Visual acuity as an outcome measure in clinical trials of retinal diseases.
PURPOSE: Visual acuity (VA) is the primary outcome measure in many studies involving eye diseases. A standard statistical approach for comparing a continuous measurement such as a VA letter score between 2 treatment groups is to perform a t test comparing the means. However, frequently a binary variable is created from the continuous VA letter score based on whether or not there has been a worsening (or gain) of > or =15 letters (equivalent to > or =3 lines), and a chi square or similar statistical test is performed to compare the proportions of success (or failure) between groups. The purpose of this article is to contrast these 2 approaches.
METHODS: Clinical trial reports of retinal disorders were used to compare results using mean change in the VA letter score versus binary proportions created from the VA letter score. Additionally, analyses were performed using generated data to gain a perspective on the magnitude of differences that might be expected between the 2 methods.
RESULTS: Studies from the literature showed that differences of 6% to 15% in > or =15-letter worsening corresponded to mean differences in letter scores between groups of 3.0 to 7.0 (approximately 0.6 to 1.4 lines). Analyses using generated data demonstrated that a mean improvement in the VA letter score of 5 corresponded to a doubling of the proportion of eyes with > or =15-letter improvement and a 28% relative reduction in the proportion of eyes with > or =15-letter worsening.
CONCLUSIONS: How VA data should be analyzed in a clinical trial depends to large extent on the research question. The frequently used outcome of > or =15-letter change has several drawbacks, including loss of efficiency (need for a larger sample), misclassification of the outcome, and potential for a ceiling or floor effect. Therefore, for most clinical trials we believe that the primary outcome analysis should be a comparison of changes in the VA letter score, and created binary variables should be reported as secondary outcomes. This approach maximizes the information gained from the data and accommodates both improvement and worsening of acuity.