The MuIS changed general practice referral behaviour compared to the PaIS.
We used a combination of outcomes recommended for studies on LBP [
13]. The patient-reported outcomes were validated measures, whereas questions regarding satisfaction with the received treatment and satisfaction with treatment outcomes were formulated to fit our setting. The fact that referrals to secondary healthcare could be assessed by registry data allowed a 100 % follow-up rate of the included patients for the primary outcome and for the cost-effectiveness analysis. The change in data source for the primary outcome is a limitation. Use of registry data might have introduced uncertainty in the estimates due to difference in registration practices among different hospital departments, varying completeness of registration, as well as by changes in registration practice over time. However, we found it unlikely that this should have affected the two allocation groups unevenly in our study. Hence, we did not expect bias to be introduced by the change in data source. This study was powered to recruit 2700 patients from 100 practices. The recruitment for this project unfortunately coincided with a conflict between the Danish regions and the Organization of General Practitioners in Denmark, and this affected GPs’ willingness to participate. Consequently, after 15 months, the inclusion of 60 practices and approximately 1200 patients was accepted. This decision was made while assessors were still blinded and before registry data were collected. We consider the risk of bias caused by a smaller sample very unlikely given the large effect size (5.0 vs. 10.5 %). However, generalisability might be reduced. The response rate for the secondary outcomes was low, and with responders representing 50 (83.3 %) of the practices, this may harm the validity of the secondary outcomes. The probability of referral from primary care to secondary care within 12 weeks was used as an effectiveness measure in the cost-effectiveness analysis. It is a limitation in the economic evaluation. This might be considered double counting as costs related to primary and secondary healthcare services are included in the denominator of the incremental cost-effectiveness ratio. A more appropriate effectiveness measure could be QALYs. By participating in this study, the included practices agreed to receive the intervention, which included an outreach visit while other practices might have declined participation because of resistance to elements within the MuIS. Subsequently, practices participating in the study were generally more likely to have had an outreach visit from the regional quality unit the year prior to the study compared with the non-participating practices. This might imply that the participating practices were more likely to work with quality improvement compared with non-participants.
Reviews did not offer compelling evidence for the superiority of multifaceted interventions compared to single-component interventions [
19]. Several randomised controlled trials have studied the implementation of LBP clinical guidelines in general practice using different strategies [
20,
24]. The use of physiotherapists as facilitators in general practice was novel in our study. A previous randomised study used outreach visits to raise the awareness of LBP guidelines in 24 centres with 2187 patients, but the management of patients was mostly unchanged by those outreach visits [
20]. In another large cluster randomised controlled trial with 118 general practices and 1378 patients, the effect of two multifaceted implementation strategies was compared to a postal dissemination of guidelines. No effect on patients’ functional levels was found when using an implementation strategy including four basic education modules and flyers for patients versus postal dissemination of the guidelines. When adding motivational counselling to patients (each patient had up to three sessions of 10–15 min), a small significant difference was found in patients’ functional level, measured by the 12-item Hannover Functional Ability Questionnaire [
21]. As with our study, these two studies included outreach visits at GPs’ work environments, which, in our setting, typically took place during an extended lunch break. This may have contributed to a relaxed and informal learning environment. Other studies have used workshops to implement LBP guidelines. A Dutch guideline implementation study with 67 GPs and 531 patients included a 2-h educational and clinical practice workshop in addition to a screening tool for patients with LBP and a tool for patient education. The intervention succeeded in reducing referrals from general practice to therapists (physical, exercise, or manual therapists) [
22]. In the IMPLEMENT study, the researchers found a change in clinician behaviour (knowledge, attitudes, and intentions) among 92 general practices offering two facilitated interactive workshops in a total of 6 h with the purpose of decreasing X-ray referrals and increasing advice to stay active [
23]. However, the change in behavioural attitude was not reflected in a difference in the actual referral rate to X-ray or CT-scan amongst GPs receiving the intervention compared with GPs receiving the usual dissemination strategy (access to guidelines) [
23]. An additional study with 462 GPs and their LBP patients with accepted compensation claims were unsuccessful in improving concordance with Canadian LBP guideline recommendations. The intervention consisted of a passive knowledge transfer method that involved postal letters with guidelines and reminders [
24]. In contrast, a successful intervention included a clinical decision support system as part of a multifaceted strategy, together with quality reports and peer-to-peer consultations in a large cohort study with 1200 GPs and 23,685 patients. This multifaceted strategy was found effective in reducing MRI referrals from 5.3 to 3.7 % [
25]. Our study likewise found a high effect size in the rate of referral following a broad MuIS that included clinical decision support, feedback (statistics regarding LBP patients), and outreach visits. Compared with the other trials aimed at GPs, our intervention dose (consultation time with clinicians) was slightly below average. GPs in the MuIS group could use the STarT Back Tool to categorise patients into (i) low risk patients, where advice and information can stand alone, (ii) medium risk patients with extra needs for exercise treatment or manual therapy and who may benefit from a referral to physiotherapy or chiropractic treatment, or (iii) high risk patients with an additional need for addressing psycho-social barriers for recovery. This tool has been found both effective and cost effective as a tool for subgrouping LBP patients and targeting their treatment [
11,
26]. The STarT Back Tool is now widely used, and as of November, 2015, it has been translated into Danish and 20 other languages [
27,
28]. The STarT Back Tool was known by many physiotherapists in Denmark and was described in the Regional LBP guideline. However, training in the stratified care management programme developed at Keele University was not generally available in Denmark. Hence, to our knowledge, none of the Regions’ GPs or physiotherapists had been trained in the stratified management programme from Keele University. The GPs’ and physiotherapists’ knowledge about the STarT Back Tool has been adapted from scientific journals, from discussions with colleagues, at conferences, from the regional LBP guidelines, and, probably most importantly, from the outreach visits at the MuIS practices. In addition to the STarT Back Tool, GPs in the MuIS group had additional questions regarding psycho-social risk factors built into their electronic medical record system regarding work problems, compensation claims, and psychological or social barriers to recovery. The tools available to the GPs may be viewed in the context of the coloured flags [
3]. In combination, the STarT Back Tool and the questions about psycho-social risk factors incorporate biological, psychological, and social aspects. Patients with red flags (serious pathology) were excluded from our study. Yellow flags (beliefs, emotional responses, and pain behaviour) were addressed by the STarT Back Tool. Blue flags (perceptions about the relationship between work and health), black flags (rehabilitation/compensation system or contextual obstacles, such as legislation and injury claim conflicts), and orange flags (psychiatric factors) were encompassed by the additional questions regarding psycho-social problems. Satisfaction was planned to be analysed as a continuous variable but a non-parametric distribution of data led us to dichotomise the two satisfaction variables and this may have led to loss of information regarding satisfaction. We found a reduced satisfaction among patients in the MuIS group. This discrepancy between functional outcome measures and patients’ satisfaction has also been reported by Takeyachi et al. [
29]. We do not know why patients in the MuIS group were more dissatisfied than patients in the PaIS group. One reason could be GPs advice to stay active regardless of patients’ pain, or that patients with low risk were recommended minimum treatment, or it could be related to unfulfilled expectations induced by GPs’ information of an expected good prognosis.