Considerations in Poisson and Negative Binomial Model Selection for Identification of Risk Factors for Caries Development in Potentially Heterogeneous High-Risk Populations

• Activity Number: 554
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Keyla Pagan-Rivera* and Deborah Dawson and Karin Weber-Gasparoni and John J. Warren and Katherine W.O. Kramer and Teresa A. Marshall and David C. Johnsen
• Companies: and The University of Iowa and The University of Iowa and The University of Iowa and Health Integrity LLC and The University of Iowa and The University of Iowa
• Keywords: Model Selection ; Negative Binomial ; Posion ; Information Criteria

Abstract

Data were pooled from four studies conducted in four eastern Iowa pediatric populations at high risk for dental caries (N=882). All studies scored caries at the surface level using the d1d2mfs caries score criteria. Multiple risk factors were evaluated using similar or identical protocols, including demographics, and children's dietary and oral hygiene habits. A set of candidate variables were chosen based upon bivariate analyses (Spearman and Kruskal-Wallis tests). Poisson and Negative Binomial models were fitted with those candidate variables. The inclusion of Poisson and Negative Binomial zero-inflated models was also considered. Goodness-of-fit tests were computed for a subset of candidate models and information criteria were used to choose the best model. Challenges to model selection were posed by the possible heterogeneity of the different populations represented, and strategies to address this consideration were addressed. The Negative Binomial models were always preferred over the Poison models and Zero-inflated models were necessary with these data.