• Sung C Choi, Guy L Clifton, Anthony Marmarou, and Emmy R Miller.
• Department of Biostatistics, Medical College of Virginia, Virginia Commonwealth University, Richmond 23298-0032, USA. choi@hsc.vcu.edu
• J. Neurotrauma. 2002 Jan 1; 19 (1): 17-22.

AbstractThe power of clinical trials depends mainly on the choice of the primary outcome measure, the statistical test, and the sample size. The most widely used outcome measure has been the five-category Glasgow Outcome Scale (GOS). Contrary to intuition, we show that more categories do not necessarily increase the power of a trial and actually can decrease power. This is so for two reasons. The more categories of outcome measure used, the more the likelihood for misclassifications. The effect of 0%, 10%, and 20% misclassification rate upon power is illustrated. Misclassification rates in two completed trials are examined based on comparative overlap in GOS and Disability Rating Scale (DRS) categories. The outcome results of the "National Acute Brain Injury Study: Hypothermia" indicate that the ideal number of categories also depends upon the effect of study treatment. In the recently completed hypothermia trial, the use of a dichotomized GOS (good recovery/moderate disability versus severe disability/vegetative/dead) is shown to be more sensitive than use of three or more categories of the GOS. The results point to the importance of training study investigators who will collect the outcome data. The results also indicate that the number of categories should be carefully determined using the pilot data or the data from phase II trials.

16 keywords...

hide…