Volume 89, Issue 8 , Pages 1542-1549, August 2008
Classifying Subgroups of Chronic Low Back Pain Patients Based on Lifting Patterns
Article Outline
Abstract
Slaboda JC, Boston JR, Rudy TE, Lieber SJ. Classifying subgroups of chronic low back pain patients based on lifting patterns.
Objective
To compare self-reported measures of chronic lower back pain (CLBP) patients who were assigned to 2 subgroups based on their lifting patterns performed during a repetitive lifting task.
Design
Cross-sectional study.
Setting
Research laboratory
Participants
CLBP subjects (n=81) and pain-free controls (n=53).
Interventions
Not applicable.
Main Outcome Measures
Measures of lifting patterns and self-reported disability, pain, and psychosocial aspects.
Results
Two CLBP subgroups were found: 1 group that lifts similarly to control subjects (n=35) and 1 group that lifts very differently from controls (n=46). The CLBP group that lifted differently than controls reported higher pain intensity (P=.005), higher pain severity (P=.025), and lower self-efficacy (P=.013) than the CLBP group that lifted similarly to controls.
Conclusions
A classification system based on lifting patterns identified 2 CLBP subgroups that were significantly different on lifting and self-reported measures, indicating the importance of physical functioning measures in classification systems.
Key Words: Lifting, Low back pain, Rehabilitation
List of Abbreviations: ANOVA, analysis of variance, CLBP, chronic low back pain, MANOVA, multivariate analysis of variance, MEDICS, Medical Evaluation and Diagnostic Information Coding System, MPI, West Haven–Yale Multidimensional Pain Inventory, ODI, Oswestry Disability Index
PAIN PATIENTS TOO OFTEN are assumed to be a homogeneous population when in fact pain impacts these patients in different ways. By identifying differences and designing diverse treatment options based on these differences, treatment effectiveness may improve. For example, Delitto et al1 identified subgroups of low back pain patients. When the patients entered a treatment program specialized to treat the symptoms of the subgroup, the group had greater return to work status and lower self-reported disability compared with patients who received a standardized treatment protocol.2
Subgroups of chronic pain patients have been identified based on the patients' responses to self-reported measures such as the MPI,3, 4, 5 disability, self-efficacy, pain intensity, fear of movement, and catastrophizing,6, 7 and clinical examinations such as pain during specific exercises.8, 9 These classifications provide important information about the differential psychologic impact of pain in chronic pain populations, but ignore the differential impact of pain on physical functioning. For instance, the dysfunctional group from the MPI classifications reported difficulty performing activities of daily life, but no information about the patients' physical performance during these tasks was provided.5 Because physical functioning and self-report measures have been suggested to be distinct domains,10 identifying classifications based on physical functioning could provide additional information on multidimensional impacts of pain that could lead to diverse and more effective treatment programs.
The classification system described in this article focuses on identifying subgroups based on physical functioning of CLBP patients. Motion differences between CLBP subjects and pain-free control subjects have been widely studied with differences found in flexibility,11 spinal loading,12 muscle activation,13, 14 segmental motion,15 jerk,16 and lifting strategies over time.17 For example, Rudy et al17 found that controls and CLBP subjects show different changes in lifting parameters over the duration of a repetitive lifting task. These studies have shown that CLBP subjects and controls show different motion patterns during tasks such as lifting, but little research attention has been focused on identifying motion differences among CLBP groups.
We recently developed a classification procedure that uses hidden Markov models to assign CLBP subjects to a subgroup based on their lifting patterns performed during a repetitive lifting task.18, 19, 20 Hidden Markov models are statistical models that have been used extensively to describe time series data such as speech. The models, which include several states and an observed sequence of output symbols (the time series), are doubly stochastic, where one stochastic process describes state transitions (which are not observable) and a second set of processes produces the sequence of observed symbols.21 The particular symbol observed depends on the current, hidden state of the model. In classification studies, multiple hidden Markov models are designed, with each hidden Markov model describing a specific classification group. The probability that an individual time series or sequence of observations was produced by a particular model can then be computed and the sequence classified to the model with the highest probability. For this study, we have designed a hidden Markov model that describes the lifting patterns used during the lifting task by CLBP subjects and a hidden Markov model that describes the lifting patterns used by control subjects. Two groups were identified: the CLBP subjects who lifted similar to control subjects and the CLBP subjects who lifted differently from control subjects. The purpose of this study was to determine whether the CLBP subgroups identified by the classification procedure show differences on measures other than lifting.
Methods
We obtained the data used for this study from a database that contained lifting parameters, medical findings, and self-reported measures collected during a clinical study conducted at the University of Pittsburgh Medical Center Pain Evaluation and Treatment Institute.15, 16, 17 During the clinical study, subjects completed a medical evaluation, a psychologic evaluation, and a repetitive lifting task. Data from the lifting task were used to define the hidden Markov models that were used to classify the patients into 2 subgroups, and the measures from the 2 evaluations were compared between the CLBP subgroups to determine whether the groups were different on measures other than lifting patterns.
The methods section is separated into 4 parts. The first part describes the subjects recruited for the study, the second part describes the clinical study protocol, the third part describes the hidden Markov model classification procedure, and the fourth part describes the statistical analyses.
Participants
Eighty-one CLBP subjects and 53 control subjects participated in the clinical study. Subjects were defined as CLBP subjects if the subject reported having pain every day or almost every day for the past 3 months that was of moderate or greater intensity. Control subjects were defined as having no pain and no history of back pain. The control subjects were recruited from the community, and athletes and college students were excluded. The CLBP subjects were recruited from a treatment group at the University of Pittsburgh Medical Center Pain Evaluation and Treatment Institute. Before the start of the treatment program, patients were asked if they were interested in participating in the clinical study. All subjects gave written informed consent as approved by the University of Pittsburgh Biomedical Institutional Review Board before the start of the study. The ages of the subjects ranged from 36 to 63 years and the average age ± SD of CLBP subjects was 37.8±10.1 years. Thirty-eight of the CLBP subjects were men and 43 of the subjects were women. All of the 81 CLBP subjects had a history of prolonged back pain with mean pain duration of 4.1±5.4 years.
We used a standardized protocol to determine whether the control and CLBP subjects were able to participate in the repetitive lifting task. During the medical evaluation, the examining physician asked the subject to lift an item from 30cm above the floor to waist height with both hands using a leg lift. The subject was asked to complete 5 lifts. If the subject was successful, he/she was included in the study. Because the initial medical evaluation and the lifting task often were separated by several weeks, a physical therapist conducting the lifting task also used the same standardized protocol to ensure the subject's ability to perform the motion required. No subject, control, or CLBP was excluded based on this protocol.
Clinical Study Protocol
The clinical study protocol was separated into domains of medical, psychosocial, cognitive, disability, spinal mobility, task self-efficacy, and the repetitive lifting task.
Medical domainThe 2 measures collected for the medical domain are described below. In one, a standardized scoring system (MEDICS) that assigns linear weights to medical procedures according to their relevance in diagnosing chronic pain patients was used. Twenty-three medical procedures were listed in the MEDICS form and diagnosis of each procedure was assigned a score by the examining physician.22 In the other, the body mass index of the CLBP subjects was calculated.
Pain domainTwo measures were used to assess each CLBP subject's perception of his/her pain: in one, for pain severity, we used the Pain Severity Scale from the MPI3, 4; and in the other, for pain intensity, we used the pain intensity questions from the Jan van Breemen examination.23
Psychosocial domainThe psychosocial domain consisted of the MPI, which assessed pain-relevant psychosocial aspects and responses of the significant other to the patient's pain.3, 4, 5 CLBP subjects completed the 3 sections of the MPI. The first section contained 5 subsections that assess the pain patients' (1) report of pain severity, (2) perceptions of how pain interferes with their lives, (3) appraisals of the amount of support received from significant others, (4) perceived life control, and (5) affective distress. The second section of the MPI assessed the responses of significant others to the patients' pain with 3 subsections: (1) punishing response, (2) solicitous responses, and (3) distracting responses. The subjects' responses to the MPI questions were analyzed with the MPI computer program, version 3.24 The MPI program computes 2 higher order composites scores, dysfunctional and interpersonal distress, based on the scores from the MPI scales. The MPI composite scores were based on factor analysis of 6545 heterogeneous chronic pain patients.
Cognitive domainThe items represent common negative cognitive appraisals and self-statements. Subjects rate the frequency with which they have each of these thoughts and feelings when their pain is particularly severe. Two scale scores are derived, emotionality and worry. The Pain Experience Scale is a 19-item self-report questionnaire.25
Disability domainWe used 5 measures to assess the disability domain as described below. First, gait speed: subjects were requested to walk at a comfortable speed over a 15-m (50-ft) expanse of a level floor. The time in seconds was recorded. Second, the ODI: a 10-item questionnaire that measured the patient's perceived level of disability due to pain during activities of daily life.26 Third, functional status: the Jan van Breemen examination23 was used to assess functional status. Subjects rated how well during the past week they were able to perform activities of daily life. Fourth, pain behavior checklist: the pain behavior checklist asked subjects the frequency that they performed certain pain behaviors.27 Fifth, general activities from MPI: in the general activities section of the MPI, CLBP subjects were asked how often they participated in 19 common activities.3, 4, 5
Spinal mobility domainA physical therapist collected flexion and flexion/extension measures using the methods described in the Jan van Breemen examination.23 The assessment was performed only on CLBP subjects because the measure was developed and geared to the CLBP patient.
Task self-efficacy domainA task-specific self-efficacy questionnaire asked subjects to rate on a scale ranging from 0% (very uncertain) to 100% (very certain) how confidently they felt they could perform the lifting task for a period of time.16 The lifting task was described to the subjects and they viewed the equipment and the evaluator simulating a lifting pattern before giving a rating. The questionnaire was completed before the subjects performed the repetitive lifting task.
Repetitive lifting taskThe protocol of the repetitive lifting task has previously been described.15, 16, 17 Briefly, the task required each subject to lift a resistant load attached to a handle located 33cm from the floor to waist height. The resistant load was equal to 40% of the subject's mean voluntary static strength. The 40% level of static strength was selected because it is the minimum amount of resistance necessary to show fatigue on electromyography recordings and because it would enable the subjects to perform for a sustained period of time.28, 29 A force gaugea attached to a platform was used to measure voluntary static strength. Subjects were instructed to assume a bilateral symmetrical leg lift position with the forearm in supination and the handle of the force gauge adjusted to knee height. The subject was then instructed to pull steadily on the force gauge for approximately 4 seconds. This process was repeated 3 times with a 15-second rest period between each attempt, during which the subject was instructed to return to a standing position. We used 40% of the mean of the 3 trials as the resistant force in the repetitive lifting task.
The BTE Work Simulatorb provided resistant load during the up-phase of the lift. The BTE Work Simulator also provided signals that indicated the start and end of the lift. Reflective markers placed on the ankle, apex of the patella, greater trochanter of the femur, and acromion of the shoulder were used with a Motion Analysis Model 110c to track the markers at a sampling rate of 30 frames per second. The motion of these markers was used to define changes in body angles as functions of time during each lift. Subjects performed lifts for a maximum of 20 minutes with a 15-second rest interval between each of the lifts. Subjects were instructed to lift until they felt physically unable to continue or until the time limit was reached, and they were given no visual or verbal feedback during the task. CLBP subjects were asked to rate their pain intensity before the start of task (baseline rating), after the static strength task, and at the end of the dynamic lifting task. The pain intensity rating is a self-reported measure that asks the subject to rate their pain on a scale from 0 (no pain) to 10 (extreme pain).
Hidden Markov Model Classification Procedure
We classified CLBP subjects to a subgroup based on lifting patterns that were performed during the repetitive lifting task. The lifting patterns were derived from the lifting parameters that describe body posture, timing of body angle motion, and force of each lift. Because the task was performed for a maximum of 20 minutes and the parameters were calculated for each lift performed during the task, a multidimensional time series of lifting parameters was obtained for each subject. Each of the lifting parameters has previously been described and has been shown to be significantly different between controls and CLBP subjects.15, 16, 17
A data reduction procedure was applied to the lifting parameters to reduce redundancy and to combine the multidimensional parameters into discrete lifting patterns.18, 19, 20 The procedure used factor analysis to reduce the parameters into 4 factors and cluster analysis to assign the 4 factor scores of each lift to 1 of 5 clusters, determined using the statistical methods of Milligan and Cooper.30, 31 The subject's lifting sequences were the sequence of cluster assignments for each lift, where each of the 5 clusters described a different lifting pattern. The 5 lifting patterns were: (1) slow, low jerk lift, (2) squat starting posture lift, (3) fast, high jerk lift, (4) torso starting posture lift, and (5) 2 segment lift (lower body moves faster than the upper body).18, 19, 20
Hidden Markov models were designed using the sequence of cluster assignments as the observable outputs for each state. A temporal 3-state hidden Markov model was designed and trained using the lifting patterns of all control subjects and a temporal 2-state hidden Markov model was designed and trained using the lifting patterns of all CLBP subjects. These models are shown in figure 1. Temporal models are hidden Markov models that contain only forward transitions and no backward transitions, for example, transition from state 1 to state 2 but not from state 2 to state 1.
Fig 1.
Diagrams of the CLBP hidden Markov model (HMM) and control hidden Markov model.
To determine whether these hidden Markov models could identify subgroups of CLBP subjects, we performed a simulation study to assess reliability of the classifications. The simulation study mimicked the clinical problem of identifying subjects as belonging to 1 group (controls) when they had been assigned to the other group (CLBP group). The methods of the simulation study have been previously described.18, 19, 20 Briefly, simulated lifting sequences were generated from the control and CLBP hidden Markov models. Several of the simulated lifting sequences were then intentionally mislabeled to the wrong group, creating a simulated CLBP sample that contained some control simulated sequences and a simulated control sample that contains some CLBP simulated sequences. Two hidden Markov models were trained with the simulated samples and individual sequences were classified to either the simulated CLBP hidden Markov model or the simulated control hidden Markov model using the leave-one-out technique.18, 19, 20 Several variations on the model structures were evaluated, and the 2-state CLBP and 3-state control hidden Markov models provided the highest number of correct classification. These 2 hidden Markov models identified 82% of the simulated sequences to the correct group; for example, the CLBP simulated sequences that were mislabeled to the control group were classified as CLBP simulated sequences. These 2 hidden Markov models were then used to classify the CLBP subjects into subgroups.
Each subject within the 81 CLBP group was classified to either the control hidden Markov model or the CLBP hidden Markov model based on the model that best fit that subject's lifting sequence. To assess the consistency of the classification procedure, it was applied to a group of 53 control subjects. Eighty-seven percent of these subjects were classified to the control hidden Markov model, suggesting that the classification procedure can identify lifting sequences to the appropriate hidden Markov model. When the classification procedure was applied to the CLBP subjects, approximately half were classified to the control hidden Markov model and half were classified to the CLBP hidden Markov model.18, 19, 20 The 2 groups of CLBP subjects were studied to identify differences in measures other than lifting.
Statistical Analyses
The hypothesis of the study was that CLBP classified to the control hidden Markov model will be different from the CLBP subjects classified to the CLBP hidden Markov model for one or more of the measures other than the lifting patterns. Before we assessed the comparisons of the self-report measures between the 2 CLBP subgroups, we compared the demographics of the CLBP subjects classified to the control hidden Markov model and the CLBP subjects classified to the CLBP hidden Markov model with ANOVA models and chi square statistics. If any of the demographics were significantly different between the 2 groups, the variable was treated as a covariate.
We used a MANOVA model to test that the CLBP groups were significantly different for the psychosocial, pain, medical, cognitive and lifting domains. The only exception was the self-efficacy domain that contained a single measure, for which an ANOVA was performed to assess subgroup differences. If the P value of the MANOVA was significant, follow-up ANOVAs were performed on the individual measures within the domains. A stepwise discriminant function analysis was performed to identify the measures that best separated the CLBP groups.
CLBP subjects were asked to rate their pain intensity before testing (baseline), after the static lifting task, and at the end of the repetitive lifting task. Differences in the pain intensity ratings between the 2 CLBP subgroups were assessed using repeated measures ANOVA. A P value of .05 or less was considered statistically significant in all statistical analyses.
We calculated an effect size index to compare the magnitude of the differences between the 2 CLBP groups on each of the domains. Effect size is calculated as the difference in the means divided by a pooled SD. An effect size range of 0 to .32 is small, .33 to .55 is medium, and .56 and greater is large.32
Results
Thirty-five of the 81 CLBP subjects were classified to the control hidden Markov model and 46 CLBP subjects were classified to the CLBP hidden Markov model. The CLBP subjects that were classified to the CLBP hidden Markov model were labeled as guarded CLBP lifters because these subjects performed a slow, low jerk lifting pattern most frequently (76% of the lifts performed) during the repetitive lifting task.18, 19, 20 The CLBP subjects that were classified to the control hidden Markov model were labeled as high-performing CLBP lifters. The high-performing CLBP lifters frequently used all of the lifting patterns except the slow, low jerk lifting pattern. The frequency of the lifting patterns that the high-performing CLBP group, guarded CLBP group, and the control group used during the lifting task are shown in table 1. None of the demographics were significantly different between the high-performing CLBP lifters and the guarded CLBP lifters (table 2).
Table 1. Frequency That the 5 Lifting Patterns Were Used During the Lifting Task for the High-Performing CLBP Group, Guarded CLBP Group, and the Control Group
| Lifting Patterns | High-Performing CLBP Lifters | Guarded CLBP Lifters | Controls |
|---|---|---|---|
| Slow, low jerk lift | 4 | 76 | 3 |
| Squat starting posture | 19 | 3 | 20 |
| Fast, high jerk lift | 16 | 4 | 34 |
| Torso starting posture | 35 | 12 | 13 |
| Two-segment lift | 26 | 5 | 29 |
Table 2. Subjects in Each of the Demographic Variables
| Variables | Guarded CLBP Lifters | High-Performing CLBP Lifters | P |
|---|---|---|---|
| Pain duration (y) | 3.72±5.13 | 4.52±5.69 | .523 |
| Age (y) | 36.36±9.24 | 39.66±11.03 | .162 |
| Sex | .439 | ||
 Males | 55 | 63 | |
| Females | 45 | 36 | |
| Ethnicity | .247 | ||
| White | 67 | 77 | |
| Black | 33 | 19 | |
| Other | 0 | 3 | |
| Education | .589 | ||
| ≤12th grade | 20 | 16 | |
| High school graduate | 34 | 31 | |
| Trade/technical school | 25 | 31 | |
| Some college | 16 | 16 | |
| College degree or higher | 5 | 6 | |
| Marital status | .851 | ||
| Single | 25 | 22 | |
| Separated/divorced | 20 | 28 | |
| Married | 50 | 44 | |
| Widowed | 5 | 6 | |
| Employment status | .255 | ||
| Full-time (>30h/wk) | 18 | 32 | |
| Part-time (<30h/wk) | 3 | 4 | |
| Unemployed | 0 | 4 | |
| Retired | 11 | 0 | |
| Working part-time because of pain | 8 | 8 | |
| Unemployed because of pain | 55 | 44 | |
| Retired early because of pain | 0 | 4 | |
| Other | 5 | 4 | |
| How pain began? | .765 | ||
| Accident at work | 63 | 65 | |
| Accident at home | 7 | 3 | |
| Following surgery or illness | 5 | 3 | |
| Pain just began | 7 | 16 | |
| Other | 19 | 13 | |
| Pain hours (h/d/wk) | .206 | ||
| 0–4 | 2 | 3 | |
| 4–8 | 10 | 3 | |
| 8–12 | 20 | 7 | |
| >12 | 68 | 87 | |
| Surgery | .265 | ||
| Never | 73 | 58 | |
| Once | 14 | 13 | |
| Twice | 14 | 19 | |
| >Twice | 0 | 10 |
The MANOVAs showed that the high-performing CLBP lifters were statistically different from the guarded lifters for the domains of pain (P=.007), self-efficacy (P=.013), and lifting (P<.001) as shown in table 3. Follow-up ANOVAs indicate that within the lifting domain, the difference in the number of lifts completed (P<.001) was significant, and in the pain domain the MPI pain severity scale (P=.025) and pain intensity (P=.005) from the Jan van Breemen examination were significantly different between the 2 CLBP subgroups. The results indicate that the high-performing CLBP group performed more lifts, reported lower pain intensity, lower pain severity, and had greater task self-efficacy than the guarded CLBP group. A stepwise discriminant function analysis found that number of lifts, pain severity, pain intensity, and self-efficacy were all independent contributors to CLBP group separation (table 4). The discriminant function was significant (P=.015).
Table 3. Comparison of the 2 CLBP Subgroups
| Measures | CLBP Groups | Effect Size | P | DFA Entry | |
|---|---|---|---|---|---|
| High-Performing CLBP Lifters | Guarded CLBP Lifters | ||||
| Sample size | 35 | 46 | |||
| Medical domain† | .665 | ||||
| MEDICS scale | −0.15±0.63 | 0.00±0.86 | .205 | .383 | |
| Body mass index | 27.65±5.9 | 28.04±6.77 | .062 | .812 | |
| Pain domain† | .007⁎ | ||||
| MPI: pain severity | 4.47±0.71 | 4.89±0.86 | .535 | .025⁎ | 3 |
| Jan van Breemen: pain intensity | 5.57±1.72 | 6.66±1.56 | .665 | .005⁎ | 2 |
| Psychosocial domain† | .100 | ||||
| MPI dysfunctional composite score | 57.27±9.24 | 62.60±10.30 | .547 | .035 | |
| MPI interpersonally distressed composite score | 39.12±13.36 | 38.10±11.74 | .081 | .737 | |
| Cognitive domain† | .119 | ||||
| Coping strategies: emotionality | 3.21±1.32 | 3.85±1.31 | .487 | .039 | |
| Coping strategies: worrying | 4.13±1.35 | 4.62±1.18 | .387 | .095 | |
| Disability domain† | .122 | ||||
| MPI: general activities | 2.08±0.9 | 1.77±0.79 | .367 | .068 | |
| ODI | 51.09±14.43 | 51.06±14.43 | .002 | .991 | |
| Gait speed | 35.29±7.95 | 41.63±15.89 | .426 | .057 | |
| Jan van Breemen: functional status | 4.50±1.62 | 3.91±1.15 | .532 | .123 | |
| Pain behavior checklist | 6.91±3.60 | 7.77±3.94 | .228 | .238 | |
| Spinal mobility domain (cm)† | .662 | ||||
| Jan van Breemen: flexion | 5.19±4.01 | 5.21±1.41 | .007 | .975 | |
| Jan van Breemen: flexion/extension | 6.10±2.18 | 6.48±1.88 | .187 | .414 | |
| Self-efficacy domain† | 3.42±0.91 | 2.88±0.91 | .593 | .013⁎ | 4 |
| Lifting domain† | .002⁎ | ||||
| No. of lifts | 35.06±23.28 | 18.61±16.39 | .829 | .001⁎ | 1 |
| Static strength | 84.31±54.47 | 60.09±57.68 | .432 | .121 | |
⁎Significant P values. |
†Indicates domain. |
Table 4. Coefficients and Classification Rate of the Discriminant Function Analysis for the High-Performing CLBP Group and the Guarded CLBP Group
| Variables | Coefficients | |
|---|---|---|
| Guarded CLBP Group | High-Performing CLBP Group | |
| No. of lifts | −0.08 | −0.05 |
| Pain intensity | 9.47 | 8.815 |
| Pain severity | 2.72 | 2.46 |
| Self-efficacy | 9.24 | 9.14 |
| Constant | −45.52 | −41.54 |
| Classification (%) | 79 | 65 |
All CLBP subjects were asked their pain intensity before the start of the lifting task (baseline), after the static lifting task, and after the repetitive lifting task. The results of the repeated-measure ANOVA are shown in table 5. The high-performing CLBP group reported significantly lower pain intensity ratings than the guarded CLBP group (P<.001) and both CLBP groups reported increased pain intensity ratings from the baseline to the end of the lifting task (P<.001). There was no significant group-by-time interaction.
Table 5. Pain Intensity Ratings of the CLBP Subjects at the 3 Time Points During the Functional Capacity Evaluation
| Classifications | Average Pain Ratings | P | ||||
|---|---|---|---|---|---|---|
| Baseline | After Static Lifting Task | After Dynamic Lifting Task | Group | Time | Group by Time | |
| Guarded CLBP group | 5.5±2.13 | 6.54±2.08 | 7.72±1.57 | .001 | .001 | .931 |
| High-performing CLBP group | 3.91±2.71 | 4.83±2.60 | 6.06±2.20 | |||
Discussion
The hypotheses of this study were that CLBP subgroups found with the hidden Markov model classification procedure are different on measures other than lifting patterns. The guarded CLBP lifters were found to report significantly higher pain intensity, higher pain severity, and lower self-efficacy, indicating that the hypothesis was supported and that the hidden Markov model classification procedure can identify 2 meaningfully different CLBP subgroups.
The classification procedure identified a group of CLBP subjects that perform lifts similar to control subjects, named the high-performing CLBP group, and a group of CLBP subjects that perform lifts very differently from control subjects, named the guarded CLBP group. The 2 CLBP groups performed very different lifting patterns during the repetitive lifting task. The guarded CLBP lifters used the slow, low jerk lifting pattern in almost all of the lifts that were performed by these subjects (see table 1). The high-performing CLBP group did not perform the slow, low jerk lifting pattern frequently. In fact, they used this lifting pattern less frequently than any of the other 4 lifting patterns. Because jerk is the rate of change of acceleration and can be related to muscle force, the lower values of jerk suggest that the guarded CLBP groups were not using maximum muscle force to lift the load but may instead be co-contracting antagonistic muscles. This guarded motion may be used to restrict motion that could produce further pain exacerbation33 or to protect the body from increases in pain. The higher levels of self-reported pain, smaller numbers of lifts completed and type of lifting style performed by the guarded CLBP group suggest that, for them, pain may have a greater impact on body motion than it does on the motion of the high-performing CLBP subjects.
The high-performing CLBP subjects used lifting patterns that were similar to control subjects with the exception of the lifting patterns related to starting posture and jerk. The high-performing CLBP group used a squat style lifting pattern more frequently than controls. In contrast, control subjects performed a fast, high jerk lifting pattern more frequently than the high-performing CLBP group. CLBP subjects may be more likely than controls to perform lifts that started in a squat posture because the CLBP patients are often told to lift with their legs in treatment programs. Because a majority of the CLBP subjects were likely to have had prior treatment before entering the study, the treatment instructions could explain this lifting pattern difference. The difference in the jerk lifting pattern could be related to the significant difference in the amount of weight lifted and/or to the significant difference in pain report during the repetitive lifting task, because controls lifted almost twice as much weight as CLBP subjects and were pain-free. It is reasonable to assume that the greater weight would require more muscle force. Because jerk is related to the rate of change of force, lifting a heavier weight might result in a higher jerk lifting style than lifting a lighter weight. CLBP subjects may have avoided the higher jerk lifts that control subjects performed because this type of lifting pattern may increase pain.
The guarded CLBP lifters reported higher levels of pain severity and pain intensity and lower self-efficacy than high-performing CLBP lifters, suggesting that these measures may have an impact on body motion during lifting. Specifically, higher pain intensity may translate to guarded lifting style and lower endurance. The observation is not surprising that the CLBP subjects who reported lower pain severity and pain intensity lifted more like control subjects than did those CLBP subjects who reported higher pain levels. The significant difference in self-efficacy between the 2 CLBP subgroups and the result that self-efficacy is an independent contributor to CLBP subgroup separation in the discriminant function analysis may be of more interest. This result suggests that patient's perception of their ability to complete a task is related to their physical performance, and that incorporating methods to change these perceptions in rehabilitation programs might be beneficial to the CLBP patient. Future research should focus on evaluating treatment programs that include changing self-efficacy and pain perception on treatment effectiveness.
The results found in this study are consistent with previous studies that have found correlations between self-reported measures and performance on physical functioning tasks. Rudy et al34 found that self-efficacy expectations and perceived emotional and physical health were significant predictors of subjects' performance on a lifting task in a sample of chronic pain patients with lower-extremity amputation. Vlaeyen et al35 found a significant covariation between lumbar muscular activity and the pain report, suggesting that the presence of pain results in tensing of the muscles in patients. Lackner and Carosella36 found that self-efficacy was a better predictor of lifting ability than measures of perceived control over pain or psychologic distress. Verbunt et al37 found an association between decreased quadriceps muscle strength of CLBP subjects and increased self-reported pain intensity and psychologic distress. Our results combined with these previous studies suggest a link between physical performance during tasks such as lifting and the self-report measures of pain and self-efficacy. Future studies should focus on determining how this link may change after CLBP patients have completed a treatment program.
Study Limitations
A limitation of the study was sample size. Several measures, such as disability, coping strategies, and MPI dysfunctional composite score had large to moderate effect sizes indicating potential for significant differences in larger sample sizes. Although increasing the sample size would provide further information, the present study was relatively large, and a larger clinical study would require extensive resources that are not easily available.
Conclusions
This study identified 2 groups of CLBP subjects based on their lifting patterns that were performed during a repetitive lifting task. The 2 groups were significantly different in measures of pain and self-efficacy. A high-performing CLBP group lifted more similarly to control subjects than a guarded CLBP group, completed more lifts and reported lower pain severity, lower pain intensity, and higher self-efficacy. The guarded CLBP group lifted very differently from control subjects, completed fewer lifts, and reported higher pain intensity, higher pain severity, and lower self-efficacy. This study shows a relationship between self-reported measures and physical performance and shows the importance of examining physical functioning in classification systems for pain patients.
Suppliers
References
- . A treatment-based classification approach to low back syndrome: identifying and staging patients for conservative treatment. Phys Ther. 1994;75:470–489
- . 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;28:1363–1372
- . Towards a comprehensive assessment of chronic pain patients. Behav Res Ther. 1987;25:237–249
- . Toward an empirically derived taxonomy of chronic pain patients: integration of psychological assessment of data. J Consult Clin Psychol. 1989;56:233–238
- . Differential treatment response of TMD patients as a function of psychological characteristics. Pain. 1995;61:103–112
- . Characterizing the course of low back pain: a latent class analysis. Am J Epidemiol. 2006;163:754–761
- . Musculoskeletal pain in primary health care: sub-groups based on pain intensity, disability, self-efficacy, and fear-avoidance variables. J Pain. 2007;8:67–74
- . Clinical instability of the lumbar spine: its pathological basis, diagnosis and conservative management. In: Boyling JD, Jull G editor. Grieve's modern manual therapy. 3rd ed. Philadelphia: Elsevier; 2004;p. 311–331
- . Differences in sitting postures are associated with nonspecific chronic low back pain disorders when patients are subclassified. Spine. 2006;31:698–704
- . Physical functioning: self-report and performance measures are related but distinct. Spine. 2003;28:2407–2413
- . The effect of load on coordination of the trunk for subjects with and without chronic low back pain during flexion-extension and lateral bending. Clin Biomech (Bristol, Avon). 2000;15:407–416
- . Spine loading characteristics of patients with low back pain compared with asymptomatic individuals. Spine. 2001;26:2566–2574
- . Activation imbalances in lumbar spine muscles in the presence of chronic low back pain. J Appl Physiol. 2003;94:1410–1420
- . Back and hip extensor fatigability in chronic low back pain patients and controls. Arch Phys Med Rehabil. 1998;79:412–417
- . A measure of body coordination during repetitive dynamic lifting. IEEE Trans Rehabil Eng. 1993;1:137–144
- . The use of splines to calculate jerk for a repetitive lifting task involving chronic lower back patients. IEEE Trans Neural Syst Rehabil Eng. 2005;13:406–414
- . Body motion during repetitive isodynamic lifting: a comparative study of normal subjects and low-back pain patients. Pain. 2003;105:319–326
- . Classifying chronic lower back pain groups using a time series model of lifting. [dissertation] Pittsburgh: Univ Pittsburgh; 2007;
- . Evaluating the reliability of hidden Markov models that describe the lifting patterns of chronic lower back pain patients and controls. Conf Proc IEEE Eng Med Biol Soc. 2006;1:3238–3241
- Slaboda JC, Boston JR, Rudy TE. Using HMMs to identify groups in patient population: a simulation. Presented to: IEEE 14th Workshop on Statistical Signal Processing; 2007 Aug 26-28; Madison (WI). p 355-7.
- . An introduction to the Hidden Markov models. IEEE ASSP Magazine. 1986;Jan:4–16
- . Quantification of biomedical findings of chronic pain patients: development of an index of pathology. Pain. 1990;42:167–182
- . Objectivity and repeatability of measurements in low back pain. Scand J Rehabil Med. 1982;14:21–26
- . Multidimensional Pain Inventory, version 3.0 user's guide. University of Pittsburgh http://www.pain.pitt.edu/mpiAccessed May 6, 2008
- Turk DC, Rudy TE. Pain experience: assessing the cognitive component. Paper presented to: 5th Annual Meeting of the American Pain Society; 1985; Dallas (TX).
- . The Oswestry low back pain disability questionnaire. Psychotherapy. 1980;66:271–273
- . An empirical examination of the “pain behavior” construct. J Behav Med. 1985;8:119–130
- . Effects of frequency and load of lift on endurance time. Ergonomics. 1989;23:51–56
- . Trunk force development during static and dynamic lifts. Hum Factors. 1989;29:19–29
- . An examination of procedures for determining the number of clusters in a data set. Psychometrika. 1980;50:159–179
- . Methodology review: clustering methods. Appl Psychol Meas. 1987;11:329–354
- . Design sensitivity: statistical power for experimental research. Newbury Park: Sage; 1990;
- . Chronic myofascial pain syndrome. In: Tollison CD editors. Handbook of chronic pain management. Baltimore: Williams & Wilkins; 1989;p. 509–529
- . Psychosocial predictors of physical performance in disabled individuals with chronic pain. Clin J Pain. 2003;19:18–30
- . Fear of movement/(re)injury in chronic pain and its relation to behavioral performance. Pain. 1995;62:363–372
- . The relative influence of perceived pain control, anxiety, and functional self-efficacy on spinal function among patients with chronic lower back pain. Spine. 1999;24:2254–2261
- Pain-related factors contributing to muscle inhibition in patients with chronic lower back pain. Clin J Pain. 2005;21:232–240
Supported by the National Institutes of Health (grant nos. 1R01 AR38698, 1R01 AG18299) and the University of Pittsburgh (Provost Development Fund).No commercial party having a direct financial interest in the results of the research supporting this article has or will confer a benefit on the authors or on any organization with which the authors are associated.
PII: S0003-9993(08)00351-1
doi:10.1016/j.apmr.2008.01.016
© 2008 American Congress of Rehabilitation Medicine and the American Academy of Physical Medicine and Rehabilitation. Published by Elsevier Inc. All rights reserved.
Volume 89, Issue 8 , Pages 1542-1549, August 2008
