Article type
Year
Abstract
Introduction/Objective: To find the best way to handle studies with a negative outcome in a systematic review of the literature.
Methods: A systematic review of studies on low back pain and vocational outcome was carried out. After retrieval the internal validity of studies was assessed. For studies with a negative outcome a statistical power analysis was performed. A level of evidence for the efficacy of different interventions was assessed with and without a statistical power analysis. Studies with low internal validity or insufficient statistical power were excluded.
Results: Exclusion of studies with insufficient power led to a different picture of evidence of inefficacy. No evidence for the efficacy was distinguished from evidence of inefficacy.
Discussion: The statistical power criterion was based on a rather large 25% difference between experimental and reference treatment as a clinical relevant difference. This is a mild criterion for exclusion as a smaller difference would be more difficult to detect. Systematic reviews will be biased towards evidence of inefficacy of interventions if the statistical power of studies is not assessed.
Methods: A systematic review of studies on low back pain and vocational outcome was carried out. After retrieval the internal validity of studies was assessed. For studies with a negative outcome a statistical power analysis was performed. A level of evidence for the efficacy of different interventions was assessed with and without a statistical power analysis. Studies with low internal validity or insufficient statistical power were excluded.
Results: Exclusion of studies with insufficient power led to a different picture of evidence of inefficacy. No evidence for the efficacy was distinguished from evidence of inefficacy.
Discussion: The statistical power criterion was based on a rather large 25% difference between experimental and reference treatment as a clinical relevant difference. This is a mild criterion for exclusion as a smaller difference would be more difficult to detect. Systematic reviews will be biased towards evidence of inefficacy of interventions if the statistical power of studies is not assessed.