Quantile-Based Sample Size Planning for Surveillance-Based Prevalence

• Activity Number: 517
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Clinton Alverson*+ and Adolfo Correa and Mike Atkinson
• Companies: CDC/ONDIEH/NCBDDD and University of Mississippi Medical Center and CDC/ONDIEH/NCBDDD
• Keywords: Poisson ; Surveillance ; Prevalence ; Planning

Abstract

In birth defects surveillance, a basic measure of disease burden is the population prevalence. Since birth defects are rare with respect to population size, and the usual populations (and samples) of interest are large, the Poisson count distribution is commonly used when performing inference and modelling in this setting. The Poisson process is driven by its mean, in this setting Mean = NP, N being population or sample size, P being the true prevalence. When planning surveillance with prevalence estimation in mind, there are two basic considerations: given a fixed sample size, what is the minimum prevalence that may be reliably estimated; and given a specified prevalence value, what is the minimum sample size required to permit reliable estimation of that prevalence. This amounts to determining an acceptable lower bound for the corresponding Poisson mean. We employ exact calculations and/or simulations to examine the distribution quantiles of prevalence estimators and related quantities. These quantiles can guide the identification of minimally reasonable sample sizes or prevalence values, and supplement the mean-based planning methods.