The risk of bias and sample size of trials of spinal manipulative therapy for low back and neck pain: analysis and recommendations.

OBJECTIVE

The purpose of this study was to evaluate changes in methodological quality and sample size in randomized controlled trials (RCTs) of spinal manipulative therapy (SMT) for neck and low back pain over a specified period. A secondary purpose was to make recommendations for improvement for future SMT trials based upon our findings.

METHODS

Randomized controlled trials that examined the effect of SMT in adults with neck and/or low back pain and reported at least 1 patient-reported outcome measure were included. Studies were identified from recent Cochrane reviews of SMT, and an update of the literature was conducted (March 2013). Risk of bias was assessed using the 12-item criteria recommended by the Cochrane Back Review Group. In addition, sample size was examined. The relationship between the overall risk of bias and sample size over time was evaluated using regression analyses, and RCTs were grouped into periods (epochs) of approximately 5 years.

RESULTS

In total, 105 RCTs were included, of which 41 (39%) were considered to have a low risk of bias. There is significant improvement in the mean risk of bias over time (P < .05), which is the most profound for items related to selection bias and, to a lesser extent, attrition and selective outcome reporting bias. Furthermore, although there is no significant increase in sample size over time (overall P = .8), the proportion of studies that performed an a priori sample size calculation is increasing statistically (odds ratio, 2.1; confidence interval, 1.5-3.0). Sensitivity analyses suggest no appreciable difference between studies for neck or low back pain for risk of bias or sample size.

CONCLUSION

Methodological quality of RCTs of SMT for neck and low back pain is improving, whereas overall sample size has shown only small and nonsignificant increases. There is an increasing trend among studies to conduct sample size calculations, which relate to statistical power. Based upon these findings, 7 areas of improvement for future SMT trials are suggested.