Statistical power is one key element of robust results and scientific rigor, so sample size and power calculation is a crucial step in designing a study and especially important for dental outcomes with excess zeros when many subjects have not experienced the oral disease. Different approaches have been developed to model outcomes with excess zeros (Duan et al 1983; Lambert 1992; Long 1997; Min and Agresti 2002; Hilbe 2011; Kassahun et al. 2014; Neelon et al. 2016). However, the existing models may have limited power (Williamson et al. 2007; Follman et al. 2009). This project will examine intensively the statistical power of existing analysis methods specifically on longitudinal dental outcomes with a large amount of zeros under different realistic settings, and develop software for sample size and power calculation for overall treatment effects and mediation effects specifically on longitudinal zero-inflated (ZI) dental outcomes for more accurate estimate on the sample size needed. The software we will develop will provide investigators a tool to design a study with good power to examine not only the overall treatment effects but also potential mechanisms or pathways the treatment works. A larger sample size is usually required with ZI models than standard models (Williamson et al. 2007). However, recruiting large samples will significantly increase the cost. Therefore, given the potentially limited power of existing methods, this project will develop a more powerful tool allow investigators to detect a meaningful treatment effect on ZI dental outcomes with a realistic sample size. With successful completion of this project, we expect to fill the gap in literature to have better understanding of how excess zeros observed in common dental outcomes affect the statistical power of existing methods, provide dental researchers software for power calculation on overall and mediation treatment effects, and develop more powerful tools for treatment effect evaluations on longitudinal dental outcomes with excess zeros.
With successful completion of this project, we will provide dental researchers software for power calculation on overall and mediation treatment effects to fully understand the treatment working mechanism, and more powerful tools for treatment effect evaluations on longitudinal dental outcomes with excess zeros, allowing investigators to detect a meaningful treatment effect on ZI dental outcomes with a realistic sample size. We expect these tools to be readily interpretable and more powerful than existing methods, allowing better understanding of the clinical and community benefit of dental health interventions.