WITHDRAWN: Sample Size Determination for Quadratic Inference Functions Method in Longitudinal Design with Dichotomous Outcomes

• Author(s): Youna Hu and Peter Song and Youna Hu
• Keywords: Clinical trial ; Correlation structure ; Efficiency ; GEE ; QIF ; Power

Abstract

This paper calculates sample size with quadratic inference functions (QIF) method in the longitudinal marginal model with binary outcomes. An improved method to the generalized estimating equations (GEE) method, QIF has demonstrated higher estimation efficiency. To capture this gain into the design of longitudinal studies, we developed sample size and power formulas for the QIF method. We show that with the same power the QIF-based design requires smaller sample sizes than the GEE-based design when correlation are misspecified. Depending on its magnitude, the sample size saving can be substantially - ranging from 3% to 54% among the cases considered in our examples. In addition, unlike the GEE-based design that requires larger sample sizes for more follow-up visits,the QIF-based design is insensitive to varying numbers of follow-up visits. Thus, we recommend QIF for the design stage.