A Practical Way to Standardize Test Scores in Some Clinical Measures

• Activity Number: 171
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Liansheng Zhu*+
• Companies: Hoffmann-La Roche
• Keywords: Normalize ; standardization ; discrete uniform distribution ; multinomial distribution

Abstract

In clinical studies, many important endpoints are captured through a set of tests, each test with a different score rante. Test scores need be normalized in order to give similar weights for the set of tests before used for analysis. In current practice, scores are normalized by using the mean and standard deviation of baseline scores. There are three major concerns around this approach. Namely, 1) Correlation between any two subjects is induced; 2) Transformed scores are not invariant from study to study, which results that measures across studies can not be compared; 3) Transformed scores and changes from baseline are not interpretable. We propose to standardization test scores by using the mean and variance from the underlying distribution function such as discrete uniform distribution, which addresses three concerns mentioned above without scarifying any statistical properties.