Interpreting the outcomes of stroke rehabilitation trials

Article type
Authors
Langhorne P
Abstract
Introduction/Objective: Conducting systematic reviews of stroke rehabilitation services are complicated by the range of different outcome measures used in the primary trials. Statistical approaches such as the standardised effect size can permit some pooling of outcome data but raise both methodological difficulties and problems of interpretation. We have therefore surveyed the available randomised trials of stroke rehabilitation services to establish; 1) the range of outcome measures used, 2) their comparability of content, 3) whether apparently comparable measures of disability and handicap had comparable utility in identifying important patient outcome groups.

Methods: The survey of stroke rehabilitation trials identified the most commonly used measures of disability and handicap. These were then applied in two cohort of stroke patients within 6 months of discharge from hospital. These patients were categorised into three (reference) disability groups; 1) full independence - capable of living independently at a private address, 2) limited independence - capable of personal activities of daily living but requiring help with domestic and social activities, 3) dependent - requiring nursing support in daily activities. Measures of handicap were compared against a self-rated quality of health ("good/excellent" vs "poor/very poor"). The ability of different outcome measures to discriminate patient outcomes were analysed using receiver operator curves and reported as their peak accuracy.

Results: Over 40 outcome measures were identified although the 10 commonest measures (3 disability, 5 handicap, 2 "global" outcome measures) were used in 41/54 (76%) of trials. The disability and "global" measures were good discriminators of the disability groups (accuracy 0.81-0.96 and 0.79-0.82 respectively). Handicap and "global" measures were poorer predictors of self-rated quality of health (accuracy 0.67-0.76; 0.57-0.58 respectively).

Discussion: When faced with different measures of similar outcome domains, it may be possible to establish reliable points at which to dichotomise different outcome measures to provide comparable and meaningful results. This approach appears more promising for measures of disability than for handicap.