Statistical Challenges in the Analysis of Longitudinal Symptom Count Data
Nicole V. Brown, Center for Biostatistics, The Ohio State University  Xiaobai Li, Medimmune  *Jianliang Zhang, Medimmune 

Keywords: symptom count data, longitudinal, modeling, overdispersion

In clinical studies, the symptom counts, which is based on the number of symptoms observed on a symptom checklist, are usually collected at multiple time points for each subject. The count variable is bounded at zero and the total number of questions the checklist contains. Typically these questions are not independent of each other and having a symptom may not be a rare event at all. All these issues cast doubt on the use of a Poisson based approach, such as generalized linear mixed effects models with a Poisson distribution and log link. Treating the counts as ordinal and using a proportional odds ratio type of model makes the results hard to interpret and the assumptions hard to evaluate. Semiparametric mixed effects models could be a promising alternative. In this talk we will review different analysis methods and illustrate their applications on real clinical data and simulated data. We will compare the utilities of these models and further discuss how to handle overdispersion.