This longitudinal prospective cohort study was embedded within a three armed RCT with an educational intervention in a primary care setting . The cohort encompasses all patients enrolled in the trial. The primary goal of the RCT was to asses the impact of treatment on functional capacity. A predefined secondary goal of the study was to explore the variation of health care services for LBP. The intervention consisted in quality circles for general practitioners (GPs) on an evidence-based LBP guideline (in both intervention arms) and in training of practice nurses in motivational counselling to promote physical activity (in one intervention arm). The guideline is in accordance with the European guidelines [17, 18]. These guidelines do not advocate acupuncture for LBP, but rather consider the treatment method as a second line intervention with uncertain effectiveness. The study was conducted in two centres (Marburg, Göttingen). Ethical approval was obtained from both study sites.
The goal was to recruit 120 practices. In all practices, both GPs and practice nurses had to agree to participate in the educational intervention, in case they would be randomized to the intervention arm with motivational counselling. We contacted 818 general practices surrounding the study centres. Addresses were obtained from local health authorities. From 118 practices who agreed to participate, 2 dropped out after randomization. The GPs were on average 12.7 years in practice (Range 1 to 31 years), the average age was 48 years (SD ± 6) (national average 50.4 years) and 42 % of them were female (national average 36 %). A total of 68 (59 %) practices were run by a single GP. The basic demographic data of our sample is not meaningfully different from the national average . Of the 116 participating practices, 25 (21 %) offered acupuncture treatment to their patients.
During the recruitment period of 8 weeks, practice nurses asked every patient with LBP to participate in the study and registered these individuals in order to estimate the number of screened patients. Patients were recruited form November 2002 to March 2003. Inclusion criteria were (1) consulting for LBP in general practice, (2) age above 18, (3) ability to read and understand German and (4) written consent.
Instruments and data collection
After written consent was obtained, baseline socio-demographic data was collected prior to the consultation with a baseline questionnaire. In addition, patients mailed a questionnaire filled in at home to the study centres. During the consultation, GPs assessed warning signs for complicated LBP ("red flags"). At follow-ups 4 weeks, 6 months and 12 months later, study nurses conducted standardised telephone interviews and patients were asked about health care utilization.
The Hanover Functional Ability Questionnaire (HFAQ) was used for the assessment of functional capacity. The HFAQ is a frequently used instrument for the assessment of back pain disability and a scale with good psychometric properties. It consists of 12 items in which patients can rate their limitation in day-to-day functional abilities. It can be compared to the Roland & Morris Scale, but is advantageous in telephone interviews . The scale ranges from 0 (extreme functional limitation) to 100 (no functional limitation), scores below 70 are considered to be a significant impairment.
In order to classify the natural history of LBP, we used a modification of the von Korff procedure as follows :
▪ Acute LBP single episode of LBP of less than 90 days duration
▪ Recurrent LBP multiple episodes LBP of less then 90 days duration within the last 12 months
▪ Chronic LBP more then 90 consecutive days of LBP within the last 12 months.
To estimate the proportion of patients with radicular symptoms, we relied on the reported level of pain radiation into the leg. We considered radiation below the knee as an indicator of possible nerve root irritation. This is a frequently used and pragmatic approach given the absence of reliable methods for assessing radicular pain in large cohorts .
For assessment of depression, we applied the German version of the Centre for Epidemiologic Studies Depression Scale (CES-D) . Scores above 23 are considered clinically relevant .
In the telephone interview, we also collected data on health care utilisation, e.g. specialist visits, medication, and non-pharmacological treatment for LBP within the last 6 months. In the interview, the study nurses actively presented a list of 42 possible interventions for LBP. They were trained in conducting standardized interviews and able to describe each method in more detail if necessary. Acupuncture sessions were not considered as consultations. We asked them whether they had received injections in the back or in the buttocks. Injections in the back most likely indicate the use of local anaesthetics with or without steroids, injections in the buttocks most likely indicate the application of an analgesic medication.
In a first step, we conducted univariate analyses in order to compare patients who received acupuncture with those who did not. In case of missing data, we provide the number of subjects analysed. Dependent on the biometric properties of the scales, we either used chi2-tests for categorical data, or t-tests and non-parametric tests (Kruskall-Wallis) for continuous data. We considered data to be statistically significant at a 5 % level. A calculation of odds ratios allowed to depict 95 % confidence intervals (CI). When data for health care utilization was missing, it was assumed that patients did not receive that particular service.
In a second step, we performed logistic regression analyses. This procedure calculates the probability of receiving or using a specific health service in relation to acupuncture. Only those factors were incorporated into the model that had proved to be significant in previous univariate tests (chronicity, duration of pain, CES-D score, radiation of pain below the knee, baseline HFAQ). Continuous data were dichotomised. For depression we used a cut-off score of >23 and for functional capacity (HFAQ) a cut-off score of > 70. Due to missing data, the models included only 1320 (98 %) patients.
Given the fact that the present study was an embedded cohort study, we also checked if one of the study arms was a significant factor in the model, which was not the case.
Comparison of consultation frequencies and duration of sick leave were adjusted with ANCOVA including the same covariates as the logistic regression models.
In a final step, we calculated a comprehensive model to predict the probability of receiving acupuncture. This model included all of the factors that had demonstrated their significance in previous univariate analyses. This model included only 1163 (87 %) subjects due to list wise deletion. The software package SAS 9.1 was used for analysis.