Advertisement

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

Mark W Werneke, Dennis L Hart

Abstract

Background and Purpose. Quebec Task Force Classification (QTFC) and pain pattern classification (PPC) procedures, including centralization and noncentralization, are common classification procedures. Classification was done to estimate validity of data obtained with QTFC and PPC procedures for differentiating patient subgroups at intake and for use in predicting rehabilitation outcomes at discharge and work status at 1 year after discharge from rehabilitation. Subjects. Patients (n=171, 54% male; mean age=37 years, SD=10, range=18–62) with acute work-related low back pain referred for physical therapy were analyzed. Methods. Patients completed pain and psychosocial questionnaires at initial examination and discharge and pain diagrams throughout intervention. Physical therapists classified patients using QTFC and PPC data at intake. Patients were classified again at discharge by PPC (time-dependent PPC). Results. Analysis of variance of showed QTFC and PPC data could be used to differentiate patients by pain intensity or disability at intake. Analysis of covariance showed that intake PPC predicted pain intensity and disability at discharge, but QTFC did not. Logistic regression showed that PPC predicted work status at 1 year, but QTFC did not. Classifying patients over time using time-dependent PPC data reduced the false positive rate by 31% and increased percentage of change in pretest-posttest probability of return to work by 16% compared with classifying patients at intake. Discussion and Conclusion. Results support the discriminant validity of the QTFC data at intake and predictive validity of the PPC data at intake. Tracking PPC over time increases predictive validity for 1-year work status.

Classifying patients with nonspecific low back pain into meaningful subgroups is thought to provide assistance for clinical management14 and to increase the power of outcomes assessments1,3,5,6 and has been targeted as an important research priority.7,8 Use of homogeneous subgroups of people with low back pain is considered by many experts to be essential for the improvement of clinical trials related to patient management and clinical outcomes.9,10

Several classification systems have been designed to categorize patients with low back pain into homogeneous subgroups that could guide clinical management decisions or predict pain and disability.4,6,811 Of these classification systems, the Quebec Task Force Classification (QTFC) system11 has received the widest review.1214 Health care professionals using the QTFC procedure classify patients into 1 of 11 diagnostic categories according to presence of pain, anatomical location of pain, presence of neurologic signs, findings from radiological imaging techniques, and surgical history.11 Categories are further subdivided according to pain duration and patient working status. Simpler versions of the QTFC system have been recommended for use by primary care practitioners,12,14 emphasizing anatomical location of pain and results from clinical neurological assessments.

Although the QTFC system is commonly used to classify patients, the predictive validity of data obtained with it is debated. Atlas et al13 reported that changes in pain and perceived disability were associated with QTFC category for patients following nonsurgical management. Patients with spinal nerve root compression with pain below the knee and positive neurologic signs (QTFC category 4) and confirmed by imaging techniques (QTFC category 6) showed more improvement at the 1-year follow-up evaluation compared with similar patients with pain above the knee or pain below the knee but negative neurologic signs (QTFC categories 2 and 3). In contrast, Loisel et al12 reported that patients with pain below the knee with or without neurologic signs (QTFC categories 3 and 4) were more likely to have poorer pain and return-to-work outcomes at 1 year compared with patients with pain but without radiation (QTFC category 1). O'Hearn14 reported that all patients regardless of classification category (QTFC categories 1–4 and 6), using a modified QTFC scheme, reported improvements in perceived disability from initial physical therapist evaluation to discharge.

Investigators examining the prognostic validity of data obtained with the QTFC system for future work-related back troubles based their work on predictive models utilizing medical information,1214 physical examination,1214 and diagnostic imaging studies.13,14 Using this approach, the QTFC system has been identified as a potential predictive factor. However, we believe confidence in prior results12,13 supporting the predictive value of the QTFC system is diminished because of the failure to analyze psychosocial and other physical examination factors.

Psychosocial factors are important predictors in patients with acute low back pain at risk for future work-related disability.1526 Fritz and George,26 for example, reported that fear-avoidance beliefs were associated with work status after 1 month from physical therapy intervention. Biopsychosocial multivariate models are recommended to enhance prediction of occupational low back disability.16,20,2225 Data from one recent study27 supported the QTFC procedure for differentiating patient categories on the basis of intake psychological distress measures. Frank et al27 reported that patients with pain below the knee were more disabled and depressed than patients without pain radiation into the leg. However, we found no biopsychosocial multivariate studies that investigated the predictive validity of data obtained using the QTFC system.

The centralization phenomenon has been reported to be a key physical examination finding in the classification5,6,28 and evaluation and management of patients with spinal impairments.10,2934 McKenzie originally defined centralization as “a situation in which pain arising from the spine and felt laterally from the midline or distally is reduced and transferred to a more central or near midline position when certain movements are performed.”10(p22) The reliability for the clinical documentation of centralization has been shown to be high, with kappa values ranging from .7 to .8.34,35 Data from classifying patients into centralization or noncentralization categories have been shown to be valid for predicting short- and long-term outcomes following rehabilitation.2832,36,37 Patients with centralizing symptoms report better outcomes compared with similar patients without centralizing symptoms. Yet, despite evidence that centralization can be a reliably identified physical examination finding, centralization has not been extensively investigated using biopsychosocial predictive models identifying patients with acute work-related low back pain who are at risk for developing chronic disability.

We found only 2 studies in which the prognostic validity of centralizing symptoms was compared with that of psychosocial variables.6,36 Karas et al36 reported that a high Waddell score was more predictive of return to work regardless of the patient's ability to report centralization of symptoms. In contrast, Werneke and Hart6 reported that a pain pattern classification (PPC) system, including centralization and noncentralization, predicted work status, as determined by telephone interview with the patient 1 year after discharge from physical therapy services, compared with Waddell signs and other psychosocial factors, including fear-avoidance beliefs, depression and somatization symptoms, and high perceived pain and disability ratings.

The PPC system is a method of categorizing patients with low back pain according to the pain they report in response to repeated trunk movements during an initial evaluation28,31,32 or after multiple treatment visits.5,6,28 We defined classification using initial evaluation data as a “one-point-in-time classification.” Classifying patients according to specific anatomical changes in pain location over multiple visits we defined as a “time-dependent classification.” One-point-in-time classification is common,4,9,10,1214,32 but use of time-dependent data has been recommended as a means of improving our understanding of the long-term prognosis of low back pain.38,39 For example, Hunt et al38 and van der Weide et al39 purported that there may be evolving stages of recovery from low back pain and speculated that medical factors may be more predictive than psychosocial factors immediately after the onset of acute back pain. If the patient's pain is protracted, however, psychosocial factors may play a prominent predictive role as medical factors become less prognostic. This hypothesized temporal relationship between physical and psychosocial factors is consistent with Waddell's observation that predictive models may be enhanced by assessing a patient's progress over time versus patient assessment at only one point in time (ie, initial evaluation).40 Data collected using the time-dependent PPC procedure over the episode of therapy were more precise than data collected at the time of intake using one-point-in-time PPC for discriminating pain and disability outcomes following physical therapy intervention.28

Because of the importance of psychosocial issues for predicting recovery following an episode of low back pain, the importance of classifying patients into meaningful subgroups for predicting outcomes, and interest in the validity of classifications of patients obtained using intake or time-dependent data, we conducted this study to assess the validity of the modified QTFC and PPC systems using intake data for the purposes of: (1) classifying patients on the basis of pain and disability at initial evaluation and (2) predicting pain and disability at the time of discharge from rehabilitation and work status 1 year after discharge from rehabilitation. We also used time-dependent data from the PPC to predict work status 1 year after discharge from rehabilitation. The results may help clarify differences between 2 classification procedures and differences between one-point-in-time versus time-dependent classification techniques for clinical practice and research.

Method

Subjects

This is a secondary analysis of a previously described cohort of patients.5,6 The original design was a prospective data collection of 351 consecutive patients between the ages of 18 and 65 years referred for physical therapy with recent onset of nonspecific neck or low back pain and having symptoms of less than 6 weeks' duration. Patients were excluded if they refused to sign a consent form, reported spinal pain or work loss within 6 months before this episode, were unable to complete intake questionnaires, or had poor English proficiency, prior spinal surgery, pregnancy, spinal stenosis, or serious spinal pathology. Fifty-one patients did not meet the admission criteria. For this study, we selected patients (n=171) who were receiving workers' compensation benefits following a work-related low back pain incident with complete data sets for independent and dependent variables (Tab. 1).25,26,4148 Patients were referred by their physicians to 1 of 2 physical therapy outpatient clinics within the same municipality. Five physical therapists participated in the study, and all therapists received advanced training in McKenzie evaluation and treatment methods.

Table 1.

Patient Characteristics (n=171)

The characteristics of the patients are shown in Table 1. At the time of intake (one-point-in-time), 123 patients (72%) reported low back pain without radiation below the gluteal fold (QTFC category 1), 25 patients (14%) reported back pain radiating to above the knee (QTFC category 2), 20 patients (12%) reported pain radiating below the knee (QTFC category 3), and 3 patients (2%) reported distal pain and had at least 2 positive neurological signs. We merged QTFC categories 1 and 2 and QTFC categories 3 and 4 for validity calculations. Intake PPC consisted of 2 classification categories: 77 patients (45%) were classified into the centralization category, and 94 patients (55%) were classified into the noncentralization category. Time-dependent PPC at the time of discharge from rehabilitation consisted of 3 classification categories: 49 patients (29%) were classified into the centralization category, 78 patients (46%) were classified into the partial reduction category, and 43 patients (25%) were classified into the noncentralization category. One patient did not have discharge data for time-dependent PPC identification. For predictive validity of time-dependent data at 1 year, centralization and partial reduction categories were merged because previous research demonstrated no difference in outcomes between these 2 groups.6

Procedure

The procedures used in this study have been described previously.5,6 Briefly, before initial physical therapist examination, patients completed a battery of questionnaires designed to gather information related to medical, demographic, pain, job, and psychosocial factors. In addition, at intake and at discharge, patients completed a pain intensity scale49 and the Oswestry Low Back Pain Disability Questionnaire.41 Body diagrams were completed before and after each visit, including the initial examination, to determine anatomical pain response from mechanical examination.5

After the patients completed intake questionnaires, a mechanical evaluation following McKenzie's assessment methods was done by one of the 5 physical therapists who were credentialed (n=2) or diplomats (n=3) in McKenzie methods.5 Interrater reliability of low back pain assessments by therapists with advanced credentialing in the McKenzie system has been previously reported.33,35 For example, Kilpikoski et al35 reported satisfactory agreement on the relevance of lateral shift (κ=.7), repeated movement tests to define centralization (κ=.7) and directional preference of exercise (κ=.9), and classification into specific McKenzie subgroups (κ=.7). In addition, the following physical examination tests were completed for patients reporting leg pain radiating below the knee: straight leg raise (SLR), knee/ankle/foot manual muscle tests (MMT), light touch for sensation tests, and ankle and knee deep tendon reflex tests. An SLR was considered positive if the patient's familiar calf/foot symptoms were below 60 degrees of leg elevation as measured with a goniometer.50 Manual muscle testing51 of knee extension (L3), ankle dorsiflexion (L4), large toe extension (L5), and ankle plantar flexion (S1) was done, although the reliability of these measurements is questionable. An MMT grade was considered positive if the muscle's score was graded as reduced compared with the uninvolved side according to the therapist's judgment. Light touch sensation testing was performed for the L3 to S1 dermatomes.52 A sensory test was considered positive if light touch was graded as reduced compared with the uninvolved side according to the therapist's judgment. Knee and ankle reflexes were tested using a standard reflex hammer and were judged as positive tests if graded as absent or reduced compared with the uninvolved side.

Patients were classified at 2 different times: at the time of intake and at the time of discharge. At the time of intake, evaluating physical therapists classified patients (first classification process) by determining if patient symptoms were centralized (centralization category) or were not centralized (noncentralization category), and patients were classified (second classification process) using the QTFC categories. There were no patients in QTFC categories 5 through 1111 because the inclusion criteria excluded these patients. The remaining 4 categories represent a method of classification based on pain location and clinical examination of neurological signs (ie, motor, sensory, and reflex). For our study, we followed a truncation recommended by Loisel et al12 for QTFC categories. Loisel et al recommended using the first 4 QTFC categories for patients without surgery who were evaluated during the early stage of nonserious back pain. The QTFC categories 5 to 11 were excluded because of the study's inclusion criteria. The QTFC categories 1 and 2 were combined and QTFC categories 3 and 4 were combined, producing a dichotomous classification system based on whether or not pain radiated below the knee. Patients were classified (third classification process) at the time of discharge from rehabilitation into groups on the basis of having one of 3 anatomical pain patterns (centralization, noncentralization, and partial reduction) after multiple treatment visits,5 which we defined as time-dependent PPC.

The evaluating physical therapist treated each patient. If the evaluating therapist's schedule was changed unexpectedly, another physical therapist participating in the study may have treated the patient. Exercises, manual techniques, and cognitive-behavioral educational strategies53 were provided as deemed necessary by the treating physical therapist and are described elsewhere.5 Our study was designed to assess discriminant and predictive validity of data obtained with the 2 classification procedures and, therefore, there was no attempt to standardize or influence care across patient classification procedures.

Outcome Measures

Pain intensity and perceived disability were assessed at the time of intake and at the time of discharge from rehabilitation. Maximal pain intensity experienced during the preceding 24 hours was assessed using an 11-point numeric pain scale: 0 (no pain) to 10 (severe emergency-department-type pain).49,54 The 11-point pain scale has been shown to yield reliable and valid measurements of pain intensity.49,54 Low back-related disability was assessed using the 10-item Oswestry Low Back Pain Disability Questionnaire.41 The disability score is expressed as a percentage, with higher scores representing more disability. Data from the Oswestry questionnaire have been shown to have good test-retest reliability41 and predictive validity.25,42 In the original study,41 22 patients with chronic low back pain completed the Oswestry questionnaire on 2 consecutive days, producing a correlation coefficient of .99. Cooper et al55 measured change in Oswestry questionnaire scores between the time of injury and a 6-month follow-up evaluation and reported that high disability at the time of injury was associated with high disability at the 6-month follow-up evaluation (P<.01). Tate et al42 reported that perceived disability as measured with the Oswestry questionnaire predicted (P<.001) duration of time loss due to back symptoms and predicted future lost work time. Nordin et al25 reported that Oswestry scores greater than 40 out of 100 predicted delayed return to work in patients with serious functional disability (odds ratio=1.40, 95% confidence interval [CI]=1.05–1.88, P<.02).

Work status was assessed 1 year after discharge from rehabilitation. Work status was considered good if the employee was working full-time at full duty. A poor outcome was defined as when a previously full-time employee was currently working less than full-time at full duty because of the low back pain problems for which the patient was managed. An occupational nurse who was experienced in conducting structured telephone interviews and who was masked (blinded) to classification categories and response to intervention called all patients to assess work status at 1 year after discharge from rehabilitation.

Data Analysis

To accomplish the 3 study purposes, several sets of analyses were conducted. For purpose 1, validity of QTFC and PPC was assessed using intake data to differentiate patients by pain intensity (0–5=low, 6–10=high)56 and disability (Oswestry questionnaire scores of 0–39=low and 40–100=high)25,41,42 at initial evaluation by calculating one-way analyses of variance (ANOVAs). We assessed discriminant validity at the time of intake by determining differences for the dependent variables pain intensity and perceived disability across PPC and QTFC categories. Power (power=1−β) analyses were performed on all ANOVA results if the results of one ANOVA were not significant.57 For all analyses, α=.05.

Our ability to use intake classification data to predict which patients would have high pain or disability at the time of discharge from rehabilitation was assessed by determining differences in dependent variables across categories of classification procedures using one-way analyses of covariance (ANCOVAs). Pain intensity at the time of discharge was compared across QTFC and PPC categories using intake pain intensity as the covariate. Perceived disability was compared across classification categories using intake perceived disability as the covariate. Power (power=1−β) analyses were performed on all ANCOVA results if the results of one ANCOVA were not significant.57

For purpose 2, our ability to use intake data to predict which patients would have less than optimal work status 1 year after discharge from rehabilitation was determined using 3 sets of analyses. First, the relationship of each of the 22 independent variables at the time of intake (Tab. 1) to work status was assessed with univariate analyses. Two-sample t tests were used to compare continuous independent variables and work status, and chi-square tests of independence were used to compare categorical independent variables and work status.

Second, independent variables (Tab. 1) related to poor work status assessed by the univariate analyses were entered into a complete multivariate logistic regression model to assess work status.58,59 We used the Hosmer-Lemeshow summary goodness-of-fit statistic to assess fit of the model to the data. Higher probability values indicate better fit.59 Likelihood ratio (LR) chi-square and McFadden rho statistics were calculated for the logistic model. A t-ratio (regression coefficient divided by associated standard error) and odds ratio with 95% CIs were calculated for each independent variable in each final logistic model.59

Third, we examined independent variables from the logistic regression analyses for their ability to predict work status by calculating sensitivity, specificity, positive and negative likelihood ratios (+LR, −LR), and positive and negative predictive values (PPV, NPV).60 To calculate sensitivity and specificity, a 2×2 contingency table was used. Patients who were unable to return to work full-time at full duty formed the disease-positive group of the target disorder. Patients returning to full-time, full-duty work formed the disease-negative group of the target disorder. Patients with pain below the knee (QTFC category 3 or 4) or whose symptoms were not centralized (PPC noncentralization) formed the diagnostic test-positive group. Patients with pain above the knee (QTFC category 1 or 2) or whose symptoms were centralized (PPC centralization) formed the diagnostic test-negative group. Sensitivity is the proportion of patients with the target disorder (ie, less than optimal work status) who have positive test results (ie, pain below the knee or noncentralized).60,61 Specificity is the proportion of patients who do not have the target disorder (ie, return to work without restrictions) and who have a negative test result (ie, no pain below the knee or centralized).60,61

Positive likelihood ratios were calculated as sensitivity/ 1–specificity, and −LRs were calculated as 1–sensitivity/specificity.60 As described elsewhere,62 LRs are summary measures of diagnostic test performance (ie, classification) that indicate how much a given classification will raise or lower the pretest probability of the target disorder of interest (ie, work status).60,61,63 Following a published guide,64 acceptable +LRs are 2 or more and acceptable −LRs are 0.5 or less because they generate at least small, but possibly important, changes in predictive value of the test. Positive predictive value is defined as the probability of having the target disorder when the test result is positive.60,61 Negative predictive value is defined as the probability of absence of the target disorder if the test result is negative.61 The PPV and NPV are affected by sensitivity, specificity, and classification prevalence. Prevalence, which is equal to pretest probability, was calculated as the number of patients with the target disorder divided by all patients tested.60 The higher the +LR, the more predictive a positive test will be for a given prevalence. Absolute values of −LR will increase with diminishing discriminative power of patient classification,61 so the smaller the −LR value, the higher the negative predictive value for a given prevalence.61 The 95% CIs were calculated for sensitivity, specificity, +LR, −LR, PPV, and NPV.65

The diagnostic accuracy of independent variables from the final logistic regression analysis was considered acceptable if: (1) either +LR was 2 or more or −LR was 0.5 or less64 and (2) the posttest probability was 15% or more. On the basis of pretest probability for a poor return-to-work outcome of 15% in this cohort of patients, +LR values of 2 or more and −LR values of <0.5 would result in a posttest probability change of approximately 10%.66

For purpose 3, ability of time-dependent PPC data to predict work status 1 year after discharge from rehabilitation was assessed using the same sets of analyses described above for purpose 2 except for one change: one-point-in-time PPC was supplanted with time-dependent PPC assessed at the time of discharge from rehabilitation (Tab. 1).

Results

Discriminant Validity of Intake Patient Classification

For one-point-in-time analyses at the time of intake, QTFC was used to differentiate patients on the basis of pain intensity and disability, and PPC was used to differentiate patients on the basis of disability (Tab. 2). Only PPC classification procedure predicted pain intensity and disability at the time of discharge from rehabilitation (Tab. 3).

Table 2.

Classification Systems for Differentiating Patients on the Basis of Pain and Disability at the Time of Initial Evaluation

Table 3.

Classification Systems for Predicting Pain Intensity and Disability at Time of Discharge From Rehabilitation

Contacted Patients Versus Noncontacted Patients Analyses

Of the 171 patients selected, 136 patients (80%) were contacted by telephone, 4 patients refused to be interviewed, and 132 patients were successfully interviewed (77% follow-up rate). Of all of the independent variables (Tab. 1), only one variable was different between groups: contacted patients were older than noncontacted patients (mean years of age=39 [SD=10] versus mean years of age=33 [SD=9], t=3.8, df=78.1, P=.001).

Predictive Validity of One-Point-in-Time Data at 1 Year After Discharge From Rehabilitation

The results of the univariate analyses are displayed in Tables 4 and 5.25,26,4148 Four independent variables affected work status: multiple sites of pain, high pain intensity, high fear-avoidance of work activities, and noncentralizing symptoms. The QTFC categories were not related to work status. Results of the logistic regression analyses demonstrated that overall fit of the model was supported (Hosmer-Lemeshow goodness-of-fit statistic59 for complete model=.17, df=2, P=.92; McFadden rho67=.12). In the final model, only intake PPC predicted work status at 1 year after discharge from rehabilitation (standardized t-ratio coefficient=2.8, P=.005; maximum likelihood-ratio statistic=12.2, df=1, P<.001).59 Patients classified as having noncentralized symptoms were almost 9 times more likely not to return to work (odds ratio=8.8, 95% CI=1.9–40.1). Findings for accuracy of PPC for predicting work status statistics were as follows: sensitivity=0.89 (95% CI=0.69–0.97), specificity=0.51 (95% CI=0.42–0.60), +LR=1.82 (95% CI=1.42–2.34), −LR=0.21 (95% CI=0.06–0.28), PPV=0.25 (95% CI=0.16–0.36), and NPV=0.96 (95% CI=0.88–0.99). A patient demonstrating a lack of centralization during initial evaluation (15% prevalence) produced a pretest-posttest probability change of 9%.

Table 4.

Univariate Tests for Continuous Intake Variables, With Work Status at 1 Year After Discharge From Rehabilitation as Dependent Variable

Table 5.

Univariate Tests for Categorical Variables, With Work Status at 1 Year After Discharge From Rehabilitation as Dependent Variable

Predictive Validity of Time-Dependent Data at 1 Year After Discharge From Rehabilitation

In addition to univariate analyses of intake data (Tabs. 4 and 5), time-dependent PPC results are displayed in Table 5. Time-dependent PPC assessed at discharge was added to multiple sites of pain, high pain intensity, and high fear-avoidance of work activities assessed at the time of intake for the logistic regression analysis.59,67 Overall fit of the model was supported (Hosmer-Lemeshow goodness-of-fit statistic59 =2.5, df=3, P=.48, McFadden rho67 =.18). In the final model, only time-dependent PPC predicted work status at 1 year after discharge from rehabilitation (standardized t-ratio coefficient=4.1, P<.001; maximum likelihood-ratio statistic=18.8, df=1, P<.001).59 Patients classified as having noncentralization of pain were almost 10 times more likely not to return to work (odds ratio=9.9, 95% CI=3.3–29). Findings for accuracy of PPC for predicting work status statistics were as follows: sensitivity=0.68 (95% CI=0.46–0.85), specificity=0.82 (95% CI=0.74–0.88), +LR=3.82 (95% CI=2.29–6.35), −LR=0.38 (95% CI=0.20–0.75), PPV=0.41 (95% CI=0.26–0.58), and NPV=0.94 (95% CI=0.87–0.97). A patient demonstrating no centralization of symptoms at the time of discharge (15% prevalence) produced a pretest-posttest probability change of 25%.

Discussion

Classifying patients with acute, occupational, and nonspecific low back pain syndromes by pain above or below the knee (QTFC categories 1–4)12 can be used to identify patients who have high or low pain intensity or perceived disability at the time of initial evaluation. Intake QTFC categories as grouped in this study were not predictive of pain intensity or disability at the time of discharge from rehabilitation or of work status 1 year after discharge from rehabilitation. Intake pain pattern classification28 differentiated patients by perceived disability, but the primary value of one-point-in-time PPC was in predicting pain and disability at the time of discharge and work status 1 year after discharge from rehabilitation. The predictive value of PPC increased when classification was followed over the rehabilitation episode. Our data supported the idea that not only is anatomical location of pain important for differentiating patients on the basis of disability at the time of intake, but a change in anatomical location of pain following clinician-directed examination procedures was predictive of future pain intensity, disability, and work status in this sample. The results of our study contribute to the existing literature by: (1) clarifying differences between 2 classification systems, (2) supporting literature recommending assessment of change in anatomical location of pain during patient examination, and (3) supporting time-dependent data3840 and patient classification procedures6,28 as stronger predictors of work status than the same data or classification assessed at one point in time.

Using the presence of pain above or below the knee at the time of initial evaluation for differentiating patients on the basis of baseline disability is consistent with the findings of previous studies.12,13,27 Our results do not support the use of location of leg pain for predicting pain intensity and disability at the time of discharge from rehabilitation. This finding is similar to the findings of O'Hearn,14 who reported decreases in perceived disability at the time of discharge from rehabilitation for all patients with acute or subacute symptoms regardless of baseline modified QTFC categories 1 through 4 and 6.

Pain below the knee has been reported to be an important predictor of poor outcomes in people with low back pain.12,18,39 Loisel et al,12 for example, reported that patients with pain below the knee during initial evaluation (QTFC categories 3 and 4) were less likely to return to regular work compared with patients with back pain without radiation (QTFC category 1). In contrast, pain below the knee with or without positive neurological signs (QTFC categories 3 and 4) was not predictive of work loss at 1 year after discharge from rehabilitation in our study. There are 2 reasons that might explain the poor predictive validity for leg pain. First, back and leg pain were grouped using a dichotomous classification, which might have decreased precision secondary to lost information. Patients in our study, however, were referred for physical therapy early with acute low back pain, so QTFC categories 5 through 11 would not be expected for this sample unless radiological imaging was required to rule out serious pathology. In addition, few patients in our sample reported pain below the knee with neurological signs, which supports findings from other studies.12,14,27 Second, in only one prior study6 were one-point-in-time PPC and leg pain as independent variables used in a predictive biopsychosocial multivariate model. In that study, lack of centralization during clinical evaluation improved the likelihood of poor outcomes following conservative intervention regardless of distal leg pain location. Anatomical location of pain at the time of initial evaluation (QTFC categories 1–4) lost predictive power when entered into multivariate biopsychosocial predictive models, but change in anatomical location of pain in response to standardized repeated lumbar movement tests did not.6

Our results allow comparison of the predictive accuracy of the PPC system determined at one point in time versus classifying patients over the rehabilitation episode (ie, time-dependent classification). In our sample, one-point-in-time PPC had relatively high sensitivity (95% CI=0.69–0.97), acceptable −LR (95% CI=0.06–0.28), and high NPV (95% CI=0.88–0.99). One-point-in-time PPC produced a modest 9% change in pretest-posttest probability of return to work given a low (15%) prevalence of patients not working at 1 year after discharge from rehabilitation. The result of any clinical test (eg, repeated trunk movements) can be interpreted as an argument to strengthen or weaken conviction of prediction of the target disorder (eg, return-to-work status) based on the available information on the patient.61 In our study, PPC had good sensitivity, so if the clinician finds that his or her patient's symptoms are centralized, poor work status could be effectively ruled out.60

One purpose of the PPC procedure is to identify patients at risk for poor work status at 1 year after physical therapy intervention. Targeting patients who might have difficulty returning to work for costly comprehensive multidisciplinary interventions designed to prevent future disability during the acute phase of pain could be advantageous while avoiding unnecessary and costly interventions for patients who are likely to return to work easily. Subsequently, reducing false positive results is beneficial. The false positive rate (ie, probability of noncentralization given a patient who returns to work) for one-point-in-time PPC was 49%.61 The −LR, which in our study was good, expresses how many times less likely a normal test result (ie, symptoms centralized) is to be expected in patients who do not return to work as compared with patients who return to work.61 The smaller the −LR, the higher the negative predictive value of PPC for a given prevalence.61 Thus, although sensitivity, −LR, and NPV were adequate, the false positive rate was not. A high false positive rate may cause unnecessary testing or intervention.61

Time-dependent PPC had relatively high specificity (95% CI=0.74–0.88), modest sensitivity (95% CI=0.46–0.85), acceptable +LR (95% CI=2.29–6.35), acceptable −LR (95% CI=0.20–0.75), and high NPV (95% CI=0.87–0.97). The time-dependent PPC identified a 25% change in pretest-posttest probability of return to work given the same low (15%) prevalence of patients whose symptoms were not centralized and on whom we had return-to-work data. Results of time-dependent patient classification following testing of repeated trunk movements are clearly more predictive of 1-year work status compared with one-point-in-time classification.61 A higher specificity and reduced false positive rate (18%) were found. When specificity is high, a positive result (ie, symptoms do not centralize) effectively rules in poor work status.60 If one considers that the +LR can be interpreted as a cost-benefit ratio with the numerator or true positive rate (68%) representing a benefit criterion and the denominator or false positive rate (18%) representing cost,61 time-dependent PPC appears to be a better clinical tool to direct intervention than one-point-in-time PPC if work status is of interest.

In contrast to previous studies,6870 our data do not support psychosocial factors as important predictors of future work-related disability. There is a consensus among experts that psychosocial factors are better than medical factors for explaining chronic low back pain and disability.16,20,22,23,71 Despite this popular belief, we believe this is an oversimplification. Both physical and nonbiological factors may play important predictive roles depending on when the predictive model is applied during the course of back pain.38,39 Our research suggests to us that physical factors (ie, changes in anatomical pain location in response to repeated trunk movement testing) are better predictors of future disability than are psychosocial variables measured during the acute phase of back pain. The patient's responses to physical clinical examination tests, however, may be affected by psychosocial influences.72 The ability to use centralization for prediction as analyzed by biopsychosocial multivariate models for subacute and chronic low back pain and disability warrants future investigation.

Limitations

The noncontacted group in our study was younger than the contact group. We do not believe that the age difference between the 2 groups was a factor in our results. Although the difference was significant, we believe it was small enough to be considered clinically unimportant. Both groups were in their fourth decade of life, and the difference between groups was 6 years. There is no consensus on the importance of age as a predictive factor, and researchers1820,73 have reported conflicting affects of age on disability and work status.

We investigated anatomical pain patterns utilizing a prospective cohort design. Such a design, we contend, is optimal for examining a diagnostic test (eg, centralization) and its relationship to the reference standard (ie, return-to-work status). However, randomized control trials are required to elucidate whether the results of classifying patients based on pain patterns lead to more effective interventions. Effects of exercises such as those proposed by McKenzie10 have not been rigorously investigated,74 and effects of exercises prescribed for specific patient classifications are still under review.1,3 Whether general exercises can be designed to modify patients' beliefs concerning pain and activity following an acute episode of low back pain also warrants investigation.

Generalizability of the PPC model for patients with chronic low back pain has not been investigated. In addition, in our study, QTFC categories 3 and 4 were merged because only 3 patients were classified into QTFC category 4. The small number of patients in the QTFC category 4 limits our results for patients with positive neurological findings. However, the predictive role of neurological clinical signs for identifying patients with low back pain who are at risk for future work-related disability can be questioned.22,23,25,71 Future research is needed to investigate centralization classification models for patients with either chronic low back pain syndromes or neurological deficits.

The predictive validity of the modified QTFC system investigated in our study may have been affected by the dichotomous subgroupings we analyzed (ie, QTFC categories 1 and 2 and categories 3 and 4 were combined). However, we chose these 2 subgroups based on Loisel and colleagues' QTFC stratification recommendations for patients in primary care.12 In addition, these subgroups appear logical in light of previous research suggesting that patients with sciatica (ie, pain below the knee) are at risk for poor treatment outcomes.18,39

Summary and Conclusion

When clinicians want to predict pain and disability at the time of a patient's discharge from rehabilitation or a patient's long-term work status, classifying the patient according to change in anatomical location of pain over the treatment episode is more predictive than classifying the patient by anatomical location at one point in time or classifying the patient with pain above or below the knee during an acute episode of low back pain.

Footnotes

  • Both authors provided concept/idea/research design and writing. Mr Werneke provided data collection, project management, subjects, facilities/equipment, institutional liaisons, and clerical support. Dr Hart provided data analysis and consultation (including review of manuscript before submission).

  • Received May 15, 2003.
  • Accepted October 17, 2003.

References

View Abstract