To identify subgroups of workers absent from work due to low back pain who are more or less likely to return to work earlier as a result of a graded activity intervention, and to investigate whether this intervention is effective in reducing pain-related fears and if so, whether these reductions in pain-related fears mediate return to work.
A subgroup analysis was conducted on data from a previous randomized controlled trial of 134 Dutch airline workers, which found that a behaviorally-oriented graded activity intervention was more effective than usual care in stimulating return to work. The subgroup analyses added interaction terms to a Cox regression model that described the relationship between treatment allocation and return to work over 12 months of followup. Furthermore, we studied the effects of graded activity on pain-related fears and added variables indicating a reduction in pain-related fears to the model in order to investigate their influence on return to work.
Statistically significant interactions were found for disability, fear-avoidance beliefs about physical activity, and fear-avoidance beliefs about work. No indication was found that the reduction in pain-related fears in the graded activity group mediated more favorable return-to-work results in this group.
Workers who perceive their disability to be moderate and workers with moderate scores for fear-avoidance beliefs return to work more rapidly as a result of the graded activity intervention than workers with higher scores. The return to work of workers receiving the graded activity intervention is possibly independent from the reductions in pain-related fears caused by this intervention.
In occupational health care it is a great challenge to find ways to improve the management of workers with low back pain. More specifically, there is a need for effective rehabilitation interventions for workers who are at risk for prolonged disability and absence from work due to low back pain (1, 2). The aim of such return-to-work interventions is to promote functional recovery and return to work of the disabled worker. So far, many efforts have been made in both clinical practice and in research to develop interventions with maximal chances of a successful outcome. Despite these efforts, the content of interventions that are subject to effectiveness research still varies widely, making it difficult to evaluate and summarize the effects in a systematic review or meta-analysis (3).
Assuming that, in general, a specific return-to-work intervention is found to be effective, it seems to be a logical step to identify relevant subgroups of workers for whom the intervention is likely to be more or less effective. This knowledge would help those who provide the treatment to decide which worker should receive which return-to-work intervention, in order to increase the probability of a successful result. Tailoring return-to-work interventions to specific subgroups of workers might eventually increase the cost-effectiveness of care that is provided for disabled workers with low back pain.
In addition to efforts to evaluate the effects of a return-to-work intervention at a group or subgroup level, it is also important to gain insight into the working mechanisms and the most appropriate elements of the intervention (2). Obviously, it is possible to label an intervention as either exercise therapy or cognitive–behavioral treatment, or both. However, the question is whether such a label reflects what exactly happens with the individual and whether it describes the way in which the intervention works.
We previously performed a randomized controlled trial (RCT) on the effects of a graded activity intervention among workers employed by an airline company who were absent from work due to low back pain (4, 5). This graded activity intervention can be described as a return-to-work intervention consisting of individual counseling, education, and physical exercises, applied according to operant conditioning behavioral principles. The results of our RCT were promising: the number of days of absence from work due to low back pain in the intervention group over 12 months of followup was reduced substantially compared with the usual care group. However, no significant influences were found for disability and pain (4, 5). Furthermore, we investigated the effects of the graded activity intervention on pain-related fears (fear-avoidance beliefs and fear of movement/[re-]injury). Ideally, the content of an intervention that aims at improving return to work should match with modifiable prognostic factors for return to work (6). Pain-related fears are assumed to be associated with a poor return-to-work prognosis (7–9) and are therefore subject to modification by the graded activity intervention. In accordance with the behavioral orientation of the intervention, we assumed that reassurance and a gradually increasing activity level would lead to a reduction in pain-related fears. As an intermediate variable, this reduction in pain-related fears might then be associated with an earlier return to work for the disabled worker (6).
In the current study we first describe the results of subgroup analyses, as applied to data from our RCT, in order to investigate which workers are most likely to benefit from the graded activity intervention. Subgroups were created on the basis of demographic characteristics, dichotomized values of the duration of the last episode of absence from work due to low back pain prior to randomization, and dichotomized baseline values of outcome measures. Subsequently, we address the question of whether the graded activity intervention was effective in reducing pain-related fears and if so, whether the reductions in pain-related fears were associated with the reduction in absence from work in the graded activity group.
PATIENTS AND METHODS
We conducted an RCT in an occupational health care setting and compared the effects of a graded activity intervention with usual care. The primary outcomes of this study (i.e., absence from work, disability, and pain) have been published previously (4, 5). The study took place in The Netherlands and was carried out in cooperation with a center that provides occupational health services for all employees of a major Dutch airline (KLM Royal Dutch Airlines). From April 1999 through December 2001, eligible workers who were absent from work due to nonspecific low back pain were included after giving their informed consent.
Workers who met the inclusion criteria were randomly assigned to graded activity or usual care. The intervention was based on an operant conditioning behavioral approach. According to the theory of operant conditioning as applied to low back pain, the future occurrence of healthy behavior tends to increase as a result of graded activity (i.e., gradually increased exercise intensity) and positive reinforcement, in contrast to inactive pain behavior (10, 11). The graded activity intervention consisted of 2 sessions of physical exercises per week, supervised by skilled physiotherapists. During the course of the intervention the load of the exercises was gradually increased toward a preset exercise goal, following a time-contingent exercise scheme. The exercise goals were connected with return-to-work goals. The usual care consisted of advice as usual from the occupational physician during work absenteeism as in the graded activity group. There were no requirements with regard to additional treatment for workers allocated to the usual care group. More specific information about the inclusion and exclusion criteria and the content of the interventions (graded activity and usual care) has been described in detail elsewhere, as well as information about the sample-size calculation, the treatment allocation, and the blinding of the outcome assessors (4, 5). The study was approved by the Medical Ethical Committee of the VU University Medical Center, Amsterdam, The Netherlands.
Collection of data and statistical analyses.
At baseline, before randomization, the workers filled out a questionnaire that included questions on demographic characteristics and the company department in which the worker was employed. Outcomes were measured at 3, 6, and 12 months of followup. The primary outcome measures of this study were return to work, disability, and pain severity (4, 5).
Return to work was defined as full return to regular work with a minimum duration of 28 days, which was in accordance with the definition of the then-valid Dutch Sickness Benefit Act (5). Workers who returned to work partially or with modified duties remained on the sick list until they made a full return to regular work. Data on days of absence from work were continuously collected by means of the electronic medical records of the company and analyzed by means of survival analysis. Cox proportional hazards model was applied in order to calculate hazard ratios (HRs) for return to work. Kaplan-Meier curves showed that the graded activity intervention was effective from ∼50 days after the randomization and onward (4, 5). By adding a time-dependent covariate to the Cox regression model, HRs for return to work were calculated for the period from ∼50 days after randomization to 12 months postrandomization. Both groups were also compared with regard to the number of recurrent episodes of absence from work due to low back pain and the total number of days of absence from work due to these recurrent episodes within the 12-month followup period. No significant differences were found, as earlier reported (5). The same holds for disability, measured according to the Roland Disability Questionnaire (12), and the severity of pain complaints, scored on an 11-point numerical scale (13). The effects of the graded activity intervention on the primary outcomes in comparison with usual care have been described more extensively in our previous publications (4, 5).
The results with regard to the secondary outcomes of this trial (i.e., fear-avoidance beliefs and fear of movement/[re-]injury) are presented in this article. Fear-avoidance beliefs were measured with the Fear-Avoidance Beliefs Questionnaire (FABQ) (14). This questionnaire focuses on the beliefs of the patient with regard to the way physical activity and work affect low back pain. The FABQ consists of 2 scales: fear-avoidance beliefs about physical activity (FABQ-PA) with 4 items, and fear-avoidance beliefs about work (FABQ-W) with 7 items. The items were scored on a 7-point scale, ranging from 0 (completely disagree) to 6 (completely agree) (14). One item on the FABQ-W scale can only be administered to workers who are off work, and was therefore excluded from the analyses, resulting in a modified FABQ-W scale (15). The possible scores range from 0–24 for the FABQ-PA and 0–36 for the modified FABQ-W. The Dutch translation of the FABQ has been found to be valid and reliable (16–19). Fear of movement/(re-)injury was measured according to the Tampa Scale for Kinesiophobia (TSK). This questionnaire consists of 17 items rated on a 4-point Likert scale (20, 21). Possible scores range from 17–68. The Dutch version of the TSK has been found to be a valid and reliable outcome measure (17, 18, 22, 23).
The study data were analyzed using SPSS statistical software, version 13.0 (SPSS, Chicago, IL).
Who benefits most?.
The following variables, measured at baseline, were considered to be possible modifiers for the effect of the graded activity intervention on return to work: age, sex, duration of the last episode of absence from work prior to randomization, disability, severity of pain, fear-avoidance beliefs about physical activity, fear-avoidance beliefs about work, and fear of movement/(re-)injury.
These variables were dichotomized by taking the median of the baseline values of the variables as a cutoff point. Separate Cox regression analyses were carried out for each of the dichotomized variables, with an interaction term added to the basic model that was previously used for calculating HRs for return to work. The interaction terms were constructed by multiplying the group indicator (i.e., graded activity or usual care) and each specific dichotomized variable.
If there was a P value <0.10 for these interaction terms, HRs were reported for each subgroup. We used a cutoff value of 0.10 to indicate statistical significance for the subgroup analyses because of a lack of statistical power for doing interaction tests.
In order to analyze the effects of the graded activity intervention on fear-avoidance beliefs and fear of movement/(re-)injury, linear regression analysis was applied, in which the outcome variable at 3, 6, and 12 months of followup was used as an outcome and the baseline value of the particular outcome variable was used as a covariate. The analyses were carried out according to the intent-to-treat principle and P values <0.05 were considered statistically significant.
How does it work?.
We hypothesized that if the graded activity intervention would be effective in reducing pain-related fears (i.e., fear-avoidance beliefs and fear of movement/[re-]injury), these effects may influence the behavior of the disabled worker and stimulate return to work. In that case, positive results of the graded activity intervention with regard to return to work over a 12-month period were potentially due to the differences in the reduction in pain-related fears between the 2 groups at 3 months.
In order to investigate this hypothesis we added the reductions in pain-related fears over 3 months of the entire study group as separate variables to the basic Cox regression model that was previously used to calculate HRs for return to work. If our hypothesis is correct, adding these variables to the model should lead to a reduction in the HR for return to work.
Patient flow and followup.
The flow of workers through the different phases of the trial is shown in Figure 1. A total of 134 workers were equally allocated to the 2 groups: 67 to the graded activity group and 67 to the usual care group. Table 1 shows the baseline characteristics and the baseline values of the pain-related fears for both groups. In cases of skewed distributions, the median values and the 25th and 75th percentiles (interquartile ranges) were calculated.
Table 1. Baseline characteristics and baseline values of outcome measures*
Graded activity (n = 67)
Usual care (n = 67)
Values are the number unless otherwise indicated. IQR = interquartile range; FABQ = Fear-Avoidance Beliefs Questionnaire; TSK = Tampa Scale for Kinesiophobia.
Age, mean ± SD years
39 ± 9
37 ± 8
Baggage and aircraft turnaround services
Engineering and maintenance
Cabin and cockpit crew department
Passenger services and other departments
Partial/full absence from work at start of study
Absence from work of current episode of low back pain prior to randomization, median (IQR) days
Duration of complaints prior to randomization, median (IQR) weeks
FABQ score, mean ± SD
Physical activity (range 0–24)
16.9 ± 5.1
15.3 ± 4.9
Work, modified (range 0–36)
21.8 ± 9.2
22.0 ± 9.9
TSK score, mean ± SD (range 17–68)
39.4 ± 6.8
40.0 ± 7.8
Graded activity and usual care.
The mean ± SD number of treatment sessions completed by the workers in the graded activity group was 13 ± 5.4, and the total intervention had an average duration of almost 7 weeks. The types of treatment received in the usual care group consisted of physiotherapy (63%), manual therapy (10%), Mensendieck therapy (10%), and chiropractic (5%). One person attended a return-to-work intervention program comparable with graded activity that was organized by a private company. The workers in the usual care group attended a mean ± SD of 13 ± 8.4 treatment sessions provided by the above-mentioned caregivers.
Who benefits most from graded activity?.
The median values and the range of baseline values for the variables that were used for creating subgroups are shown in Table 2.
Table 2. Median and range of baseline values of continuous variables used for subgroup analyses*
RDQ = Roland Disability Questionnaire; FABQ-PA = Fear-Avoidance Beliefs Questionnaire physical activity index; FABQ-W = Fear-Avoidance Beliefs Questionnaire work index; TSK = Tampa Scale for Kinesiophobia.
Duration of absence from work prior to randomization, days
Severity of pain (range 0–10)
Disability (RDQ score, range 0–24)
Fear-avoidance beliefs about physical activity (FABQ-PA score, range 0–24)
Fear-avoidance beliefs about work (FABQ-W score, range 0–42)
Fear of movement/(re-)injury (TSK score, range 19–62)
The subgroup analyses resulted in the following P values for the different interaction terms of the subgroups that were added to the model: age, P = 0.27; sex, P = 0.60; duration of last period of absence from work prior to randomization, P = 0.50; severity of pain complaints, P = 0.47; disability, P = 0.07; fear-avoidance beliefs about physical activity, P = 0.03; fear-avoidance beliefs about work, P = 0.07; and fear of movement/(re-)injury, P = 0.29.
The P values of the interaction terms with regard to disability, fear-avoidance beliefs about physical activity, and fear-avoidance beliefs about work met our criterion of <0.10. Table 3 shows the HRs for return to work for the different subgroups of workers that were created by dichotomizing these variables.
Table 3. Hazard ratios (HRs) and corresponding 95% confidence intervals (95% CIs) per subgroup of workers*
Indicating the chance for return to work in the graded activity group compared with the usual care group for the period from 50 days of absence from work since randomization and onward. RDQ = Roland Disability Questionnaire; FABQ-PA = Fear-Avoidance Beliefs Questionnaire physical activity index; FABQ-W = Fear-Avoidance Beliefs Questionnaire work index.
Adjusted for age, sex, duration of absence from work due to low back pain prior to randomization, and the proportion of workers who were partially absent from work at baseline.
Disability (RDQ score, range 0–24)
Fear-avoidance beliefs about physical activity (FABQ-PA score, range 0–24)
Fear-avoidance beliefs about work (FABQ-W score, range 0–42)
Fear-avoidance beliefs and fear of movement/(re-)injury.
The mean improvements in outcomes from baseline for fear-avoidance beliefs and fear of movement/(re-)injury and the results of the linear regression analyses are shown in Table 4. The regression coefficients indicate the differences between the mean improvements in the outcome variables of the graded activity group and the usual care group at 3, 6, and 12 months. The results show that the graded activity group had statistically significant greater improvements on the FABQ-PA (3.6, 2.9, and 2.9 points more reduction at 3, 6, and 12 months followup, respectively) than the usual care group. The graded activity group also had statistically significant greater improvements on the modified FABQ-W (4.2, 4.2, and 4.2 points more reduction at 3, 6, and 12 months followup, respectively) compared with the usual care group. The results of the TSK scores demonstrated that the reduction in fear of movement/(re-)injury was significantly greater in the graded activity group at 3 months (i.e., 3.3 points higher reduction) than in the usual care group.
Table 4. Mean improvements from baseline and differences between groups in improvements from baseline for fear-avoidance beliefs and fear of movement/(re-)injury at 3, 6, and 12 months*
Differences between groups (95% CI)
Values are the mean ± SD unless otherwise indicated. Mean difference in improvement adjusted for age, sex, duration of absence from work prior to randomization, being either partially or fully absent from work at baseline, and baseline values of the outcome variable. 95% CI = 95% confidence interval; FABQ-PA = Fear-Avoidance Beliefs Questionnaire physical activity index; FABQ-W = Fear-Avoidance Beliefs Questionnaire work index; TSK = Tampa Scale for Kinesiophobia.
Statistically significant. The mean score for fear-avoidance beliefs about physical activity is 3.6 points lower in the graded activity group than in the usual care group at 3 months.
Table 5 shows the results of the analyses after adding the intermediate variables. After adding the reductions in pain-related fears over 3 months as separate variables to the basic Cox regression model, the HR for return to work increased from 1.9 to 2.0 for fear-avoidance beliefs about physical activity, from 1.9 to 2.2 for fear-avoidance beliefs about work, and from 1.9 to 2.3 for fear of movement/(re-)injury. However, if there was an existing causal association, the HRs for return to work would be expected to decrease. Therefore, these results gave no indication that the favorable return-to-work outcome in the graded activity group compared with the usual care group was due to reductions in pain-related fears.
Table 5. Hazard ratios (HRs) and corresponding 95% confidence intervals (95% CIs) for return to work for the period from 50 days of absence from work since randomization and onward, adjusted for reductions in pain-related fears*
HRs adjusted for age, sex, duration of absence from work prior to randomization, and the proportion of workers in the groups who were partially absent from work at baseline.
Basic Cox regression model
Intermediate variable added at 3 months
Reductions in fear-avoidance beliefs about physical activity
Reductions in fear-avoidance beliefs about work
Reductions in fear of movement/(re-)injury
The subgroup analyses demonstrated that workers in the graded activity group with scores equal to or above the baseline median for disability, fear-avoidance beliefs about physical activity, and fear-avoidance beliefs about work returned to work more slowly than workers with scores for these variables that were below the baseline median. These results indicate that workers who perceive their disability to be moderate and workers with moderate scores for fear-avoidance beliefs have better chances of a successful treatment result (i.e., return to work) than workers with higher scores. Because there is a risk of chance findings, the results of post hoc subgroup analyses should be interpreted with caution (24). Although we were aware of this drawback, we decided to perform these subgroup analyses for explorative reasons. In our view, an explorative approach in this area of research is legitimated, given the generally assumed heterogeneous composition of the group of patients with nonspecific low back pain. The question “what works for whom” needs to be addressed to provide directions for further research (25). We took the baseline median as a cutoff point for creating subgroups because to our knowledge, more meaningful cutoff points for the FABQ and the TSK are currently lacking. However, the results of the subgroup analyses in this study make sense from a clinical standpoint, because it is imaginable that workers who are more severely disabled and who have higher scores for fear-avoidance beliefs have more problems in coping with their back pain, and therefore respond less positively to the graded activity treatment than other workers. It is possible that these workers need greater support and guidance from caregivers than the kind of support that is provided during a graded activity treatment. The intervention in the present study only consisted of 2 hour-long exercise sessions per week for a maximum of 13 weeks supervised by a physiotherapist, and appointments with the occupational physician once every 3–4 weeks, usually by telephone. In our opinion, this rather modest level of intensity should primarily be considered as a strength of the graded activity intervention for reasons of cost-effectiveness, but this intensity may be too low for a subgroup of workers with high scores for disability and fear-avoidance beliefs.
Furthermore, it might be argued that more cognitive-oriented interventions such as graded exposure in vivo are more appropriate for the subgroup of workers with greater pain-related fears (26). In these interventions, patients with high scores for pain-related fears are exposed more specifically to individually tailored, fear-eliciting, and hierarchically ordered physical movements, rather than in a standard graded activity program. Vlaeyen et al (26) reported on patients with low back pain with high scores for pain-related fears (i.e., TSK score >40 points) who were exposed consecutively to graded activity and graded exposure in vivo in a replicated single-case crossover study. Their time-series analyses showed that reductions in pain-related fears and pain disability only occurred during the graded exposure in vivo (26). Although these findings still need to be confirmed in an RCT, the results indicate that graded exposure in vivo might be a valuable alternative when targeting patients with high levels of pain-related fears.
The positive effects that were found for fear-avoidance beliefs and fear of movement/(re-)injury tend to support the behavioral hypotheses that underlie the graded activity intervention. They show that reassurance and the enhancement of physical exercise and activity are associated with a significant reduction in pain-related fears.
However, reduction in pain-related fear was not found to play an intermediate role between the graded activity intervention and return to work (6), which suggests that these effects were independent of each other. The return-to-work behavior of workers receiving the graded activity intervention could not be explained by reductions in pain-related fears. These results are surprising, whereas it was found in earlier studies that heightened levels of pain-related fears were associated with an increased disability level (27) and an increased risk of becoming absent from work (28, 29). Among workers with low back pain who already were absent from work, it was found that heightened pain-related fears were associated with a decreased probability of returning to work (7–9). Contrary to our findings, these results suggest that the reduction in pain-related fears may lead to an earlier return to work. The lack of an association in our study with a reduction in pain-related fears and return to work seems to be supported to some extent by a systematic review by Pincus et al (30). Based on this review that covered 9 prognostic cohort studies, it was concluded that there is little evidence for pain-related fears predicting poor prognosis in early stages of low back pain (30). However, the studies included in this systematic review, except for one (31), had been carried out in nonoccupational settings and used self-perceived disability as a study outcome instead of return to work, which limits the comparison with our study. Pincus and colleagues further assume an important role of distress and depression in the prognosis of back pain, which are factors we unfortunately did not take into account in our study (30). Our results indicate that other elements of the graded activity intervention, and not the reductions in pain-related fears, are perhaps responsible for the favorable return-to-work results. According to the operant conditioning theory for the treatment of pain, as originally postulated by Fordyce, pain behavior (e.g., staying off work for a long time) is not compatible with physical and functional activity. No further assumptions were made about cognitions such as pain-related fears (10). It is possible that the strict time-contingent nature of the treatment aims, which were linked to return-to-work goals, led to the earlier return to work in the graded activity group. The physiotherapists and the occupational physicians also worked together more closely in the graded activity group and were more aware of the working conditions at the different workplaces involved compared with the caregivers in the usual care group. Furthermore, the workers in the graded activity group were stimulated to return to work more gradually (i.e., gradual work exposure), which provided the experience of being active at work despite pain complaints. In contrast with this approach, other types of treatment executed in more clinical settings may neglect significant factors at the workplace that are critical for successful return to work (32).
Contrary to our findings and those of an earlier Swedish study (11), a more recent RCT in a Dutch occupational health setting carried out by our group reported no effect of graded activity (33). In this study, the graded activity intervention was part of a multistage return-to-work program. An explanation for this lack of effect might be the combination with another ergonomic intervention in almost 50% of the cases, which caused a delay in return to work (33). Furthermore, the intervention was performed by many physical therapists working in different practices, whereas the intervention in the present study was carried out at an in-company physical therapy clinic by a limited number of physical therapists who had been trained extensively prior to the study. In addition, referral in our study was done by the researchers, whereas in the study by Steenstra et al, this was done according to daily practice by the occupational physician after notification by the researchers. This might have caused a delay in return to work (33).
To date, the results reported in the literature suggest that graded activity might be an effective intervention for facilitating the return to work of workers with low back pain. In clinical practice, however, the structure and process of care as well as the skills of caregivers appear to be important factors to achieve an earlier and successful return to work after attending a graded activity intervention.
Dr. van Mechelen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Staal, Hlobil, Smid, van Mechelen.
Acquisition of data. Staal, Hlobil.
Analysis and interpretation of data. Staal, Hlobil, Köke, Twisk, van Mechelen.
Manuscript preparation. Staal, Hlobil, Köke, Twisk, Smid, van Mechelen.