Background: Ordinal outcomes are extremely common in medical research - approximately 70% of Cochrane Stroke Group reviews of interventions include an ordinal outcome measure. However, in individual studies, these scales are often reduced to a binary outcome during analysis. This type of approach could be throwing away potentially useful information. Cohen [1983] and MacCallum [2002] both report that the loss of power when dichotomising at the median is equivalent to throwing away a third of the data. Better methods, such as proportional odds regression, are available for analysing ordinal outcomes, but they are not currently being used in Cochrane reviews.

Objectives: We aim to show how often sufficient data are available in randomized stroke trials for performing an ordinal outcome meta-analysis.

Methods: As it is a widely used outcome measure in stroke trials, we have focussed on the modified Rankin Scale, a measure of disability. We examined all 132 systematic reviews of interventions published by Cochrane Stroke Group in The Cochrane Library, 2010 Issue 11, to find included trials that measured the modified Rankin Scale. Using the Cochrane Stroke Group Specialized Trials Register, we obtained the full reports of these trials.

Results: We will present the answers to the following questions: i) How is the scale reported in each study? (i.e. full scale, partial scale, binary or other), ii) Are recent trials more likely to use/report the full scale?; and iii) Is the full scale more likely to be reported if the modified Rankin Scale is a primary outcome measure, rather than a secondary?