Interconversion of stroke scales. Implications for therapeutic trials.

BACKGROUND AND PURPOSE Stroke scales are intended to measure stroke severity for the purpose of clinical trials. Scores have been used to determine trial entry, to compare patient groups within or between trials, or as a secondary end point. The use of scores as an end point in meta-analysis has not been validated, but such analyses have nevertheless been performed when equivocal results have been obtained using the main outcome measure. The different scale designs suggest that conversion of scores may not be possible. We sought to determine whether scores on different scales could be interconverted. METHODS A single observer scored 433 consecutive admissions to an acute stroke unit on the Canadian Neurological Scale, the middle cerebral artery Neurological Score (or Orgogozo scale), and the National Institutes of Health stroke scale. Data were separated into training and test sets, and linear regression was used to model conversion between scales. Prediction errors were calculated. Strokes were subdivided according to the Oxfordshire Community Stroke Project classification, and coefficients of determination were calculated for different subtypes. RESULTS Conversion between Canadian and middle cerebral artery Neurological scales was satisfactory (R2 = 94.7%), and prediction errors were acceptable (absolute prediction error, 5.0 +/- 5). Conversion from the National Institutes of Health scale was worse (R2 = 87.5% to Canadian and 89.0% to Neurological Score), and prediction errors were significantly greater (Neurological Score error, 8.7 +/- 7; Canadian Neurological Scale error, 8.5 +/- 7.3; P < .005 for both). Coefficients of determination for interconversion were significantly worse for dysphasic patients with total anterior circulation strokes than for other stroke types (P < .01). Reweighting the motor component of the National Institutes of Health scale improved coefficients of determination and reduced prediction errors, but prediction error for conversion to the Canadian scale remained significantly greater than other conversions (P = .001). CONCLUSIONS The Canadian Neurological Scale and the middle cerebral artery Neurological Score may reliably be converted. The National Institutes of Health scale cannot be used to predict these scores reliably, even with reweighting of the motor score. Interconversion is poorest for patients with dysphasia and total anterior circulation strokes. These results suggest that there will be more general difficulty in interconverting scales that use different test items and weighting. Meta-analysis using sequential changes in averaged scores from various stroke scales is not valid.