study does not provide information on how the traction classification differs from the other Classification groups. Last, the previously outlined concerns regarding the potential age limitation associated with the specific exercise classification should be considered when interpreting the results of our study.

Directions for Future Research

The discriminant functions proposed in this study are tentative, and future prospective studies are necessary to confirm or refute their role in predicting TBC group membership. Another consideration for future research is whether implementation of the discriminant functions outlined in our study improves the reliability associated with TBC. The reliability reported with a previous investigation of TBC was described as "moderate" ([kappa] = .56), (5) but other investigators (21) have reported lower reliability ([kappa] = .14-.41). If these functions truly discriminate among TBC groups, then their inclusion should improve the reliability of classification. Different ways of improving TBC's reliability have been discussed, (5,21) but emphasizing groups of clinical examination variables that discriminate among TBC groups has not been previously considered. A study comparing the reliability of TBC using the originally described clinical factors with these statistically derived factors would add to the efficient management of patients with acute LBP.

Conclusion

This study supported the discriminant validity of TBC and provided additional, preliminary diagnostic information related to TBC for LBP. Future investigations are necessary to confirm these variables as predictors of TBC group membership and to determine whether they improve the reliability associated with TBC.

* SPSS Inc, 233 S Wacker Dr, Chicago, IL 60606.

SZ George, PT, PhD, is Assistant Professor, Department of Physical Therapy, Brooks Center for Rehabilitation Studies, University of Florida, PO Box 100154, Gainesville, FL 32610-0165 (USA) ([e-mail]). Address all correspondence to Dr George.

A Delitto, PT, PhD, FAPTA, is Chair and Associate Professor, Department of Physical Therapy, University of Pittsburgh, Pittsburgh, Pa.

Both authors provided concept/idea/research design, writing, and fund procurement. Dr George provided data analysis. The authors acknowledge Joel Bialosky, PT, MS, OCS, FAAOMPT, Douglas Donald, PT, MPT, Tara Ridge, PT, MPT, MS, SCS, and Michelle Vignovic, PT, MS, OCS, FAAOMPT, for their assistance with the manuscript.

Support for this study was provided by the Foundation for Physical Therapy in the form of a PODS II scholarship and the National Institute of Disability and Rehabilitation Research (NIDRR) in the form of a postdoctoral fellowship.

This study was approved by the University of Pittsburgh's Institutional Review Board.

This article was received December 22, 2003, and was accepted October 1, 2004.

References

(1) Borkan JM, Koes B, Reis S, Cherkin DC. A report from the Second International Forum for Primary Care Research on Low Back Pain: reexamining priorities. Spine. 1998;23:1992-1996.

(2) Delitto A, Erhard RE, Bowling RW. A treatment-based classification approach to low back syndrome: identifying and staging patients for conservative treatment. Phys Ther. 1995;75:470-485.

(3) Scientific approach to the assessment and management of activity-related spinal disorders: a monograph for clinicians. Report of the Quebec Task Force on Spinal Disorders. Spine. 1987;12(7 suppl): S1-S59.

(4) McKenzie RA. The Lumbar Spine: Mechanical Diagnosis and Therapy. Waikanae, New Zealand: Spinal Publications Ltd; 1989.

(5) FritzJM, George SZ. The use of a classification approach to identify subgroups of patients with acute low back pain: interrater reliability and short-term treatment outcomes. Spine. 2000;25:106-114.

(6) Fritz JM. Use of a classification approach to the treatment of 3 patients with low back syndrome. Phys Ther. 1998;78:766-777.

(7) Erhard RE, Delitto A, Cibulka MT. Relative effectiveness of an extension program and a combined program of manipulation and flexion and extension exercises in patients with acute low back syndrome. Phys Ther. 1994;74:1093-1100.

(8) Delitto A, Cibulka MT, Erhard RE, et al. Evidence for use of an extension-mobilization category in acute low back syndrome: a prescriptive validation pilot study. Phys Ther. 1993;73:216-222.

(9) Fritz JM, Delitto A, Erhard RE. Comparison of classification-based physical therapy with therapy based on clinical practice guidelines for patients with acute low back pain: a randomized clinical trial. Spine. 2003;28:1363-1371.

(10) Atlas SJ, Deyo RA, Patrick DL, et al. The Quebec Task Force classification for spinal disorders and the severity, treatment, and outcomes of sciatica and lumbar spinal stenosis. Spine. 1996;21: 2885-2892.

(11) Loisel P, Vachon B, Lemaire J, et al. Discriminative and predictive validity assessment of the Quebec Task Force Classification. Spine. 2002;27:851-857.

(12) Werneke MW, Hart DL. Categorizing patients with occupational low back pain by use of the Quebec Task Force Classification system versus pain pattern classification procedures: discriminant and predictive validity. Phys Ther. 2004;84:243-254.

(13) Rantanen P. Physical measurements and questionnaires as diagnostic tools in chronic low back pain. J Rehabil Med. 2001;33:31-35.

(14) Potter NA, Rothstein JM. Intertester reliability for selected clinical tests of the sacroiliac joint. Phys Ther. 1985;65:1671-1675.

(15) Van Dillen LR, Sahrmann SA, Norton BJ, et al. Reliability of physical examination items used for classification of patients with low back pain. Phys Ther. 1998;78:979-988.

(16) Riddle DL, Freburger JK. Evaluation of the presence of sacroiliac joint region dysfunction using a combination of tests: a multicenter intertester reliability study. Phys Ther. 2002;82:772-781.

(17) Levangie PK. Four clinical tests of sacroiliac joint dysfunction: the association of test results with innominate torsion among patients with and without low back pain. Phys Ther. 1999;79:1043-1057.

(18) Levangie PK. The association between static pelvic asymmetry and low back pain. Spine. 1999;24:1234-1242.

(19) Cibulka MT, Delitto A, Koldehoff RM. Changes in innominate tilt after manipulation of the sacroiliac joint in patients with low back pain: an experimental study. Phys Ther. 1988;68:1359-1363.

(20) Cibulka MT, Koldehoff RM. Clinical usefulness of a cluster of sacroiliac joint tests in patients with and without low back pain. J Orthop Sports Phys Ther. 1999;29:83-89.

(21) Heiss DG, Fitch DS, Fritz JM, et al. The interrater reliability among physical therapists newly trained in a classification system for acute low back pain. J Orthop Sports Phys Ther. 2004;34:430-439.

(22) George SZ, Fritz JM, Bialosky JE, Donald DA. The effect of a fear-avoidance-based physical therapy intervention for patients with acute low back pain: results of a randomized clinical trial. Spine. 2003;28:2551-2560.

(23) Flynn T, Fritz JM, Whitman J, et al. A clinical prediction rule for classifying patients with low back pain who demonstrate short-term improvement with spinal manipulation. Spine. 2002;27: 2835-2843.

(24) Waddell G, Somerville D, Henderson I, Newton M. Objective clinical evaluation of physical impairment in chronic low back pain. Spine. 1992;17:617-628.

(25) Fairbank JC, Couper J, Davies JB, O'Brien JP. The Oswestry Low Back Pain Disability Questionnaire. Physiotherapy. 1980;66: 271-273.

(26) Fairbank JC, Pynsent PB. The Oswestry Disability Index. Spine. 2000;25:2940 -2952.

(27) Roland M, Fairbank J. The Roland-Morris Disability Questionnaire and the Oswestry Disability Questionnaire. Spine. 2000;25: 3115-3124.

(28) Chan CW, Goldman S, Ilstrup DM, et al. The pain drawing and Waddell's nonorganic physical signs in chronic low-back pain. Spine. 1993;18:1717-1722.

(29) Roach KE, Brown MD, Dunigan KM, et al. Test-retest reliability of patient reports of low back pain. J Orthop Sports Phys Ther. 1997;26: 253-259.

(30) Marcoulides GA, Hershberger SL. Discriminant Analysis--Multivariate Statistical Analysis Methods: A First Course. Mahwah, NJ: Lawrence Erlbaum Associates; 1997:85-131.

(31) Afifi AA, Clark V. Discriminant Analysis. 3rd ed. New York, NY: Chapman and Hall; 1996:243-280.

(32) Coste J, Delecoeuillerie G, Cohen de Lara A, et al. Clinical course and prognostic factors in acute low back pain: an inception cohort study in primary care practice. BMJ. 1994;308:577-580.

(33) McIntosh G, Frank J, Hogg-Johnson S, et al. Prognostic factors for time receiving workers' compensation benefits in a cohort of patients with low back pain. Spine. 2000;25:147-157.

(34) Miller PA, McKibbon KA, Haynes RB. A quantitative analysis of research publications in physical therapy journals. Phys Ther. 2003;83: 123-131.

(35) Fritz JM, Erhard RE, Hagen BF. Segmental instability of the lumbar spine. Phys Ther. 1998;78:889-896.

Table 1.
Key History and Clinical Findings for Treatment-Based
Classification Decision Making

Classification   Key History and             Intervention
                 Clinical Findings

Specific exercise (a)

Extension        Postural preference for     Extension exercises
syndrome         extension
                                             Avoidance of flexion
                 Flexion activities          activities
                 increase pain

                 Status improves
                 (centralizes) with
                 extension lumbar
                 movement testing

                 Status worsens with
                 flexion testing

Flexion          Postural preference for     Flexion exercises
syndrome         flexion
                                             Avoidance of extension
                 Extension activities        activities
                 increase pain

                 Status improves
                 (centralizes) with
                 flexion lumbar movement
                 testing

                 Status worsens with
                 extension testing

Lateral          Visible frontal plane       Manual or self-performed
shift            deformity--shoulders        pelvic translocation
syndrome         compared with pelvis        exercises

                 Unilateral side bending
                 restriction during lumbar
                 movement testing

                 Status improves
                 (centralizes) with pelvic
                 translocation

Mobilization (a)

Lumbar           Report of local,            Lumbar mobilization or
mobilization     unilateral low back pain    manipulation technique

                 "Opening pattern" during    Lumbar range-of-motion
                 lumbar movement testing     exercises

                 "Closing pattern" during
                 lumbar movement testing

Sacroiliac       Report of local pain at     Sacroiliac region
mobilization     PSIS, (b) buttock, or       manipulation or muscle
                 lateral thigh               energy technique

                 Three fourths of            Lumbar range-of-motion
                 sacroiliac test cluster     exercises
                 are positive

Immobilization (a)

Immobilization   Frequent prior episodes     Trunk strengthening
syndrome         of low back pain, with      exercises
                 minimal perturbation

                 History of trauma

                 Generalized ligamentous
                 laxity

                 "Instability catch"
                 during flexion lumbar
                 movement testing

Traction (a)

Traction         Signs and symptoms of       Intermittent mechanical
syndrome         nerve root compression      traction

                 No improvement              Autotraction
                 (centralization) with
                 lumbar movement testing

Lateral          Visible frontal plane       Autotraction
shift            deformity-shoulders
syndrome         compared with pelvis

                 Unilateral side-bending
                 restriction during lumbar
                 movement testing

                 Status worsens with
                 pelvic translocation

(a) Indicates a collapsed treatment-based classification.

(b) PSIS = posterior superior iliac spine.

Table 2.
Baseline Differences in Treatment-Based Classification Groups for
Continuous Variables (a)

                            Treatment-Based Classification

                            Specific
                            Exercise      Mobilization   Immobilization
Variable                    (n = 51)      (n = 42)       (n = 28)

Age (y)                     38.6 (10.4)   36.5 (9.2)     36.6 (11.0)
Duration (d) (b)            14.0 (14.5)   14.5 (15.0)    23.6 (18.4)
Pain intensity (0-10) (b)    6.2 (2.2)     5.5 (2.0)      4.8 (2.7)
Disability (0-100)          40.0 (14.5)   39.3 (12.6)    39.5 (15.4)
Flexion ([degrees])         58.1 (28.2)   67.8 (31.5)    70.5 (33.9)
SLR (c) ([degrees])         66.2 (11.4)   69.1 (12.0)    66.8 (20.0)

Variable                      F        P

Age (y)                       0.61     .544
Duration (d) (b)              3.94     .022
Pain intensity (0-10) (b)     3.29     .041
Disability (0-100)            0.03     .974
Flexion ([degrees])           1.90     .155
SLR (c) ([degrees])           0.52     .596

(a) All values are reported as mean (SD), df = 2,57. Boldfaced type
indicates variable was selected for discriminant function analysis
(P < .20).

(b) Bonferonni adjusted post hoc testing indicates significant
difference between specific exercise and immobilization groups.

(c) SLR = straight leg raise.

Table 3.
Baseline Differences in Treatment-Based Classification Groups
for Categorical Variables (a)

                             Treatment-Based Classification

                             Specific
                             Exercise      Mobilization   Immobilization
Variable                     (n = 51)      (n = 42)       (n = 28)

Female sex                   23/51 (45%)   23/42 (55%)    18/28 (64%)
History of back pain (b,c)   26/51 (51%)   16/42 (38%)    22/28 (79%)
Presence of leg pain         24/51 (47%)   10/42 (24%)    11/28 (39%)

Variable                       [chi square]     P

Female sex                         2.74        .254
History of back pain (b,c)        10.08        .006
Presence of leg pain               5.35        .069

(a) All values are specific to classification category as reported as
total number (%). Degrees of freedom for all [chi square] tests were 2.
Boldfaced type indicates variable was selected for discriminant
function analysis (P < .20).

(b) Mann-Whitney U post hoc testing indicates significant difference
between specific exercise and immobilization groups.

(c) Mann-Whitney U post hoc testing indicates significant difference
between mobilization and immobilization groups.

Table 4.
Structure Coefficients for Predicted Treatment-Based
Classification Group Membership

Variable                   Function 1 (b)     Function 2 (c)

History of LBP                 0.754              0.293
Duration of LBP                0.631             -0.116
Leg pain with LBP              0.106              0.678
Present pain intensity        -0.454              0.473
Lumbar flexion                 0.244             -0.478

(a) Bold, italicized text indicates structure coefficient
>0.3. LBP = low back pain.

(b) Function that predicts immobilization classification.

(c) Function that predicts specific exercise classification.