Comparison of Adaptive Gaussian Quadrature, PQL, and Laplace 6 Estimates of Hierarchical Logistic Models When Comparing Small Incident Rates in Cluster-Randomized Trials

• Activity Number: 548
• Type: Contributed
• Date/Time: Thursday, August 2, 2007 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Rafael Diaz*+
• Companies: California State University, Sacramento
• Keywords: hierarchical logistic models ; cluster-randomization with dichotomous outcomes ; PQL ; Laplace 6 approximation ; Adaptive Gaussian Quadrature

Abstract

Hierarchical logistic models allows for the comparisons of groups of proportions from cluster-randomized trials with binary outcomes. Parameter estimation techniques for these models can yield considerably different estimates when the cluster proportions are small. This study compares the commonly used Penalized quasi-likelihood technique, the relatively new Laplace 6 approximation, and the Adaptive Gaussian Quadrature technique used in mainstream statistical packages using Monte Carlo simulations. In these simulations the difference between two groups of small proportions, modeled after those commonly found in epidemiological interventions involving small incident rates, is compared by means of a hierarchical logistic model. Recommendations for the use of each of these techniques when comparing groups of small proportions are provided based on the results of this study.