Abstract
Background Chiropractors throughout the world by and large focus on patients with musculoskeletal complaints who are generally in good health. Currently, it is widely accepted that neck pain and low back pain are best understood as biopsychosocial phenomena.
Objective The purpose of this study was to determine whether certain psychosocial factors provide added value for predicting recovery.
Design This was a prospective, multicenter, chiropractic, practice-based cohort study in Belgium and the Netherlands.
Methods A total of 917 participants, 326 with neck pain and 591 with low back pain, completed self-administered questionnaires at baseline and at 3, 6, and 12 months. They provided information on several demographic, biomedical, and psychosocial variables. Lasting perceived recovery was used as the outcome measure, that is, recovery at all follow-up assessments from 3 months on. Twenty-seven potential predictors of outcome were used to build the predictive model. Stepwise, backward generalized estimating equation regression models were used to take into account the clustering of participants within practices. For assessment of the added value of psychosocial variables, 2 model fit indexes were compared.
Results After the addition of psychosocial variables, predictors in the final model for neck pain included occupational status, body mass index, duration of complaints, previous treatment, and participant expectations (the model fit was marginally improved from 0.684 to 0.695 for the area under the curve and from 65.0% to 66.1% for the percentage correctly predicted). In the final model for low back pain, the selected predictors included country of treatment, age, duration of complaints, previous imaging, and somatization (the area under the curve changed from 0.669 to 0.715, and the percentage correctly predicted changed from 68.6% to 69.5%). Only a minority of participants had high scores on psychological variables.
Limitations The reliability and validity of lasting recovery as an outcome measure have not been tested. The cohort needs to be seen as a convenience sample. Selection bias, therefore, not be ruled out. There are no indications, however, that patients with complex psychosocial profiles were excluded from this study.
Conclusions Psychosocial variables provided little added value for predicting outcome in people who had neck pain or low back pain and sought chiropractic care. Therefore, chiropractors should not screen extensively for them at baseline. With regard to the identification of the small subgroup of people with high scores on psychosocial variables and a high risk for chronic pain, further investigation is needed.
In 1977, Engel highlighted the overall limitations of the traditional biomedical model of disease.1 Shortly thereafter, Waddell and colleagues2,3 made an important translation of Engel's idea to the area of low back pain (LBP). It is now widely accepted that neck pain (NP) and LBP are best understood as biopsychosocial phenomena.4–6 As a result, more emphasis has been placed on baseline screening for psychosocial aspects alongside the traditional biomedical and demographic characteristics of patients with apparently mechanical or structural musculoskeletal problems. These considerations are of particular importance to chiropractors because the vast majority of their patients have spine-related complaints and chiropractic treatments, by and large, are mechanical in nature.7,8
A 2005 survey of 1,045 chiropractors in private practice in Australasia and North America revealed that 80% to 90% of the chiropractors believed that psychological factors influence pain syndromes.9 Thus, these factors would appear to be relevant for chiropractors. However, the study by Kongsted et al10 showed that there is a low prevalence of patients with high levels of psychological risk factors or profiles in a chiropractic practice. Therefore, Kongsted et al10 suggested screening for these risk factors with a short and general tool instead of many large questionnaires with items that would be irrelevant to most patients. Given the low prevalence of psychosocial factors in chiropractic practice, it is questionable to what extent chiropractors should screen for these. More importantly, it is questionable to what extent psychological or social factors predict outcome beyond what can be gleaned from biomedical factors.
We conducted a prospective study in a chiropractic setting in Belgium and the Netherlands. For the purpose of this study, we used lasting recovery as the outcome measure. We defined lasting recovery as recovery at all follow-up assessments (6 and 12 months) from 3 months on.
The purpose of this study was to determine whether psychological or social factors add predictive value to biomedical factors known to be associated with a better or worse outcome in people with NP or LBP treated by chiropractors. In other words, can chiropractors predict, with more accuracy, whether a person will recover if they also consider certain psychosocial factors at baseline?
Method
Study Design and Population
A prospective, multicenter, practice-based cohort study was conducted for people with NP or LBP. Participants were recruited by 97 chiropractors in private practices in Belgium and the Netherlands. All participants received chiropractic care, including spinal manipulation. Treatment was left to the discretion of the chiropractors: the use of soft tissue techniques, advice about ergonomics, or instruction with regard to home exercises was allowed.
Recruitment of Chiropractors and Participants
Chiropractors.
All 72 Dutch-speaking Belgian chiropractors and all 189 members of the Netherlands Chiropractors' Association were invited to participate. They spoke the same language, with small regional differences. Recruitment was conducted in various ways: through personal email communication to all members, emails sent by the respective professional associations to all of their members, and a presentation at one of the general meetings organized by the professional associations.
Participants.
Each chiropractor was asked to recruit 15 participants who visited their practice for NP or LBP between August 26, 2010, and December 30, 2010. At the request of the chiropractors, who did not want to disturb normal office routine, the selection of the participants did not need to be consecutive. People who had shown interest in the project when they made their first appointment were contacted by one research assistant. She explained the entire study protocol over the telephone. People were considered participants after they gave consent over the telephone. In Belgium, participants also were required by the Medical Ethics Committee of the University Ghent, Ghent, Belgium, to complete and sign a written informed consent form.
Inclusion and exclusion criteria.
People who had NP or LBP and who had not been treated by a chiropractor within the preceding 6 months were eligible. In addition, they had to be between 18 and 64 years of age and had to have NP or LBP with or without radiation to an extremity as their chief complaint. Participants who had both NP and LBP were asked to indicate which of the complaints was worse. Participants needed to have a basic understanding of the Dutch language, both in reading and in writing. People were excluded from participation if they were suspected of having a tumor, fracture, or infection (“red flags”) or any other condition considered to be a contraindication for spinal manipulative therapy, such as severe osteoporosis, acute rheumatic episode, or extremely high blood pressure values.
Procedure for Data Collection and Variables of Interest
Before the start of the study, the primary investigator (L.A.) organized an information session for the participating chiropractors and their office staff to explain the research protocol and the expected office routine in detail. This information session was held in the summer of 2010 at 11 separate locations throughout Belgium and the Netherlands.
The participating chiropractors completed, for each participant at baseline, a paper questionnaire that included a description of clinical variables, the diagnosis, and the estimated or expected prognosis for each participant. A Web-based questionnaire was sent to the participating chiropractors at 12 months to document how many treatment sessions were needed and whether the participant was still under treatment. Self-administered participant questionnaires were sent at baseline; at the end of the second visit; and at 1, 3, 6, and 12 months. Participants could choose paper or Web-based questionnaires. The responses on the Web-based questionnaires were collected by NETQ (NetQuestionnaires Nederland BV, Utrecht, the Netherlands), which is an online data collection system. Paper questionnaires were returned to and processed at data collection centers in Belgium and the Netherlands.
At baseline, the following items were assessed: participant characteristics, including age, sex, country of residence, level of education, employment status, and compensation benefits as a result of the complaint11–16; biomedical items, including the duration and type of complaint, history pertaining to the complaint (including patterns of pain over the preceding year, whether it was treated by a chiropractor before, whether it was the first episode, previous treatment for the complaint, previous imaging for the complaint, previous pain in a different spinal region, previous surgery in the affected area, and ligamentous hypermobility), body mass index, and baseline pain13–22; and psychological and social factors, including distress, depression, anxiety, and somatization (assessed with the 4 scales of the Four-Dimensional Symptom Questionnaire),23,24 participants' beliefs with regard to the effects of physical activity and work on their spinal complaint (assessed with the Fear-Avoidance Beliefs Questionnaire),25 social support (assessed with the Feij Social Support Scale),26 participant expectations with regard to treatment efficacy,27 and fear of or apprehension about the therapist or treatment.28
Six pain patterns were selected; 4 were based on Dunn and colleagues' latent class analysis in characterizing the course of LBP,29 and 2 were based on the clinical expertise of 2 members of the research team. Dunn et al named the pain patterns “persistent mild,” “recovering,” “severe chronic,” and “fluctuating.” The 2 additional patterns were named “first time acute” and “episodic” (eFigure 1). These 6 patterns were presented to the participants at the first consultation; at that time, the participants selected the pattern they thought best fit the course of their NP or LBP.
The items assessed at baseline were used to build the prognostic model. Two chiropractor-specific variables were added to the model: chiropractor expectations of prognosis30 and immediate improvement reported for a participant at the first visit31; both were assessed immediately after a participant's first treatment. All variables were analyzed categorically, with the exception of age, body mass index, and Fear-Avoidance Beliefs Questionnaire (all 3 were continuous and numerical measures), current level of pain, fear of or apprehension about the treatment, and participant expectations. The latter 3 also were continuous measures and were scored on a numeric rating scale ranging from 0 to 10. On this numeric rating scale, a score of 0 represented no pain at the moment, no fear of the treatment, and low to no expectations (of the participants) about the result of the therapy.
Outcome Measure
We used lasting perceived recovery as the outcome measure. Perceived recovery was measured on a 7-point Likert scale ranging from “completely recovered” (score of 0) to “worse than ever” (score of 6). The scores were dichotomized into “recovered” (completely recovered or much improved) and “not recovered.” Participants were considered to have lasting recovery if they were reported to be completely recovered or much improved at the 3-month follow-up assessment and remained recovered at the 6- and 12-month follow-up assessments. The models were analyzed separately for NP and LBP.
Data Analysis
Baseline variables are presented as percentages for categorical variables and as means (and standard deviations) for continuous variables. For the development of the prognostic model, we followed the guidelines of Harrell32 and Royston et al.33 Candidate variables were first checked for their univariable associations with the outcome. Linearity assumptions were checked for continuous predictors; in the case of nonlinearity, predictors were recoded as categorical variables and analyzed as dummy variables.
For predictive modeling, we conducted stepwise, backward generalized estimating equation regression analyses, which helped to account for the clustering of participants within practices.34 After every step, the variable with the highest P value was eliminated from the model; the model was run repeatedly until all remaining variables scored a P value of less than or equal to .10. Model fit was assessed for the resulting model. Next, the psychosocial variables were added to the model. The model was again run repeatedly until all variables scored a P value of less than or equal to .10, and model fit was reassessed. The results of the backward regression analyses are reported as odds ratios, P values, and 95% confidence intervals.
To value the overall fit of the model and to assess the added value of the psychosocial variables, we computed 2 fit indexes. First, we assessed the area under the receiver operating characteristic curve (AUC). The AUC can be interpreted as the probability of correctly identifying a participant who showed improvement from randomly selected pairs of participants who showed improvement and stability.35 An AUC of greater than 0.70 is considered satisfactory.36 A value of 0.50 represents a model without any discriminating power, whereas a value of 1.0 represents a perfectly discriminating model. Second, we calculated the percentages of participants who were correctly classified on the basis of the predictors included in each of the 2 models. These percentages were calculated on the basis of the predicted probabilities.
The internal validity of the final models was assessed by calculation of the heuristic shrinkage factor.37 This factor quantifies the amount of possible “overfitting” present and demonstrates the likelihood that a model can reliably predict new observations.
We performed sensitivity analyses to determine how robust the findings for lasting recovery were relative to the findings for other, more traditional outcome measures (eg, recovered or not recovered at 3 months). For these analyses, we ran the same stepwise, backward generalized estimating equation regression models. We used IBM SPSS Statistics for Windows version 20.0 (IBM Corp, Armonk, New York) for all statistical analyses.
Role of the Funding Source
This work was supported financially by the European Chiropractic Union (ECU); l'Institut Franco-Européen de Chiropraxie à Paris, Paris, France (IFEC); the Chiropractic Association of the Netherlands (NCA); and the Belgian Chiropractic Union (BCU). The funding sources had no role in the design, conduct, or reporting of the study.
Results
The representation of participants at the various stages of the study and chiropractors is shown in eFigures 2 and 3. The response rates after 3, 6, and 12 months were 81.4% (746/917), 77.4% (710/917), and 73.2% (671/917), respectively. The response rates did not differ significantly between men and women, between participants with NP and participants with LBP, and between Belgian participants and Dutch participants (data not shown).
For participants with NP, 73.8%, 70.1%, and 63.9% were much improved or completely recovered at 3, 6, and 12 months, respectively. However, only 50.6% of participants with NP had lasting recovery. For participants with LBP, 66.7%, 73.2%, and 67.9% were much improved or completely recovered at 3, 6, and 12 months, respectively. However, only 51.2% of participants with LBP had lasting recovery.
The demographic, biomedical, and baseline psychological and social characteristics of participants with NP, stratified by “lasting recovery” and “nonlasting recovery,” are shown in Table 1. Participants in both groups were predominantly middle-aged and women, had a college degree, and were in good overall health, with a moderate amount of pain. The most profound differences between the groups were for the duration of complaints (twice as many acute cases with a duration of <6 weeks in the group with lasting recovery), less previous treatment and imaging in the group with lasting recovery, and half the number of markedly elevated somatization scores in the group with lasting recovery.
Descriptive Baseline Characteristics of 243 Participants With Neck Paina
Table 2 shows the demographic, biomedical, and baseline psychological and social characteristics of participants with LBP, also stratified by “lasting recovery” and “nonlasting recovery.” Participants in both groups were predominantly middle-aged and educated and had moderate pain scores. Differences between the groups were mainly in the duration of complaints (more than 50% of acute cases with a duration of <6 weeks in the group with lasting recovery), more than double the percentage of participants treated for a first episode of LBP in the group with lasting recovery, and less previous treatment and imaging in the group with lasting recovery. A minority of participants in the group with lasting recovery had markedly elevated scores on the psychological subdomains distress, depression, fear, and somatization, with twice as many in the group with nonlasting recovery. The results of the univariable analyses are shown in eTables 1 and 2. Selected predictors in the final model for NP are shown in Table 3 and included occupational status, body mass index, duration of complaints, and previous treatment (AUC=0.684, P<.001; percentage correctly predicted=65.0%). Of the 7 added psychosocial variables, only “participant expectations” was a statistically significant predictor of lasting recovery. After the addition of participant expectations, model fit improved marginally (AUC: from 0.684 to 0.695, P<.001; percentage correctly predicted: from 65.0% to 66.1%).
Descriptive Baseline Characteristics of 424 Participants With Low Back Paina
Multivariable Analysis for Lasting Recovery From Neck Paina
Selected predictors in the final model for LBP are shown in Table 4 and included country of residence, pain patterns, age, duration of complaints, and previous imaging (AUC=0.669, P<.001; percentage correctly predicted=68.6%). Of the 7 added psychosocial variables, only “somatization” was a significant predictor of lasting recovery. After the addition of somatization, model fit improved slightly (AUC: from 0.669 to 0.715, P<.001; percentage correctly predicted: from 68.6% to 69.5%), but the variable “pain patterns” was no longer a significant predictor in the model. The heuristic shrinkage factors of the final models for LBP and NP were 0.97 and 0.98, respectively, indicating a very low chance of overfitting (ie, good internal validation).
Multivariable Analysis for Lasting Recovery From Low Back Paina
Tables 5 and 6 show the results of the sensitivity analyses. After the addition of the psychosocial factors, the fit indexes for the NP model changed marginally for the AUC (from 0.729 to 0.726) and changed from 82.5% to 86.6% for the percentage correctly predicted. The fit indexes for the LBP model did not change because psychosocial factors did not contribute to the prediction model (AUC=0.696; percentage correctly predicted=62.0%). The baseline biomedical and demographic variables that were used to build the prediction model were not the same for the models with recovery at 3 months as the outcome measure and for the model with lasting recovery as the outcome measure. Only the variable “duration of complaints” was observed in all 4 models.
Multivariable Analysis for Recovery From Neck Pain, With Outcome at 3 Monthsa
Multivariable Analysis for Recovery From Low Back Pain, With Outcome at 3 Monthsa
Discussion
We conducted the present study in Belgium and the Netherlands to ascertain which—if any—psychological or social factors provided value beyond that provided by biomedical factors known to be associated with outcome in people who had NP or LBP and were treated by chiropractors. The backward generalized estimating equation regression analyses resulted in prediction models that included participant expectations for participants with NP and the degree of somatization for participants with LBP. The addition of these specific psychosocial variables did not substantially improve the model fit indexes; the main reason was that only a minority of participants had high scores on psychosocial variables. The present study demonstrated that adding psychosocial factors barely changed the chiropractors' ability to predict recovery for participants with NP and LBP. This result is in line with the findings of Langworthy and Breen38 and Leboeuf-Yde et al,39 who also reported that they did not obtain a better prediction model for people receiving chiropractic treatment for LBP after the inclusion of psychosocial factors.
We did find a small group of participants who scored high on all 4 subdomains of the Four-Dimensional Symptom Questionnaire. This result is in line and consistent with the use of simple screening tools, such as the STarT Back Screening Tool, in new stratified care models for primary care. The finding of an identifiable subgroup requiring a psychologically informed approach was reported in other studies.40–45 The differences in the extent of identification of psychosocial factors may be, in part, a consequence of differences in screening instruments. The STarT Back Screening Tool is a “simple prognostic questionnaire that helps clinicians identify modifiable risk factors (biomedical, psychological, and social) for back pain disability,”46 thereby allowing clinicians to allocate participants into 1 of 3 risk-defined groups (low, medium, and high), whereas the principal aim of the Four-Dimensional Symptom Questionnaire is “to distinguish between stress-related syndromes (denoted as ‘stress,’ ‘burnout,’ and ‘nervous breakdown’) and psychiatric disorder (ie, depression and anxiety disorders).”47
We elected to use lasting recovery as the outcome measure. To our knowledge, this is the first time that lasting recovery has been considered an outcome measure in longitudinal cohort studies of NP and LBP. We chose this novel outcome measure on the basis of discussions with practitioners in the field and with members of the research team. Spine pain fluctuates.48 Considering outcome at a single moment in time is nothing more than a “snapshot”; therefore, such an outcome may not reflect real change. Axén et al48 suggested that to accurately describe a clinical course in a fluctuating condition such as spinal pain, more dense data collection may be needed over a fairly long period of time. Although Axén et al used data collected on a weekly basis over a period of 6 months to draw their conclusion, we believe that recovery at the 3 consecutive follow-up assessments that we proposed (3, 6, and 12 months) also would represent a reliable reflection of participants' actual recovery status.
Our study illustrated the discrepancies between a single measurement and outcome measured over time. For participants with NP, 73.8% were reported to be much improved or completely recovered at 3 months. However, only 50.6% of participants with NP had lasting recovery. For participants with LBP, 66.7% were reported to be much improved or completely recovered at 3 months, but the percentage declined to 51.2% when lasting recovery was considered the outcome measure. To challenge the robustness of the models in which lasting recovery was used as the outcome measure, we performed sensitivity analyses (Tabs. 5 and 6).
For the most part, baseline demographic and biomedical variables were not highly predictive of outcome in participants with NP and participants with LBP, with AUCs of 0.684 and 0.669, respectively. The addition of participants' psychosocial characteristics did not substantially improve the clinicians' ability to predict outcome, although the use of the AUC as a fit index for the model for LBP provided satisfactory values after the addition of somatization. Depending on the outcome considered, the variables constituting the prediction models varied. The most consistent predictor of improvement was duration of complaints; participants in the acute phase (<6 weeks of complaints at baseline) had a better prognosis. This finding is consistent with the literature.49,50 In the model with recovery at 3 months as the outcome, psychosocial factors did not serve as predictors for participants with LBP. In the model with lasting recovery as the outcome, the addition of psychosocial variables improved the fit indexes only marginally. With the exception of the model for NP with recovery at 3 months as the outcome, for which both fit indexes had acceptable values, and of the AUC for the model for LBP with lasting recovery as the outcome, all models introduced here had low to moderate overall predictive values. Therefore, predicting outcome and lasting recovery in the study participants proved to be very difficult.
A strength of the present study is that we examined a large cohort of participants recruited from many chiropractic practices in Belgium and the Netherlands. Furthermore, treatment was left to the discretion of the chiropractors, thereby reflecting real situations in clinical practice. As a result, our findings could be considered broadly generalizable. In addition, we examined a broad set of psychosocial variables for participants with NP and LBP by using validated instruments; therefore, our findings should also be viewed as comprehensive. We used questionnaires that had been validated and applied in previous studies. This approach allows for pooling of data from studies in which similar questionnaires were used. The fact that we resolutely opted for an outcome different from that used in similar studies in the past but more reflective of what chiropractors and patients assume to be a realistic goal also could be seen as a strength. Patients want to get better as soon as possible, and chiropractors want their treatments to be effective for the long term as well; therefore, considering lasting recovery (after 3 months) as the outcome in the present study seems defensible. We suggest that this outcome measure be used in future research, particularly in patients with acute NP or LBP (duration of <6 weeks) at baseline. However, the use of a different outcome measure hampered the ability to pool data from the present study with data from previous studies in the same setting. Therefore, data for recovery at 3 months are shown in Tables 5 and 6.
A limitation of the present study is that the reliability and validity of lasting recovery as the outcome measure have not been tested. The fact that we allowed chiropractors to select participants at their convenience to avoid disturbing normal office routine might have led to selection bias. People with high levels of psychological distress were not excluded from the present study, and we do not have any indication that they were less likely to be asked to participate. Because the participant was not asked about psychological distress when he or she made the appointment over the telephone, the chiropractor would have had no idea whether the participant had psychological issues. Whether participants with high levels of psychological distress were less likely to participate and, therefore, were overrepresented in the group of 88 people who did not want to participate is uncertain. However, the present study involved a large cohort of participants (N=917), and the participants had characteristics similar to those in previous studies in a chiropractic setting in Europe.51,52
Another limitation of the present study is that at the end of the 12-month study period, only 73.2% of the participants remained in the study. Participants lost to follow-up had comparable baseline characteristics (eTables 3 and 4), with the exception of markedly elevated scores for somatization in participants with NP: the percentage of participants with markedly elevated scores for somatization in the group of participants who dropped out was twice as high as that in the group of participants with nonlasting recovery and 4 times as high as that in the group of participants with lasting recovery. For both NP and LBP, participants who dropped out had very similar distributions with regard to duration of complaints and had predominantly chronic pain.
We did not examine the external validity of the present study; however, considering the fact that the population of participants studied here reflects well the population of patients seen by chiropractors in Europe51 and North America,52 we believe that our findings can be generalized to other patients seeking chiropractic care and perhaps to patients who have NP or LBP and are seeking care from other primary care providers. Chiropractors worldwide generally focus on patients who have musculoskeletal complaints, are generally in good health, and have low scores on psychosocial subdomains. Most often, the treatments used are manipulative or mechanical in nature.8
In conclusion, psychological and social variables have little added value for predicting outcome in people seeking care from a chiropractor for NP or LBP. Therefore, we advise chiropractors not to use extensive screening tools to collect baseline information on the psychosocial status of their patients. However, a minority of participants in the present study scored high on psychological variables. In new stratified care models (eg, STarT Back Screening Tool) for primary care, simple screening tools have been used to identify this high-risk group. With regard to the identification of the small subgroup of people with psychosocial problems and a high risk for chronic pain, further investigation in a chiropractic setting is needed.
Footnotes
All authors provided concept/idea/research design and writing. Dr Ailliet provided data collection. Dr Ailliet, Dr Rubinstein, and Dr Hoekstra provided data analysis. Dr Rubinstein, Professor van Tulder, and Professor de Vet provided project management. Dr Ailliet, Dr Rubinstein, and Professor de Vet provided fund procurement. Dr Ailliet and Dr Rubinstein provided participants. Dr Rubinstein and Professor de Vet provided facilities/equipment. Professor de Vet provided institutional liaisons. Dr Rubinstein, Dr Hoekstra, Professor van Tulder, and Professor de Vet provided consultation (including review of manuscript before submission). The authors thank Virginie Colpaert for contacting all participants by phone and explaining the entire research protocol in detail each time.
This article was part of the PhD project of Dr Ailliet.
This study was approved by the medical ethics committees of the University Ghent, Ghent, Belgium, under registration number B67020095664 and VU University Medical Center, Amsterdam, the Netherlands, under reference number 08/232.
This work was supported financially by the European Chiropractic Union (ECU); l'Institut Franco-Européen de Chiropraxie à Paris, Paris, France (IFEC); the Chiropractic Association of the Netherlands (NCA); and the Belgian Chiropractic Union (BCU).
- Received May 27, 2015.
- Accepted January 25, 2016.
- © 2016 American Physical Therapy Association
References
Issue highlights
