Model Selection Using Database Characteristics: Developing a Classification Tree for Longitudinal Incidence Data