Complex interventions, such as exercise for LBP, often have many treatment targets. Matching a primary outcome to the target(s) of exercise interventions may provide greater standardized mean differences (SMDs) than using an unmatched primary outcome. We aimed to explore whether the conclusions of exercise trials for LBP might differ with i) improved matching of outcomes to treatment targets and ii) the use of composite outcome measures. We investigated i) matching in five trials (n=1033) that used an unmatched primary outcome but included some of their matched outcomes as secondary outcomes; ii) composite outcomes in four trials (n=864). The composite consisted of standardised averaged matched outcomes. All analyses replicated the primary outcome analysis, applied to the matched or composite outcome in each dataset. When not possible, SMDs were calculated for the primary and matched outcomes. i) Of five trials, three had greater SMDs and increased statistical significance with matched outcomes (pooled effect SMD 0.35 (95% CI 0.16, 0.54), p=0.0003) compared to an unmatched primary outcome (pooled effect SMD 0.13 (95% CI 0.04, 0.23) p=0.007). ii) Of four composite outcomes: two matched trials had greater SMDs and improved statistical precision in the primary outcome than the composite outcome; two unmatched trials had greater SMDs and improved statistical precision in the composite compared to the primary outcome.Background
Methods and Results
