Predicting Response to Motor Control Exercises and Graded Activity for Patients With Low Back Pain: Preplanned Secondary Analysis of a Randomized Controlled Trial
- Luciana Gazzi Macedo,
- Christopher G. Maher,
- Mark J. Hancock,
- Steve J. Kamper,
- James H. McAuley,
- Tasha R. Stanton,
- Ryan Stafford and
- Paul W. Hodges
- L.G. Macedo, PT, PhD, Physical Therapy, University of Alberta, 2-50 Corbett Hall, Edmonton, Alberta T6G 2G4, Canada.
- C.G. Maher, PT, PhD, The George Institute for Global Health, The University of Sydney, Sydney, New South Wales, Australia.
- M.J. Hancock, PT, PhD, Discipline of Physiotherapy, Faculty of Human Sciences, Macquarie University, Sydney, New South Wales, Australia.
- S.J. Kamper, PT, PhD, EMGO+ Institute, VU University Medical Centre, Amsterdam, the Netherlands, and The George Institute for Global Health, The University of Sydney.
- J.H. McAuley, PhD, Neuroscience Research Australia, Sydney, New South Wales, Australia.
- T.R. Stanton, PT, PhD, School of Health Sciences, The University of South Australia, Adelaide, South Australia, Australia, and Neuroscience Research Australia.
- R. Stafford, PhD, School of Health and Rehabilitation Sciences, The University of Queensland, Brisbane, Queensland, Australia.
- P.W. Hodges, PT, PhD, Physiotherapy, School of Health and Rehabilitation Sciences, The University of Queensland.
- Address all correspondence to Dr Macedo at: lmacedo{at}ualberta.ca, or lucianagazzi{at}hotmail.com.
Abstract
Background Current treatments for low back pain have small effects. A research priority is to identify patient characteristics associated with larger effects for specific interventions.
Objective The aim of this study was to identify simple clinical characteristics of patients with chronic low back pain who would benefit more from either motor control exercises or graded activity.
Design This study was a secondary analysis of the results of a randomized controlled trial.
Methods One hundred seventy-two patients with chronic low back pain were enrolled in the trial, which was conducted in Australian physical therapy clinics. The treatment consisted of 12 initial exercise sessions over an 8-week period and booster sessions at 4 and 10 months following randomization. The putative effect modifiers (psychosocial features, physical activity level, walking tolerance, and self-reported signs of clinical instability) were measured at baseline. Measures of pain and function (both measured on a 0–10 scale) were taken at baseline and at 2, 6, and 12 months by a blinded assessor.
Results Self-reported clinical instability was a statistically significant and clinically important modifier of treatment response for 12-month function (interaction: 2.72; 95% confidence interval=1.39 to 4.06). Participants with high scores on the clinical instability questionnaire (≥9) did 0.76 points better with motor control exercises, whereas those who had low scores (<9) did 1.93 points better with graded activity. Most other effect modifiers investigated did not appear to be useful in identifying preferential response to exercise type.
Limitations The psychometric properties of the instability questionnaire have not been fully tested.
Conclusions A simple 15-item questionnaire of features considered indicative of clinical instability can identify patients who respond best to either motor control exercises or graded activity.
Randomized controlled trials and systematic reviews evaluating the effectiveness of interventions for patients with chronic, nonspecific low back pain (LBP) typically demonstrate small treatment effects.1,2 It has been proposed that it may be possible to identify patients who are more likely to respond to a specific treatment (compared with no treatment or an alternative treatment) and for whom treatment effects are larger.3,4 Investigating subgroups of patients with LBP with specific characteristics who respond best to different treatments has been identified as the number one research priority in this field.5,6
Exercise therapy is endorsed in guidelines7 and systematic reviews2 as an effective treatment for chronic LBP. However, although more effective than no intervention, the effect size of exercise, like other back pain treatments, is small. A meta-regression study8 investigated various aspects of exercise programs and identified that individually designed programs that included stretching or strengthening and that included supervision had larger treatment effects. To illustrate this result, the effect of exercise estimated from all 43 trials in the review was 3.4 (95% confidence interval [95% CI]=2, 4.7) points reduction in pain, measured on a 0–100 pain scale.8 However, the authors estimated that the effect would rise to 18.1 points (95% CI=11.1, 25.0) for programs including the most effective intervention characteristics. Although it is plausible that identification of characteristics of the patients who respond best to different types of exercise may further enhance the treatment effects, this possibility is yet to be investigated.
Motor control exercises (sometimes called lumbar stabilization exercises) and graded activity using the principles of cognitive-behavioral therapy are 2 popular forms of exercise therapy for people with chronic LBP, with evidence of effectiveness from randomized controlled trials9,10 and systematic reviews.11,12 Motor control exercises use a motor learning approach to optimize control of the spine and pelvis via rehabilitation of posture, movement, and the coordination of muscles involved in the control and movement of the spine.13 Therefore, this treatment is expected to work best in people who have impaired control and coordination of the spinal muscles, which have been proposed to include both reduced and excessive spinal stability.14,15 Graded activity uses a cognitive-behavioral approach to increase activity tolerance by addressing pain-related fear, kinesiophobia, and unhelpful beliefs and behaviors concerning back pain while correcting physical impairments such as impaired endurance, muscle strength, and balance.10 Theoretically, this treatment should work best in patients who are physically deconditioned and have unhelpful beliefs about their back pain. Given that both exercise interventions are based on specific rationales, plausible treatment effect modifiers related to characteristics of the patient can be explored. If characteristics of patients who respond best to one of these exercise interventions compared with the other can be identified, patient outcomes could be improved.
The aim of this study was to identify simple clinical characteristics of patients who would benefit more from motor control exercises compared with graded activity, or vice versa, by evaluating potential treatment effect modifiers, identified a priori based on each treatment rationale. Guidelines16,17 on evaluation of treatment effect modification and clinical prediction rules were followed to ensure high methodological quality.
Method
Design Overview
The data for this study were drawn from a randomized clinical trial18 comparing motor control exercises to graded activity for patients with chronic, nonspecific LBP. The trial was conducted in Sydney and Brisbane physical therapy clinics, with participants enrolled in the trial during the period October 2007 to November 2009. The trial, including effect modification analyses, was prospectively registered (ACTRN12607000432415) and the trial protocol published.19 All patients signed an informed consent form prior to their inclusion in the study.
Setting and Participants
Patients with LBP of greater than 3 months' duration were invited to participate if they met the following criteria: chronic, nonspecific LBP (>3 months' duration) with or without leg pain; currently seeking care for LBP; between 18 and 80 years of age; English speaker; suitable for active exercises (as assessed with the Physical Activity Readiness Questionnaire from the American College of Sports Medicine guidelines)20; expected to reside in the Sydney or Brisbane region for the study duration; and had a score of moderate or greater on question 7 (How much bodily pain have you had during the past week?) or 8 (During the past week, how much did pain interfere with your normal work, including both work outside the home and housework?”) of the 36-Item Short-Form Health Survey (SF-36).
Exclusion criteria were: known or suspected serious pathology, nerve root compromise (at least 2 signs of the same spinal nerve: sensation loss, reduced or absent reflexes or myotomal weakness), pregnancy, previous spinal surgery or scheduled for surgery during the trial; and contraindication to exercise program.
Randomization and Interventions
The randomization sequence with a 1:1 allocation ratio was computer-generated by an investigator not involved in recruitment or treatment allocation. Allocation codes were concealed in sequentially numbered, sealed, opaque envelopes by the same investigator who created the randomization sequence. The primary goal of the motor control exercise program was to enable the patient to regain control and coordination of the spine and pelvis using principles of motor learning.21 The intervention was based on assessment of the individual participant's trunk coordination (including consideration of muscle activation, posture, and movement) and treatment goals (set collaboratively with the therapist). The first stage of the intervention included the implementation of a protocol designed to improve the activation of muscles identified to have poor control, such as the transversus abdominis, multifidus, and pelvic-floor muscles,22 and reduce the activity of any muscles identified to be overactive, such as the obliquus externus abdominis muscle. The second stage included the progression of the exercises toward more functional activities using static and then dynamic tasks.23
The primary goal of the graded activity program was to increase activity tolerance by performing individualized and submaximal exercises. The program was based on the activities that each participant identified as problematic. The activities in the program were progressed in a time-contingent manner (rather than a pain-contingent manner) from the baseline-assessed ability to a target goal set jointly by participant and therapist. Participants received daily quotas and were instructed to perform the agreed-upon amount. Cognitive-behavioral principles were used to address negative behaviors and pain-related anxiety. Both programs are comprehensively described in the original publication, including a table delineating specific characteristics of each intervention.18
Baseline Predictors of Response
Baseline characteristics of each trial participant were collected by a researcher who was blinded to the participant's treatment allocation. Seven baseline characteristics that would plausibly predict preferential response to graded activity compared with motor control exercises were selected a priori and are presented in Table 1. The predicted direction of effect also was defined for each predictor.17
Description of Candidate Baseline Predictorsa
The 7 putative predictors were measures of walking tolerance, habitual physical activity, self-reported signs of clinical instability (Appendix), self-efficacy, coping strategies, fear and anxiety of pain, and psychosocial risk factors. We chose measures to evaluate the constructs based on ease of clinical application and ready availability. The predictors were prespecified in a published trial protocol19; however, some changes to the protocol were required. We replaced the Three-Minute Step Test with the Shuttle Walk Test based on feasibility testing prior to starting the trial. The laboratory measures of trunk proprioception, trunk stiffness, trunk muscle response, and deep muscle control were not included in this analysis, as these complex neurophysiological measures cannot be performed in a typical clinic and, therefore, did not align with the aim of this study to identify simple clinical characteristics of patients who would benefit more from motor control exercises or graded activity. The complex neurophysiological measures were only collected in a subset of patients (n=76) and will be used in a further study investigating mechanisms. The instruments used in the current study, and their interpretation, are summarized in Table 1.
Our hypothesis was that graded activity would be superior to motor control exercises in participants with low walking tolerance, low habitual physical activity, low self-reported clinical instability, low self-efficacy, low coping, high fear and anxiety, and high psychosocial risk and that motor control exercises would be superior to graded activity in participants with high self-reported instability.
Outcome Measures and Follow-up
Measurements of pain and function were taken at baseline and at 2, 6, and 12 months by a blinded assessor. Pain was measured as the average pain over the previous week using a 0–10 scale (low scores indicate less pain), and function was measured using the 0–10 Patient-Specific Functional Scale (PSFS) (high scores indicate greater function).21 We chose the PSFS because it is more responsive than other measures such as the Roland-Morris Disability Questionnaire.21
Data Analysis
We prespecified a threshold of 2 units (∼1.0 standard deviation) for a clinically important interaction effect for both outcomes.19 The sample size of 172 participants was calculated a priori to detect an interaction effect size of 1.0 standard deviation and a treatment main effect of 0.5 standard deviation, with an alpha level of .05 and power of 0.80 and allowing for 10% loss to follow-up and 10% treatment nonadherence.19 An intention-to-treat method was used in all statistical analyses.
We investigated baseline patient characteristics associated with a greater effect of graded activity versus motor control exercises for the primary trial outcomes of pain and function at 2 and 12 months (ie, separate models for each of these 2 follow-ups). We did not analyze response to treatment at the 6-month follow-up in order to decrease the number of models and, therefore, the chance of type I error. Outcomes at 2 and 12 months represent short-term and long-term follow-ups. We began by investigating individual factors, but we also investigated combinations of factors. Treatment effect modification was evaluated using a group × predictor interaction term.17 The models were built using the linear regression model commands within IBM SPSS Statistics version 21 (IBM Corp, Armonk, New York).
Univariate testing.
As the goal was to identify clinical subgroups that could be simply identified, we dichotomized the predictors using a median split and then built separate models to predict pain and function outcomes at 2 months and 12 months for the 7 predictors. The models included terms for patient group, predictor, group × predictor interaction, and baseline score for the dependent variable. Predictors that provided a P value <.20 for the group × predictor term proceeded to the multivariate testing. A median split was used to dichotomize the predictors. We did not perform analyses to choose an “optimal” cutoff point, as this approach is strongly advised against.22
Multivariate modeling.
As we also were interested in building a clinical prediction rule, we attempted to build a multivariate model. We used a backward selection procedure to build the models. We entered all predictors from the univariate models with a P value <.20 into the first multivariate model and in subsequent models removed one predictor at a time starting with the predictor with the highest P value until only predictors with a P value <.05 remained.
Role of the Funding Source
The trial received funding from Australia's National Health and Medical Research Council.
Results
Participant flow through the study is shown in Figure 1, and the baseline characteristics of the participants are shown in Table 2.
Study flow diagram. IQR=interquartile range.
Baseline Characteristics of Participantsa
Univariate Testing
The results of the univariate testing are shown in Table 3. The Lumbar Spine Instability Questionnaire (LSIQ), Coping Strategies Questionnaire, Pain Anxiety Symptom Scale, and Örebro Low Back Pain Screening Questionnaire were found to be statistically significant (P<.05) treatment effect modifiers for function at 12 months but not at 2 months. For example, the effect of graded activity versus motor control exercises was greater in participants with low instability (based on LSIQ scores); the mean point estimate for the interaction was 2.72 (95% CI=1.39, 4.06) units on the 0–10 function scale. The other 3 significant interaction effects favored graded activity over motor control exercises in participants with low coping (based on Coping Strategies Questionnaire scores), low fear (based on Pain Anxiety Symptom Scale scores), and low psychosocial risk (based on Örebro questionnaire scores) but did not reach our prespecified threshold for an interaction of 2 units. Additionally, 2 predictors (fear and psychosocial risk) did not influence outcomes in the hypothesized direction. The adjusted treatment effects in subgroups that were positive and negative on each statistically significant treatment effect modifier are shown in Table 4.
Results of Univariate Testing of the Dichotomized Predictor Variables on Function Scoresa
Adjusted Function (Patient-Specific Functional Scale) Treatment Effects (95% Confidence Interval) in Subgroup Positive and Subgroup Negative Participants for the Statistically Significant Predictors Onlya
To illustrate the effect of self-reported signs of instability on treatment effect, Figure 2 shows outcomes in both treatment groups at baseline and 2- and 12-month follow-ups for all participants and in a separate panel with the groups stratified using the median LSIQ score of 9. Participants with an LSIQ score of 9 or greater (ie, high instability) are described as “positive” and those with an LSIQ score of less than 9 are described as “negative” on the questionnaire. At 12 months, participants who were positive on the LSIQ did 0.76 points better with motor control exercises than with graded activity, whereas those who were negative on the LSIQ did 1.93 points better with graded activity than with motor control exercises. These findings mean that on a scale from 0 to 10, where 0 is low function and 10 is high function, an individual who did not have instability and received graded activity was about 2 points more functional a year after inclusion in the study than those who did not have instability and received motor control exercises. This is a large difference that has been reported to be clinically significant.23
Function outcomes at baseline and 12-month follow-up to illustrate the main effect of treatment and treatment effect modification. Values represent unadjusted means and 95% confidence intervals. Graph A shows data for all participants. Graph B shows the outcome when the groups were stratified by median score on the Lumbar Spine Instability Questionnaire (LSIQ). Participants with an LSIQ score of 9 or greater (ie, high instability) are described as “positive” (+ve), and those with an LSIQ score of less than 9 are described as “negative” (−ve) on the questionnaire.
Multivariate Modeling
The LSIQ, Coping Strategies Questionnaire, Pain Anxiety Symptom Scale, and Örebro questionnaire yielded P values <.20 for predicting function at 12 months, and the Coping Strategies Questionnaire, Pain Anxiety Symptom Scale, and Örebro questionnaire yielded P values <.20 for predicting pain at 12 months. These became candidate variables for the multivariate analysis. None of the predictors achieved P values <.20 for short-term outcomes.
We were unable to build a multivariate model for pain or function outcomes at 12 months. The final pain model included the Örebro questionnaire, which did not achieve a P value <.05. The final function model included the LSIQ, which did achieve a P value <.05.
Discussion
Statement of Principal Findings
We found that motor control exercises provided better function outcomes at 12 months than graded activity in patients with self-reported signs of clinical instability (LSIQ score <9), but in those who scored low on the questionnaire of clinical instability (LSIQ score ≥9), the situation was reversed and graded activity provided better function outcomes. This finding is consistent with the theoretical rationale for motor control training. The treatment effect modification was statistically significant and larger than the threshold of 2 units for a clinically important interaction effect we prespecified in the published trial protocol.19 Measures of coping, fear and anxiety, and psychosocial risk also moderated the effect of treatment on function at 12 months but did not reach our prespecified threshold. There were no statistically significant interactions for pain at 2 and 12 months or for function at 2 months.
Strengths and Weaknesses of the Study
The strengths of the study are that we tested effect modification in a high-quality trial, and we followed a prespecified protocol. We investigated a limited number of prespecified predictors underpinned by theory. The analysis used appropriately constructed interaction terms, and we blinded assessors of outcome to predictors and group allocation. One of the potential problems with evaluating treatment effect modification is a high type I error rate associated with repeated testing. To minimize this effect, we undertook a stepped approach. Nevertheless, we acknowledge that we have taken a fairly exploratory approach and that any positive results will need replication in an independent data set. Our view is that the results of this study should be viewed cautiously, and it would be premature to advocate clinical application at this stage. We also would caution that the LSIQ was developed from the results of a Delphi study of experienced clinicians aiming to identify features associated with instability.
Self-administration of the LSIQ required adjustment of some features. For instance, the consensus feature “Reports feelings of giving way or back giving out” became item 1: “I feel like my back is going to give way or give out on me.” We performed post hoc evaluation of the internal consistency and the influence of floor and ceiling effects for the LSIQ in our sample. Cronbach's alpha was calculated as a measure of internal consistency and interpreted according to the threshold provided by de Vet et al,24 and the threshold proposed by McHorney and Tarlov25 was used to determine the presence of floor or ceiling effects. Cronbach's alpha for the LSIQ was .69 (95% CI=.62, .76), which falls marginally below the lower bound of the acceptable range of .7 to .9 proposed by de Vet et al.24 The distribution of scores on the scale showed that neither floor nor ceiling effects are of concern. As only 1 participant (0.6%) reported the maximum score on the scale and no participants reported the minimum score, these findings fall well below the threshold of 15% of the sample at each end-point.25 Future comprehensive evaluation of the psychometric properties and interpretation of the questionnaire are needed.
Strengths and Weaknesses in Relation to Other Studies
Although numerous studies have attempted to identify subgroups of responders to a specific treatment for LBP, almost all are poorly designed, and no previous randomized trials have investigated subgroups of responders to different exercise approaches.26 Arguably, the best study investigating a subgroup of responders to a specific treatment for LBP is the study of Childs and colleagues,27 who developed a rule to identify patients with LBP who respond best to spinal manipulation treatment. Although these authors expressed effect modification in a slightly different way from that used in our study, it is possible to obtain equivalent data from interpolation of the figures in the article. Treatment effect modification was approximately 25 units on the 0–100 Oswestry Low Back Pain Disability Questionnaire at 1 week, which is similar in magnitude to the effect we observed on function.
Meaning of the Study: Implications for Clinicians and Policy Makers
Although it is essential to use interaction terms to test treatment subgroups, the interaction term itself does not describe the treatment effect within the group but rather the difference between the treatment effect in patients meeting the subgroup criteria and those who do not. What we can determine is that people with a score of 9 or greater on the LSIQ (clinical instability) did 0.76 points (Fig. 1) better with motor control exercises, whereas people who had low scores for the LSIQ (<9) did 1.93 points better with graded activity. At 12 months, the treatment effect in the negative subgroup (−1.93; 95% CI=−3.00, −0.85) was statistically and clinically significant, implying graded activity was clearly superior to motor control exercises in these patients. The treatment effect in the positive subgroup was smaller and not statistically significant (0.76; 95% CI=−0.07, 1.58). It is essential to note that this study was comparing 2 effective interventions.11,12,18 Therefore, an additional benefit of 0.76 points, by selecting motor control exercises over graded activity in these patients, is considered clinically important by the authors, especially considering the interventions have similar costs and potential harms. The 95% CI values crossed zero likely due to lack of power, and larger trials are needed in the future to validate the finding that people who score positively on the LSIQ have slightly better outcomes with motor control exercises compared with graded activity. Using the LSIQ score to contribute to decision making between graded activity and motor control exercises appears to have little downside, as both treatments are effective and carry few risks. However, as is always the case, it is important to consider other factors, such as patient preference and clinician expertise, while making treatment decisions and not rely solely on LSIQ score. This appears to be the case particularly in people with positive LSIQ scores, where the benefits of choosing motor control exercises over graded activity are relatively small.
As the current trial did not have a no-treatment control group, it is not possible to determine the effect of either intervention compared with no treatment. However, the unadjusted mean (95% CI) improvement in function from baseline in people who received graded activity and had low scores on the LSIQ was 3.7 (95% CI=2.8, 4.5), whereas the improvement was less than half this (1.8; 95% CI=1.1, 2.5) in those who received graded activity and had high scores on the LSIQ. In those patients who received motor control exercises, the improvement was 2.4 (95% CI=1.7, 3.0) in those who scored high on the LSIQ and 1.7 (95% CI=0.7, 2.7) in those who scored low on the LSIQ. If the results of the current study can be replicated, they are a major breakthrough and have important implications for management of chronic LBP. The results suggest that by targeting the right patients (those with or without clinical instability) to these 2 common exercise approaches, we may have a simple, relatively cheap method to increase the effectiveness of these treatments for patients with chronic LBP. Implementation of this approach would be very simple, as both treatments are already widely used, and the method to identify subgroups of responders to each approach is very simple.
Unanswered Questions and Future Research
An interesting finding from our study is that the interaction effect was not statistically significant for function at 2 months, although the mean estimate (0.82) suggests that at this early time point, the scores on the LSIQ were already associated with response to exercise type. This finding suggests our study was not powered to identify these smaller early effects. Larger trials in the future should provide adequate power to assess the importance of the LSIQ score to early response to graded exercise or motor control exercises. However, in patients with chronic LBP, long-term outcomes are widely considered most important. We did not find a statistically significant interaction effect for the LSIQ on pain outcomes at 12 months; however, the estimate of −0.96 (95% CI=−2.59, 0.67) again suggests our study was not powered to find these smaller effects. We used a median split to dichotomize the LSIQ and suggest the same cutoff point (ie, >9) be used in future validation studies; however, exploration of different cutoff points also may be valuable. The current trial investigated only people with chronic LBP, and it is not clear whether the findings would generalize to those with acute LBP. It is in people with chronic LBP that exercise is more widely recommended; however, the findings of the current study indicate the possibility of better effects of exercise in patients with acute or subacute LBP if better matched to the patient's presentation. Future research is needed to investigate this possibility.
The Bottom Line
What do we already know about this topic?
Motor control exercise and graded activity are two forms of exercise intervention for chronic low back pain. Both interventions are known to be effective; however, clinicians lack a method to help choose one option over the other for specific patients.
What new information does this study offer?
The results of this study demonstrated that a simple 15-item questionnaire on clinical instability can help identify patients with chronic low back pain who respond best to either motor control or graded activity exercises. Replication in future studies is required.
If you're a patient, what might these findings mean for you?
A simple questionnaire may help clinicians choose the best type of exercise for individual patients with low back pain, which might help patients achieve better outcomes.
Appendix.
Lumbar Spine Instability Questionnaire
Footnotes
All authors provided concept/idea/research design. Dr Macedo, Dr Maher, Dr Hancock, Dr Kamper, Dr McAuley, Dr Stanton, and Dr Hodges provided writing. Dr Macedo, Dr McAuley, Dr Stanton, Dr Stafford, and Dr Hodges provided data collection. Dr Macedo, Dr Maher, Dr Kamper, Dr McAuley, and Dr Hodges provided data analysis. Dr Macedo and Dr Stafford provided project management. Dr Hodges and Dr Maher provided fund procurement. Dr Macedo, Dr Maher, Dr Hancock, Dr Kamper, Dr McAuley, Dr Stanton, and Dr Hodges provided consultation (including review of manuscript before submission). The funders had no role in the study, and there are no conflicts of interest.
The study was funded by the National Health and Medical Research Council. Dr Macedo is supported by the Canadian Institutes of Health Research and the Alberta Innovates Health Solutions. Prof Maher is supported by an Australian Research Council Fellowship. Dr Kamper is supported by an NHMRC fellowship. Dr Stanton is supported by the Canadian Institutes of Health Research Postdoctoral Training Fellowship [ID 223354]. Prof Hodges is supported by a Senior Principal Research Fellowship from the NHMRC (APP1002190).
The trial was prospectively registered with the Australian New Zealand Clinical Trials Registry (ACTRN12607000432415).
- Received January 21, 2014.
- Accepted June 28, 2014.
- © 2014 American Physical Therapy Association