Carolina Ballroom
Testing Parallelism in Correlated Setup (303497)
*Meenakshi Mahanta, Cytel Statistical Software & Services Private Limited"Testing hypothesis of parallelism under assumption of independence is very common. I ran into a case of correlated slopes in a multiple exposure cosmetic trial conducted to determine anti-irritation potential of skincare products. Each subject received all product applications twice daily at 12 sites on back. During treatment session, firstly irritant and then skincare product were applied to reduce irritation.
Mixed model with repeated measures was proposed for comparison at each timepoint but experimenter was interested in overall comparison. One way was to use time-weighted averages. It was revealed that slope of regression (irritation vs. time) is a good measure of efficacy of product. Hence, slopes of all products were to be compared with slope of ‘irritant only’.
Traditional approach assumes independence of data. Mixed model with Dunnett’s adjustment is one of the well known methods used for multiple comparisons. Incorporating co-relatedness is possible by applying Hotelling T2 test where correlations were estimated from data. Comparison of these methods in terms of observed type I error and empirical power is made and the Hotelling T2 method is found to be more appealing."