Skip to main content


Would a timely analysis of secondary outcomes have led to an earlier identification of effective stroke rehabilitation interventions?

Date and Location




Sunday 22 September 2013 - 10:30 - 12:00


Presenting author and contact person

Presenting author

Peter Langhorne

Contact person

Peter Langhorne
Abstract text
Background: Randomised trials often use secondary (or surrogate) outcomes that are believed to anticipate changes in primary outcomes. Many surrogate outcomes are used in rehabilitation research in the belief that recovery patterns are predictable and improved secondary outcomes will herald improved primary outcomes. Objectives: To identify if a timely analysis of secondary outcomes would have resulted in an earlier identification of effective rehabilitation interventions. Methods: We identified 11 “effective” rehabilitation interventions (median 10 trials; IQR 6-17) that were supported by a Cochrane review and recommended in Clinical Guidelines in 2010. These were matched with 11 similar “neutral” interventions (13 trials; 5-18) that are currently not recommended. We then carried out a series of cumulative meta-analysis against; a) time before guideline publication, and b) total number of participants included in each analysis. The main outcome was the time (participants) required to achieve a significant z-score (>1.96). Results: When comparing the “effective” and “neutral” intervention reviews, the total number of participants were similar (P=0.47) but the “neutral” interventions had fewer available for secondary outcome analyses (median 232 versus 586; P=0.04). The “effective” interventions achieved a significantly higher median z-score on the primary outcome after the accumulation of 300 participants compared with only 200 participants for the secondary outcome. Within the group of “effective” interventions both primary and secondary outcomes achieved significance at the same time (8 years versus 9 years; P=0.79) prior to guideline publication, but the number of participants required was lower for the secondary outcome (160 versus 461; P = 0.056). Conclusions: In these rehabilitation reviews secondary outcome results did achieved statistical significance with fewer participants than the primary outcomes but this did not translate into an earlier demonstration of effect because data were often missing. Secondary outcomes need to be applied in a more standardised and consistent manner.