The effectiveness of shared decision-making followed by positive reinforcement on physical disability in the long-term follow-up of patients with nonspecific low back pain in primary care: a clustered randomised controlled trial

The aim is to assess the effectiveness of shared decision-making followed by a positive reinforcement of the chosen therapy (SDM&PR) on patient-related clinical outcomes in patients with non-chronic low back pain in general practice.

A cluster-randomised controlled trial was performed in the practices of 68 general practitioners (GPs) in the academic primary care network around Utrecht in the Netherlands.

GPs were recruited between August 2009 and May 2011. Each participating GP was requested to include ten patients with non-chronic nonspecific low back pain.

The inclusion criteria were as follows:

The exclusion criteria were as follows:

Because the causes and pathophysiology of low back pain might be different in patients younger than 18 or older than 65 years, those who are pregnant or in those with a longer disease duration, we excluded these patients [1, 21].

GPs were randomly assigned to the usual care (UC) group or the intervention (IV) group immediately after consenting to participate in the trial. Randomisation was done by research staff members who were not otherwise in the research project. Allocation was blinded using allocation cards in sealed envelopes in an initial block of 40 followed by blocks of ten envelopes. GPs in the control group were kept unaware of the communicative techniques that were trained. Auxiliary staff members recruited the patients. Patients and auxiliary practice staff members were not informed about the allocation of the GP or about the communicative techniques in the training programme. Auxiliary practice staff members collected questionnaires from the patients after inclusion. A follow-up questionnaire with a pre-paid envelope was given to each patient with instructions on when to complete it and send it to the research team. Patients were reminded to send the questionnaires two, six, twelve and twenty-six weeks after the consultation by email or phone just before the correct time and, if necessary, again two weeks later.

GPs in the intervention group were trained to perform SDM&PR during their consultations with the included patients. SDM followed the following process steps: inform the patient about therapeutic options, discuss the patient’s preferences, concerns and expectations, confirm the patient’s understanding, assess the patient’s preferred level of involvement in decision-making and finally make a joint decision about the optimal therapeutic regimen. GPs were trained to positively reinforce treatment outcomes after SDM.

GPs in the intervention group received two training sessions of two and a half hours. Training sessions were held in small groups of approximately three to five participants and were given by a peer GP with expertise in training SDM skills (AS).

The training was based on the learning principles of Kolb and the behavioural process elements of Elwyn [22]. To support SDM performance during consultations, the participating GPs received a desktop card summarising all consecutive process elements for SDM and a decision aid specifically developed for this trial according to the International Patient Decision Aids Standards (IPDAS)-guidelines (Additional file 1 Appendix 1) [23, 24]. Finally, they received individual feedback on their SDM performance based on observation by the trainer (AS) of videotapes of the consultation of each included patient. Details of the training are reported elsewhere [23]. The fidelity of the intervention was checked by measuring behavioural changes and consultation duration differences between the intervention and the control group using the OPTION instrument on videotaped consultations [23].

In the control group, the GPs provided the usual standard of care. Although routine management was not predefined in the instructions for the study, GPs in the Netherlands are reported to follow the professional guidelines on low back pain in 70% of patients [25]. Discussion of the favourable prognosis of low back pain is part of the suggested management in the guideline, but SDM is not [2].

The primary outcome was the difference between the intervention and the control group in the course of functional disability during the six-month follow-up. Functional disability was measured daily during the first two weeks and at two, six, twelve and twenty-six weeks after the first consultation.

As secondary outcomes, we assessed the difference in functional disability at the time of each of the separate measurements (2, 6, 12 and 26 weeks), the difference in severity of back pain and the percentage of patients with adequate relief on separate measurement dates and at the end of the study.

As indicators of economic effect, we evaluated the differences in absenteeism and health care consumption between groups over the complete study period and on the separate measurement dates.

To be able to test for potential confounding, we measured illness perceptions at the baseline. To check the fidelity of the intervention, we questioned patients after the consultation about the level of involvement in decision-making.

Functional disability was assessed by the Dutch version of the Roland-Morris disability questionnaire (RMD). This validated questionnaire contains 24 closed questions about restrictions in daily activities during the previous day. The score is the total number of positive answers [26]. Pain severity was quantified by the validated continuous visual analogue scale (VAS), in which the patients indicate the level of pain during the past week on a continuous line that ranges from zero (no pain) to ten cm (the most terrible pain I can imagine), with outcomes measured in mm. [27]. Adequate relief (AR) of pain was assessed with one closed question referring to the recovery experienced since the previous questionnaire and was expressed as the percentage of patients with AR in each group.

Absenteeism was measured by the response to a question referring to the time before the baseline or since the previous questionnaire: ‘Because of my back pain, I refrained from work (absenteeism),’ yes/no/not applicable (expressed as a percentage of patients), followed by an inquiry about the number of days of sick leave.

Healthcare consumption data were derived from patient questionnaires by counting follow-up contacts, via telephone or at the practice, and expressed as the mean number of contacts per patient since the previous questionnaire.

Illness perceptions were assessed by the Dutch version of the abbreviated illness perception questionnaire (IPQ) [28]. This instrument measures eight separate dimensions of perceptions about low back pain using a scale rating from zero to ten. This instrument has been proven to be valid [29].

The actual level of shared decision-making, as experienced by the patient, was evaluated by their response to one simple question immediately after the consultation: ‘Were you involved in decision-making?’ Mean scores were calculated from a range of one to four points corresponding to the answers ‘no’, ‘mostly no’, mostly yes’ or ‘yes’.

The observed effects of the training were reported in a separate article on the evaluation of the training [23].

The primary outcome (RMD) and the VAS was assessed daily during the first 14 days by a diary and at two, six, twelve and twenty-six weeks after consultation by questionnaires. All other secondary outcomes were assessed at two, six, twelve and twenty-six weeks (Fig. 1). Baseline measurements, potential confounders and the manipulation check were assessed through questionnaires completed by all patients before and immediately after the consultation.

To reach a minimum standardised difference of 0.3 in the primary outcome between the intervention and control groups, which is more than 1 point on RMD scores with a standard deviation of 5, using a beta of 0.80 and an alpha of 0.05, 352 patients would be required [26]. As we randomised at the level of the GP but measured patient outcomes, we controlled for clustering effects. Based on clustering effects reported in earlier trials, we applied an intra-class correlation of 0.03 [30]. Presuming a 10% dropout rate, we calculated that 426 patients should be included by 60 GPs.

Differences in baseline characteristics between dropouts and patients who completed the follow-up were tested for significance using a t-test for continuous variables and a X²-test for dichotomous and categorical variables.

The effect of the intervention on the separate measurement dates at two, six, twelve and twenty-six weeks was tested univariately. In multivariate analysis, differences in primary and secondary outcomes were estimated with a mixed model corrected for potential confounders: age, sex, educational level and the corresponding baseline value of the outcome. A random intercept was included for clustering at the level of the GP, and a random intercept and a random effect for time at the patient level were included to incorporate the effect over time. All analysis was performed on an intention-to-treat basis.

The potential confounding effect of each of the illness perception dimensions was assessed with mixed models, with restrictions at 12 weeks as the outcome variable and correction for all confounders.

Mixed models are robust for individual patients with missing follow-up measurements. In the analysis, we originally included the baseline measurement of the corresponding outcome as a covariate. Consequently, any measurement of any patient with a missing baseline measurement would be excluded from the analysis, thus reducing power and potentially introducing bias. We therefore decided to use multiple imputation to impute baseline variables and missing confounders measured at the baseline [31, 32].

Age, sex, educational level, absenteeism, all illness perceptions, the treatment allocation and all baseline measurements of primary and secondary outcomes were included in the multiple imputation. The numbers of imputed missing variables per baseline variable are described in Table 1. Five imputed datasets were created. Univariate and multivariate analyses were performed on each imputation; the results reported here were combined with Rubin’s rule [33].

Sixty-eight GPs agreed to participate and were randomised to the intervention (n = 34) or the control group (n = 34). GPs in the intervention group did not differ from control GPs with regard to sex, age, professional age, number of included patients or percentage of GP trainers per group.

Between January 2010 and January 2012, forty-seven GPs included 247 patients (range 0–10). Twenty-one of these patients did not meet the inclusion criteria because they did not align with the definition of non-chronic low back pain (n = 19), did not have back complaints (n = 1) or were younger than 18 years old (n = 1). Ultimately, 114 patients in the control group and 112 in the intervention group were included in the analysis (Fig. 1).

Patients in the intervention and control groups were comparable across most baseline measurements (Table 1 and Additional file 2 Appendix 2 for the imputed dataset). Patients in the intervention group reported more absenteeism from work due to their low back pain, and they more frequently expected their pain to last longer than did the controls.

During the follow-up, 76 (67%) of the control patients and 80 (70%) of the intervention patients completed all questionnaires. Overall, 71 (62%) patients in the control group and 75 (66%) patients in the intervention group completed the diary and all questionnaires (Fig. 1). Patients who did not complete the follow-up were more frequently of non-Dutch origin (15% non-Dutch natives in dropouts versus 5% non-Dutch natives in the analysed group; p = 0.017) and were younger (a mean age of 39.2 years for dropouts versus 47.4 years for the analysed group; p = 0.000). They did not differ significantly in other baseline measurements.

The mean disability score among the patients in the intervention and the control groups declined to 4.1 (SDM&PR group) and 4.3 (control group) after 2 weeks (difference 0.2; p-value 0.789), 2.1 (SDM&PR) and 2.3 (controls) after 12 weeks (difference 0.2; p-value 0.720) and 2.0 for both groups after 26 weeks (difference 0.0; p-value 0.949) (Table 2 and Fig. 2).

The mean pain score in the two groups was 18.9 (SDM&PR) and 20.3 (controls) after 2 weeks (difference 1.4; p-value 0.675), 14.2 (SDM&PR) and 12.4 (controls) after 12 weeks (difference 1.8; p-value 0.577) and 13.6 (SDM&PR) and 16.3 (controls) after 26 weeks (difference 2.7; p-value 0.385). The percentage of patients with adequate relief was 70% (SDM&PR) and 62% (controls) after 2 weeks (p-value 0.888), 49% in both groups after 6 weeks, 69% (SDM&PR) and 62% (controls) after 12 weeks, and 66% (SDM&PR) and 64% (controls) after 26 weeks. The mean number of days of absenteeism and mean health care consumption at 2, 6, 12 and 26 weeks did not differ between the two groups (Table 2).

In the multilevel, multivariate analysis, correcting for baseline differences, patient characteristics and the clustering effect, the mean difference in disability scores between the intervention and control groups during the six-month follow-up was − 0.259 (p-value 0.582) (Table 3). The mean difference in pain score between the two groups in the six-month follow-up was − 2.269 (p-value 0.306). During the follow-up, the two groups did not differ in the percentage of patients with adequate relief, the number of days of absenteeism or in healthcare consumption (Table 3).

Of the 8 dimensions of illness perception, only consequences (β = 1.24 confidence interval (CI) 0.14–2.35), timeline (β = 1.38 CI 0.27–2.48) and concern (β = 1.45 CI 0.34–2.56) were significantly associated with disability at 12 weeks. However, when the interaction term of each of these three items with the intervention was added to the multivariate model, no significant effect of illness perception on disability at 12 weeks was found.

When studying the fidelity of the intervention, we measured significant differences in the SDM behaviour in favour of the intervention group and a mean duration of the consultation of 16 min for the intervention group versus 13 min for the control group [23].

Most research on SDM thus far has focused on process outcomes and not on patient-related clinical outcomes [20].

We performed a randomised controlled trial among patients with low back pain recruited in daily primary care and evaluated the effectiveness of SDM on relevant clinical outcomes. Based on current knowledge, we constructed a multifaceted intervention and training programme that was grounded in a theoretical concept of SDM, involving both participants in the decision process [20].

Participating GPs were well trained, and the positive SDM performance during consultations after the training was acknowledged by the patients [23]. We used a mixed model analysis because these are robust for individual patients with missing follow-up measurements under the assumption of missing values completely at random or missing values at random dependent on another variable included in the mixed model. As in most studies, the correctness of this assumption of ‘random missing’ cannot be proven in our study. In our view, however, it is very unlikely that a treatment effect was not observed due to the multiple imputation procedure because the complete case analyses confirmed the lack of a treatment effect (Additional file 3 Appendix 3).

The patient recruitment met only half (53%) of the pre-set sample size. Participating GPs experienced problems that hampered recruitment, possibly due to unforeseen changes in the healthcare system, such as the introduction of direct access physiotherapy. However, the fact that the results of the patients in the intervention and control groups did not differ in any of the outcome measures at any moment in time demonstrates in our view the consistency of the results, and even if differences were demonstrated at the pre-set sample size, these differences would have been small and of questionable clinical relevance.

Dropout rates of GPs and patients are similar in the intervention and control groups and are in line with other studies on patients suffering from nonspecific low back pain in primary care in the Netherlands [35]. We cannot think of any reason why the intervention should have influenced the dropout rate of patients, but we estimate that dropout is not related to the intervention but is rather related to the complex disease course of nonspecific low back pain and the mismatch between patient expectations and the professional’s management.

Although we have observed significant differences in the perception of SDM between the experimental groups, we believe that the difference should be evaluated in the context of treatment fidelity. In a recent publication evaluating the intervention from an observational perspective, we found significant differences in the use of physical examination and in the consultation duration between groups [23]. Moreover, we question whether the patient perspective was sufficiently considered to incorporate the patient’s positive expectations into the actual decision despite significant differences in the SDM behaviour of GPs between the groups.

Unfortunately, as with many other studies on the effects of SDM on health outcomes, the effects are too small to allow conclusions about the impact of the separate communicative techniques, SDM and positive reinforcement of treatment expectations. Theoretically, we expected a positive interaction between the two techniques. The consistent pattern of very small potential positive treatment effects of the combined intervention above the usual standard of care could be explained by a stronger positive effect of one technique counteracted by the effect of the other (Fig. 2). For instance, in a study by a physiotherapist on the effects of SDM on the prognosis of low back pain, even negative expectations of patients were suggested to be responsible for poorer health outcomes after SDM than the usual standard of care [36]. Conversely, in a study evaluating the placebo effect on chronic low back pain, positive treatment expectations and a supportive environment were considered responsible for short-term relief from complaints [16, 23].

Studies on the effects of learned communicative strategies frequently face problems with blinding. We downsized the risk of non-blinding by sorting patients per GP, by recruiting patients via auxiliary staff members unaware of the allocation, and by not providing details of the trained communicative strategy to control GPs, auxiliary staff members and patients.

We did not perform a full health economic assessment but restricted the economic impact analysis to measuring the absenteeism of workers. However, in a detailed cost-effectiveness analysis, the reported 20% difference in the duration of the consultation time should be considered [23].

As in other studies on interventions for non-chronic low back pain, we did not find substantial or significant effects [37].

This finding could be attributed to different factors. Such as an excessively diverse study population in the duration of the complaints or in patient characteristics [36‐38].

Although we think that the risk of contamination was limited, GPs in the control group may have incorporated SDM in their consultations as well. This might be reflected in the fact that the mean score for the question of whether patients felt involved in the decision-making was between ‘mostly yes’ and ‘mostly no’. The difference in the results of the intervention group, where the mean score was ‘mostly yes’, was significant but limited. However, the observational study demonstrates low levels of SDM in both groups despite the significant effect of training in SDM behaviour [11, 23, 39, 40].

Because most patients quickly recover from their back pain, the intervention simply might not have had sufficient discriminative content above the spontaneous course. In their review of psychosocial interventions for non-chronic low back pain in primary care, Ramond et al. [38] advise the integration of several psychosocial factors with multicomponent interventions to overcome this problem.

Contextual factors play an important role in the symptom perception, prognosis and recovery of low back pain [1, 4]. We identified three subgroups of patients with a poorer prognosis for symptom recovery in the analysis of the effect of each of the illness perception dimensions on the restrictions at twelve weeks. Patients with more negative illness perceptions or with a longer duration of complaints before they contact their GP might be better helped by more positive treatment expectations, but unfortunately, our dataset did not allow subgroup analysis on the effect of the intervention for particular patient characteristics [38]. Differences between the contextual factors of patients in the intervention and control groups may have influenced the results. Although the patients were not randomised, we have no indication that the recruitment to the intervention and control groups resulted in selection bias [18, 28, 41, 42].

Although the GPs were extensively trained before participating in the intervention group and the patients recognised SDM during the actual consultations, we question, based on the results of the evaluation of the training, whether the training did result in adequate SDM performance [40, 43].

Trained GPs became more aware of the need to better inform patients about treatment options and to incorporate patients’ expectations during the intake phase of the consultation, but they persisted in providing a paternalistic, guideline-oriented choice.

In a review on the effects of the implementation of SDM in clinical encounters measured by an external observer, Couët et al. [40] reported similar training effects and noticed that only incidentally is clinical management adjusted to patient preferences. In the evaluation of the training, we confirm this observation and conclude that patient preferences were insufficiently considered in the actual decision-making to incorporate the patient’s positive treatment expectations into the treatment choice [23]. When patient preferences are not reflected in treatment choices, the impact of positive reinforcement of the therapeutic plan on patient recovery will diminish.

So far, task-oriented issues, such as performing process steps and information exchange, are emphasised in the implementation of SDM [20]. However, the effects of knowledge transfer on proportional understanding are questionable, and the effects on recovery are unclear [20, 43].

Future research on the involvement of patients in treatment decisions should therefore focus more on professional attitude and equality in the patient professional relationship as a condition for successful SDM.