Standards for the use of ordinal scales in clinical trials.

Abstract

Ordinal scales are frequently used in clinical trials to quantify outcomes which are non-dimensional. They may be regarded as either single state or transition measures based on whether they assess the outcome at a single point in time or directly examine change which has occurred between two points in time. Each has unique structural and operating characteristics, so that different methodological standards for their construction and utilisation are required. All trials employing ordinal scales published in three leading journals between 1980 and 1984 were examined. For both types of scales the individual ranks must be clearly defined, mutually exclusive, and ordered in a hierarchical progression. Further, both types must be able to detect equally both improvement and deterioration. For this to be ensured with the single state scales the population under study must not be clustered at one extreme of the scale at entry into the trial. With the transition scales internal symmetry must be achieved. Strategies for determining the comparisons to be performed should include emphasis on within patient analysis for crossover trials. Concordance between scale scores and the other measures of outcome employed in the trial must be evaluated. Frequent violations were uncovered in the studies reviewed, and it is hoped that the simple rules outlined will prove useful in the planning and evaluation of future trials.