We used baseline data from the Danish Chiropractic LBP cohort (ChiCo) study to undertake a cross-sectional analysis. This is the second of two analyses we have undertaken, the other is published elsewhere [16]. ChiCo is a prospective longitudinal observational study with one year follow-up performed between November 2016 and December 2019 [17]. As the ChiCo study followed ethical regulations [17] and only anonymised data were provided for use in this study, the Macquarie University Human Research Ethics Committee provided a waiver to use the ChiCo data without further ethical review. The Health Research Committee of Southern Denmark determined that the ChiCo project did not require ethical approval (S-20.162.000-109). Storing and processing of personal data was registered with the Danish Data Protection Agency via the University of Southern Denmark’s joint registration system (2015-57-0008; file # 16/47215). Reporting of this paper is in accordance with the STROBE statement (Additional file 1).
Cohort
As described elsewhere [16, 17], the ChiCo cohort included 2818 patients who presented to chiropractors in Denmark. To be included, patients needed to: have a new or recurrent episode of LBP, with or without leg pain; be 18 years old or over; and, be able to complete electronic questionnaires in Danish. Patients with a confirmed diagnosis of fracture, infection, cancer, or other serious pathology were excluded, as were patients who were in an ongoing course of treatment or long-term management. When determined to be eligible, participants completed an initial questionnaire before they underwent a baseline clinical assessment with the treating chiropractor, who then completed a clinical assessment questionnaire. Following the assessment, participants completed a second baseline questionnaire. Treatment provided by the chiropractors was as needed and not influenced by participation in the study. No limitation was placed on access to other healthcare services. For the current study, all participants were eligible for inclusion for Aims 1 and 2. For Aim 3, participants who were identified by the chiropractor to have previous imaging relevant to the presenting LBP complaint were excluded from analysis.
Data collection and outcome measures
Questions regarding imaging beliefs and desire for imaging were collected in the initial baseline questionnaire, completed before the clinical assessment. Participant beliefs regarding the importance of imaging were assessed by two questions used in previously published research [6]. The questions asked participants to rate their agreement on a five-point Likert scale, from strongly disagree to strongly agree, to the statements: (1) ‘X-rays or scans are necessary to get the best medical care for LBP’ and (2) ‘Everyone with LBP should have spine imaging (e.g. X-ray, CT, or MRI)’. Responses were dichotomised for analysis into those that agreed or strongly agreed with the statement compared to all other responses. Participants were then asked “What do you want from your visit with the chiropractor? You can mention things you want even if you are uncertain that they will be part of the visit”. This was followed by a list of items that could be answered “yes” or "no”, one of the items being: ‘Radiography or MRI will be performed or I will be referred for it’. Finally, whether participants were referred for imaging in the initial assessment was collected in the clinical assessment questionnaire. Chiropractors were asked to respond (yes/no) to three items: ‘Will the patient be referred to radiography’; ‘Will the patient be referred to CT’; ‘Will the patient be referred to MRI’. The participant was considered to have been referred for imaging if a response of ‘yes’ was received for any of these three items.
Covariates were assessed in the two participant baseline questionnaires and the clinical assessment questionnaire completed by the chiropractor. The initial baseline questionnaire (completed before the clinical assessment) included: age; sex (male/female); LBP intensity, measured on a 0–10 numerical rating scale as average or typical pain over the previous week; low back disability, measured using the Danish 23-item Roland Morris Disability Questionnaire (RMDQ) with results proportionally recalculated to a score from 0 to 100; duration of current episode of LBP (1–2 days; 3–7 days; 1–2 weeks; 2–4 weeks; 1–3 months; 3–12 months; more than 1 year); previous imaging for back pain (yes/no); and back pain beliefs measured using the Back Beliefs Questionnaire (scored from 9 to 45, lower score indicating more negative beliefs) [18]. The second baseline questionnaire (completed after the clinical assessment) included: education level (eight categories from no education to higher education, more than 4 years); and previous LBP (none, 1 episode, 2–3 episodes, more than 3 episodes). The LBP diagnosis determined by the chiropractor (non-specific LBP, spine-related leg pain with nerve root involvement, spine-related leg pain without nerve root involvement, or suspected fracture, infection, cancer, cauda equina syndrome, inflammatory arthritis) was collected in the clinical assessment questionnaire.
Data analysis
Multiple imputation was performed using SPSS (IBM Corp. Released 2017. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.) for covariates where there was less than 5% missing data (LBP intensity, LBP duration, back beliefs) and for two items from the second baseline questionnaire which was not completed by 27% of participants (previous LBP, education level). Missing data for outcome or exposure variables, including beliefs regarding the importance of imaging, whether participants wanted imaging, and referral for imaging were not imputed, and participants with missing data were removed from the analysis.
Sample size calculations were performed for Aim 3 as a smaller subset of the ChiCo cohort (2053/2818 participants) was available for analysis due to exclusion of participants with previous relevant imaging and missing outcome data. Sample size calculations were performed in GPower 3.1.9.2. Assuming an imaging frequency of 25% [1] and a high correlation between covariates and the outcome measure, the available sample size would have 80% power to detect an odds ratio of 1.9.
Participant beliefs of the importance of imaging (Aim 1) were presented descriptively as the percentage of respondents for each Likert scale category (strongly disagree to strongly agree) and the dichotomised responses (agree and strongly agree compared to all other responses) with 95% confidence intervals. The number of patients wanting imaging (Aim 1) was presented as a percentage with 95% confidence interval.
Univariable and multivariable logistic regression analyses were performed to assess for association between the pre-selected clinical and demographic factors and participant beliefs of the importance of imaging and expectations for imaging (Aim 2). No additional confounder variables were selected. The nine baseline factors were selected by author consensus and informed by previous research [6], and comprised: age, sex, LBP intensity, low back disability, duration of current episode of LBP, previous imaging for back pain, back beliefs, education level, and previous LBP. Pain duration was dichotomised for analysis into four weeks or less or more than four weeks. Education level was dichotomised into below further education (no education, primary school, youth education, vocational education) or further education and above (short further education, middle further education, higher education). Previous LBP was dichotomised into yes (any episodes of previous LBP) or no. Three models were created, each using the same clinical and demographic baseline factors as independent variables and the dichotomised responses to the two imaging beliefs questions and whether patients wanted imaging as dependent variables respectively. Odds ratios and 95% confidence intervals were calculated for the variables in each model. Multicollinearity of the independent variables was assessed, and any variable with a variance inflation factor (VIF) of three or more was removed from the models.
Whether patients who believed imaging to be important or wanted to receive imaging were more likely to get an imaging referral was investigated using univariable (unadjusted) and multivariable (adjusted for pre-selected potential confounders) logistic regression analysis (Aim 3). The pre-selected confounders were determined by author consensus as variables likely to be associated with the exposure (beliefs about imaging) and the outcome (imaging referral), and comprised: age, sex, LBP intensity, LBP duration, LBP disability, previous imaging for LBP, previous episodes of LBP, participant diagnosis, and practitioner imaging frequency. Pain duration (4-weeks or less/more than 4-weeks), previous episodes of LBP (yes/no), and participant diagnosis (suspicion of serious pathology/no suspicion of serious pathology) were dichotomised for analysis. Practitioner imaging frequency was calculated from the frequency with which practitioners ordered imaging for patients within the ChiCo cohort. A new item was created with four categories selected to reflect a range of lower to higher frequency of practitioner imaging referral as seen in the literature [1]: less than 15% patients referred for imaging, 15–25%, 26–40%, or more than 40%. The presence of leg pain (no leg pain/leg pain without neurological symptoms/leg pain with neurological symptoms) was added as a potential confounder after the initial analysis. The exposure variables included participant beliefs about the importance of imaging and whether patients wanted imaging.
Correlation between the two imaging beliefs questions and whether patients wanted imaging were calculated using SPSS (IBM Corp. Released 2017. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp.), with strong correlation between the two imaging beliefs questions (r = 0.7) and weak correlation between either of the imaging beliefs questions and the question on whether patients wanted imaging (r = 0.3 for each). As the imaging beliefs questions were highly correlated, the scores for the two items were summed to create a single imaging beliefs item on a 2–10 point continuous scale. The single summed beliefs item was then dichotomised for analysis into strong beliefs of the importance of imaging (score of 8–10, where imaging belief questions were either answered with a combination of agree or strongly agree on both questions, or strongly agree on one question and neutral agreement on the other) or uncertain beliefs about the importance of imaging (score of 2–7). Any participants with missing data for either of the imaging beliefs questions were removed from the dataset before the items were summed. Two models were created, one using the summed and dichotomised imaging belief questions, and the other the using the single (Y/N) question on whether participants wanted imaging as independent exposure variables respectively. Odds ratios (OR) and 95% confidence intervals (95%CI) were calculated for the exposure variable in each model. The dichotomised measure for imaging beliefs was selected for analysis to aid interpretation of results. Sensitivity analysis was performed using the summed imaging beliefs measure as a continuous variable to assess for consistency of results. Multicollinearity of the exposure and confounder variables was assessed, and any variable with a VIF of three or more was removed from the models.