Abstract WP393: Exploration Of Time-course Combinations Of Outcome Scales For Use In A Global Test Of Stroke Recovery
Background: Clinical trials for acute ischaemic stroke treatment require large numbers of patients and are expensive to conduct; any method to enhance statistical power or sensitivity is desirable. We explored whether treatment effects may be detected more strongly if outcome is assessed by using combined early and late measures (e.g. 7 day NIH Stroke Scale (NIHSS) combined with 90-day modified Rankin scale (mRS)) than either measure alone.
Methods: Data on 4077 patients were available from the Virtual International Stroke Trials Archive (VISTA). We analysed sensitivity to treatment effect, contrasting alteplase as standard care versus non-thrombolysed care. We used proportional odds logistic regression for ordinal scales and Generalised Estimating Equations for global outcomes, with all analyses adjusted for baseline severity and age. We ran simulations to assess relation between sample size and power for each ordinal scale and corresponding global outcomes. We used R 2.12.1 for simulations and SAS 9.2 for all other analyses.
Results: All scales were sensitive to treatment effect in isolation. The Table shows ordinal scales and global outcomes of combined early and late outcomes ranked in order of sensitivity to treatment effect, displayed as odds ratio (OR).
Discussion: When early and late outcomes were combined in a global test and compared to solitary ordinal scales, any enhancement of the OR appeared marginal. However, this conferred a 20% reduction in required sample size at 80% power. The greatest increment in power arose when component measures were less correlated.
- © 2012 by American Heart Association, Inc.