|Thursday, February 20|
|PS1 Poster Session I & Opening Mixer||
Thu, Feb 20, 5:15 PM - 6:45 PM
Accounting for Regression to the Mean and Natural Growth in Uncontrolled Studies (302802)Robbie Beyl, Pennington Biomedical Research Center
Jeffrey H. Burton, Pennington Biomedical Research Center
*William D. Johnson, Pennington Biomedical Research Center
Keywords: Regression to the mean, Uncontrolled studies, Changing covariate, Truncated bivariate normal distribution, Simulation
Weight-loss studies are conducted often in large samples in which the study design does not include a control group that would enable accurate assessment of treatment effect. Participants in these studies typically are enrolled because their weight is above a certain threshold that defines a need for weight reduction. It is well known that regression to the mean may obscure assessment of treatment effect in these circumstances. The problem is further exacerbated when participants are children or adolescents who experience growth during the study, because growth in height is a factor in the determination of excess weight. This presentation reviews the formulation for estimating intervention effect when the pre-post intervention assessments follow an underlying bivariate normal distribution but the first assessment cannot exceed the threshold and therefore the second assessment is subject to regression to the mean. The novel aspect of the problem lies in the fact that the assessments are correlated with a covariate that changes during the period of observation. Monte Carlo simulations are used to empirically demonstrate the methods.