Rehabilitation interventions in randomized controlled trials for low back pain: proof of statistical significance often is not relevant.

BACKGROUND

An observed statistically significant difference between two interventions does not necessarily imply that this difference is clinically important for patients and clinicians. We aimed to assess if treatment effects of randomized controlled trials (RCTs) for low back pain (LBP) are statistically significant and clinically relevant, and if RCTs were powered to achieve clinically relevant differences on continuous outcomes.

METHODS

We searched for all RCTs included in Cochrane Systematic Reviews focusing on the efficacy of rehabilitation interventions for LBP and published until April 2017. RCTs having sample size calculation and a planned minimal important difference were considered. In the primary analysis, we calculated the proportion of RCTs classified as "statistically significant and clinically relevant", "statistically significant but not clinically relevant", "not statistically significant but clinically relevant", and "not statistically significant and not clinically relevant". Then, we investigated how many times the mismatch between statistical significance and clinical relevance was due to inadequate power.

RESULTS

From 20 eligible SRs including 101 RCTs, we identified 42 RCTs encompassing 81 intervention comparisons. Overall, 60% (25 RCTs) were statistically significant while only 36% (15 RCTs) were both statistically and clinically significant. Most trials (38%) did not discuss the clinical relevance of treatment effects when results did not reached statistical significance. Among trials with non-statistically significant findings, 60% did not reach the planned sample size, therefore being at risk to not detect an effect that is actually there (type II error).

CONCLUSION

Only a minority of positive RCT findings was both statistically significant and clinically relevant. Scarce diligence or frank omissions of important tactic elements of RCTs, such as clinical relevance, and power, decrease the reliability of study findings to current practice.