The study was an observational cohort study reported according to the overall recommendations from ‘STrengthening the Reporting of OBservational studies in Epidemiology’ (STROBE) [16].
Setting
This study was conducted using a subsample of data from the Danish Chiropractic low back pain Cohort (ChiCo) collected from November 1st 2016 to September 6th 2018, which was prior to the completion of enrolments in ChiCo. ChiCo is a longitudinal observational cohort of chiropractic patients recruited from 10 private chiropractic clinics in the Central Denmark Region. Patients comprising the cohort responded to questionnaires at their initial visit for an LBP episode and after 2 weeks (unrelated to this study), at 3 months and 12 months. They received treatment at the discretion of the chiropractor. Treatment was not affected by study participation. A Danish version of the BBQ was incorporated into the ChiCo at baseline, at 3-month and 12-month follow ups. Data collection was performed electronically using REDCap licensed by the Odense Patient Explorative Network (OPEN) [17]. ChiCo data are stored and managed at the Nordic Institute of Chiropractic and Clinical Biomechanics (NIKKB) with the University of Southern Denmark (SDU) as the responsible data authority (Danish Data Protection Agency, j.nr.: 2015-57-0008/16–47,215).
Participants
To be eligible for inclusion, patients needed to be above the age of 18 and to consult the chiropractic clinic with a new episode of LBP, including both non-specific LBP and LBP with radicular pain. In this context we defined a new episode as initiating treatment for LBP and patients already in a course of treatment were not eligible. Patients were not included if LBP was suspected to be caused by serious pathology or immediate referral for surgery was required, this would also mean exclusion if occurring after study participation had started. Furthermore, the patients needed to understand and read Danish and have access to email.
Data collection
The receptionist screened patients for inclusion criteria, informed them about the study and invited those eligible to participate. The baseline survey was divided into two parts. The first part involved the patient filling out the questionnaire on a tablet in the waiting room just before their initial consultation with the chiropractor. Written information about study participation and rights of participants was provided at the beginning of the survey and additional information was provided by the chiropractor during the consultation. On the day of the consultation, participants received a link to the second part of the baseline survey and a link to follow-up surveys was mailed after 3 months and 12 months. Within a few days after enrolment in the study, a research assistant called the participants to welcome them to the study, answer practical questions about participation, repeat the rights related to study participation and remind them to fill out the second part of the baseline survey if they had not already done so.
Variables
Back belief questionnaire
The translation of the BBQ was conducted as recommended by forward and back translation [18]. The forward translation was performed by two persons with Danish as their first: A back pain researcher who is familiar with English as working language and a layperson who has a master’s degree in English Literature. After the independent translations of the questionnaire, the translations were compared, and a common version agreed on. The back translation was performed by two persons who are native English speakers and have lived in Denmark and used Danish for more than 10 years. One is a back-pain researcher and one is a layperson. The wording of the back translated version was compared to the original BBQ and two of the authors (AK and TSJ) decided on the pre-final version. The pre-final version was tested among a group of 10 people, five with content knowledge and 5 people without. Based on comments from this group, the final version was decided on by AK and TSJ. The internal consistency of the scale was tested in the study sample (see “Statistical Analyses”). Additional validation of the translated BBQ has not been performed. The original BBQ has previously been validated by the developers who found it to be one-dimensional and showing acceptable internal consistency and reliability [7]. Similar results were found in a validation study of the BBQ providing evidence supporting the structure of the BBQ [8].
The BBQ was answered in the first part of the baseline survey and again after 3 and 12 months. It consists of 14 statements regarding perceived inevitable negative consequences of an episode of LBP with five of these acting as distractors. Each statement is rated on a five-point Likert scale scored from 1 (completely disagree) to 5 (completely agree). Scores are then reversed and summed up to a final score ranging from 9 to 45 with lower scores indicating more negative beliefs about back pain.
To our knowledge, there is no consensus on the cut-point for negative versus positive back beliefs. In this study, we chose a sum score above 27 to indicate positive beliefs, as this cut-point was used in a recent systematic review on back beliefs [6]. When assessing individual items, we interpreted a score of 1 or 2 (on a reversed scale) as agreeing with the statement, as this method has been used previously [13].
Additional baseline variables
From the first part of the baseline survey, the following measures were used: age and sex (from the patient’s personal identification (social security) number); duration of current pain episode (1–2 days, 3–7 days, 1–2 weeks, 2–4 weeks, 1–3 months, 3–12 months, more than a year); number of days with pain last year (≤30 days, > 30 days); back pain intensity (Numerical Rating Scale (NRS) asking about ‘typical back pain’ the previous week. 0 = No pain; 10 = Worst imaginable pain); disability measured by the 23-item Danish Roland-Morris Disability Questionnaire (RMDQ, scale 0–23) [19]. The NRS and the RMDQ have been used widely in previous research and have been validated [19,20,21]. Previous episodes of LBP (none, 1 episode, 2–3 episodes, more than 3 episodes); whether or not the patient had attended other care for their current LBP episode (none, general practitioner, physiotherapist, another chiropractor, other) and whether or not the patient had previously attended a clinician for LBP (no care-seeking, general practitioner, physiotherapist, chiropractor, other clinician) were from the second part of the baseline survey.
BBQ was filled out in the first part of the questionnaire before the patient’s initial consultation with the chiropractor in order to prevent potential impact on back beliefs resulting from the contact. The additional baseline variables were chosen because they were considered likely to influence back beliefs which we have seen in previous studies using similar variables [7, 8, 10,11,12,13].
Study sample
The sample size was decided based on other purposes of the cohort. To be able to detect a small to moderate effect size (Cohen’s f = 0.10) in the present study using linear regression (F-test) with a power of 0.90 at the 0.05 level of significance, a sample size of n = 146 was required for categorical variables with four levels [22]. Still, we used the full cohort to obtain as precise estimates as possible. The minimum of observations at any time was n = 1613 (the regression analysis testing the association between BBQ and more than 3 episodes of LBP) and the smallest observed category in any analysis was n = 143 (other treatment for current LBP provided by a chiropractor, Table 3) .
Statistical methods
Baseline characteristics were presented as means with standard deviations (SD) or proportions. Since enrolment for the ChiCo cohort was still ongoing, not all participants had reached 3- and 12-month follow ups, and follow-up rates at 3 and 12 months were calculated from the number of participants who had received the follow-up questionnaires. We used a Wilcoxon signed-rank test to test the difference in baseline BBQ sum scores between follow-up study populations and non-responders.
To confirm that the BBQ represented one latent variable, we calculated Cronbach’s alpha. All nine items of the BBQ had an alpha score above 0.7 and a total alpha score of 0.75, indicating acceptable internal consistency.
The BBQ was analysed in separate analyses for each of the three time points (baseline, 3 months, 12 months). First, we dropped observations with six or less answers of the BBQ items. For the remaining observations, we used chained multiple imputation to fill out missing items of the BBQ. The chained multiple imputation was based on the 14 BBQ items, back pain intensity and number of days with pain last year. Because the number of imputations was low, we extracted one dataset out of five imputed versions for the analyses instead of conducting analyses across multiple datasets.
Data on the BBQ was not normal distributed and back beliefs were presented at each time point as median sum scores with interquartile range (IQR), histograms of the BBQ sum score and as means for the individual items. We also calculated the percentage of participants who agreed with each item at each time point. To investigate the association between back beliefs and patients’ baseline characteristics and symptoms, we used univariate linear regression with the BBQ sum score at baseline as the dependent variable and other baseline characteristics as explanatory variables. Pain and disability were categorised by dividing the scores into quantiles and age was divided into three categories based on the inspection of a LOWESS plot (Locally Weighted Scatterplot Smoothing). Categorical explanatory variables were introduced as indicator variables using the lowest category as the reference (for example each of the 2nd to 4th quartile of pain scores were compared to the 1st quartile). Results were presented as regression coefficients with 95% confidence intervals and p-values. The linear relationship between pain intensity on a continuous scale and the BBQ were checked in a LOWESS plot. Testing the assumption of homoscedasticity, the Breusch-Pagan test that did not make us reject the null-hypothesis of equal variance (p = 0.53).
At the 3-month and 12-month follow ups, BBQ sum scores were described separately for people who reported LBP and for those who did not. Reporting LBP was defined as scoring 1 or higher on the NRS at the respective time points. The statistical significance of group differences was tested using the Wilcoxon signed-rank test.
All analyses were performed using Stata/MP 15.1 (StataCorp LLC, TX 77845, USA).