Background
Mechanical low back pain (LBP) is the musculoskeletal disorder with the highest prevalence in the adult population [
1], and it carries considerable disability and costs for society [
2]. The chronic progression of LBP is often considered a biopsychosocial problem characterised by a combination of physical, psychological and social dysfunctions [
3] that are typically patient-reported and have a subjective nature. There are many different therapeutic interventions for chronic LBP, but none of them is universally accepted as the magic bullet [
4]. A recent review [
5] reported that non-pharmacological therapies are more common for the treatment of chronic LBP than for acute LBP [
6]. One of the most frequently proposed treatments is multidisciplinary biopsychosocial rehabilitation (MBR). MBR is based on the biopsychosocial model, where health and illness are determined by a dynamic interaction between biological (genetic and biochemical), psychological (mood, personality, and behavior), and social factors (cultural, familial, socioeconomic, and medical assistance) [
7].
Neuroimaging studies have shown that brain regions activated by nociceptive stimuli can also be affected by emotional and behavioral states [
8]. Chronic LBP involves central sensitization, a neuropathic pain component, and may induce maladaptive coping strategies and depression [
9] in which the effect of the pain becomes more complex, being both a health and a social problem that requires comprehensive care through a multidisciplinary health care team [
5]. In this context, the objective of MBR is to improve physical function and modify beliefs and attitudes by addressing psychological issues or targeting social and work-related behaviour.
Meta-analysis (MA) of randomized controlled trials (RCTs) is considered the best approach to identify the actual benefit of a health intervention. A plethora of different instruments are employed to measure rehabilitation outcomes, often using continuous scales, according to the preference of the researchers who designed the study protocol [
10‐
12]. When studies assess the same outcome but measure it differently, the summary statistic usually adopted for meta-analyses is the standardized mean difference (SMD). This metric is obtained for each study by dividing the mean differences between the intervention and the control group by the pooled standard deviation of the outcome [
13]. This approach has two drawbacks: first, the effect of the same magnitude will appear different if the study populations are heterogeneous [
13]; second, the effect size expressed in standard deviation units is difficult for most health professionals to interpret [
14]. In addition, due to the subjective nature of the outcome variables, the cumulative estimate of the treatment effect needs to be presented as a clinically relevant measure in order to illustrate the benefit of the intervention to patients.
To overcome these limitations, the minimal important difference (MID) can be adopted as the summary statistic. The MID is defined as “the smallest difference in score in the outcome of interest that informed patients or informed proxies perceive as important, either beneficial or harmful, and which would lead the patient or clinician to consider a change in the management.” [
15] Reporting study results in MID units instead of standard deviation (SD) units for individual studies, and consequently for the pooled effect, can provide a uniform metric, bypassing the issues related to the use of SMD and facilitate the interpretation of results [
16].
Discussion
We noted a limited advantage of MBR versus usual care at all follow-up times when the results of the meta-analyses were expressed in MID units. Though there was a statistically significant difference in pain relief between the two treatments, all point estimates were smaller than 1 MID, which is, by definition, the clinical sizeable benefit. Meta-analyses almost always give nominally statistically significant results at
p < 0.05 for the difference between two treatments. This is not relevant for health care professionals, however, if the effect size is not large enough to have a practical impact on patients. Reporting MID units assumes a patient-centred perspective: treatment with a MID above 1 is expected to have important benefits for the majority of patients but for very few if below 0.5 [
46]. Accordingly, comparison of MBR versus usual care for short-and medium-term relief of back pain shows that there is a real but clinically modest difference (slightly lower than 1 MID) in favour of MBR. In the long-term comparison, the benefit, albeit statistically significant, is not clinically relevant, on average, as the pooled estimated is less than one-third of 1 MID.
A systematic review published in 2007 [
47] reported no effectiveness of MBR versus usual care for pain (only one of the 7 studies reported a positive effect). In contrast, the 2014 Cochrane review that we selected as case study showed that MBR, when analyzed in SMD, significantly reduced back pain in the long-term. However, the small effect and the moderate quality of evidence rated with the GRADE system led the authors to conclude that the superiority of MBR
may be clinically relevant, a conclusion also remarked on in a recent update [
34]. In our study we better quantified the clinical relevance by using the MID as a benchmark for drawing conclusions. This could be particularly relevant in the context of clinical recommendations.
According to GRADE guidelines [
48], when evaluating
imprecision, the authors of systematic reviews should consider whether the CIs of the effect size include appreciable benefit or harm [
49]. Reporting meta-analyses in MID units can help readers and stakeholders judge at a glance the precision of the overall effect in terms of clinical relevance and the amount of benefit or harm against a clear anchor point. Furthermore, optimal information size can be directly related to both clinical and statistical significance. Drawing conclusions from meta-analyses based only on statistical significance may be misleading, however, especially if associated with a high prevalence of small studies and poor reporting, as is typical of the rehabilitation literature [
10,
50]. There is a need to move beyond the
p-value cliché and to focus on the magnitude of benefit since interventions of limited value sap valuable time and resources from other interventions that might have more substantial effects.
The goal of MBR is to teach individuals to cope with their pain. In doing so, the aim is to modify deeply-rooted attitudes and beliefs, as they may contribute to prolonging back pain by activating physical and emotional “triggers” [
51]. It is expected that interventions will produce clinical relevant benefits in both the short- and long-term. We found no evidence of clinically relevant long-term results. One possible explication is that, after the acute phase, when the specialist sees the patient, measures baseline pain, and starts the therapy, pain may decrease over time regardless of treatment. This makes potential long-term effects smaller and more difficult to detect [
52].
We showed that, in a clinically meaningful summary estimate such as the MID, the results of a meta-analysis can be interpreted differently by clinicians and patients. The main finding of our study has possible implications for recommending MBR. Multidisciplinary biopsychological rehabilitation is endorsed in the ACP/APS [
53] in the National Disease Management [
54] and the 2016 NICE draft guidelines [
55]. Based on our results, and given the potentially high cost of MBR, these indications need to be re-considered.
The MID unit approach has some limitations in particular instances. First, the use of MID units requires that previous studies have reported an estimate of the MID (possibly an anchor-based MID) from several trials. Currently, few instruments that assess an outcome have an established MID: one study including a large cohort of trials (
n = 185) on LBP rehabilitation found 70 different pain measurement instruments [
10]. In contrast, we found only 5 instruments that had an anchor-based MID. Second, the MID is informative only about the comparison of a treatment versus a control (i.e., usual care or placebo). If we compare two different treatments, the MID value needs to be modified to account for the effect of the control treatment. For example, in the comparison of MBR versus pharmacological treatment, we should not apply the same MID that we used against usual care because the latter will already have a sizeable effect on pain relief and a smaller additional increase could be interpreted as clinically relevant. In addition, in these meta-analyses, and regardless of the unit of analysis, usual care is not the same for all studies. Also, we used a MID that does not distinguish between chronic and acute pain [
43]. Finally, meta-analyses reported in MID units are vulnerable to unexperienced, oversimplified interpretation unless we keep in mind that when we define a MID, we choose a single value while, in reality, the MID is subjective, i.e., the clinical relevance of a change in outcome may be perceived differently from patient to patient.