James A Thompson1, Susan E Carozza, Li Zhu. 1. Department of Large Animal Clinical Science, Texas A&M University, College Station, TX 77843-4475, USA. JThompson@cvm.tamu.edu
Abstract
BACKGROUND: Spatial modeling of rare diseases, such as childhood cancer, has been hampered by imprecise risk estimates. Recent developments in Bayesian hierarchical modeling include the ability to adjust a disease risk estimate to be fully conditional for covariance among neighboring locations and for covariance among multiple diseases within each location. This joint modeling approach is called Multivariate Intrinsic Conditional Autoregressive. The objective of this study was to evaluate the spatial and histotype covariance among childhood cancer histotypes, in Texas. Results will be valuable for selecting appropriate models to support more specific etiologic studies of environmental factors for childhood cancer. METHODS: County level standard morbidity ratios for 13 childhood cancer histotype groups were estimated using Multivariate Intrinsic Conditional Autoregressive modeling and the results compared to results from two reduced models. The two reduced models were the base model specified with zero spatial covariance and the base model specified with zero histotype covariance. The results were compared using the Deviance Information Criterion and Geographical Information System techniques were used to compare patterns of standard morbidity ratios. RESULTS: Including histotype covariance greatly improved the Deviance Information Criterion and including spatial covariance produced a moderate improvement. Parameter evaluation by GIS techniques showed that excluding histotype covariance resulted in marked shrinkage of the risk estimates. CONCLUSIONS: Investigation of childhood cancer could benefit by incorporating histotype covariance into environmental modeling.
BACKGROUND: Spatial modeling of rare diseases, such as childhood cancer, has been hampered by imprecise risk estimates. Recent developments in Bayesian hierarchical modeling include the ability to adjust a disease risk estimate to be fully conditional for covariance among neighboring locations and for covariance among multiple diseases within each location. This joint modeling approach is called Multivariate Intrinsic Conditional Autoregressive. The objective of this study was to evaluate the spatial and histotype covariance among childhood cancer histotypes, in Texas. Results will be valuable for selecting appropriate models to support more specific etiologic studies of environmental factors for childhood cancer. METHODS: County level standard morbidity ratios for 13 childhood cancer histotype groups were estimated using Multivariate Intrinsic Conditional Autoregressive modeling and the results compared to results from two reduced models. The two reduced models were the base model specified with zero spatial covariance and the base model specified with zero histotype covariance. The results were compared using the Deviance Information Criterion and Geographical Information System techniques were used to compare patterns of standard morbidity ratios. RESULTS: Including histotype covariance greatly improved the Deviance Information Criterion and including spatial covariance produced a moderate improvement. Parameter evaluation by GIS techniques showed that excluding histotype covariance resulted in marked shrinkage of the risk estimates. CONCLUSIONS: Investigation of childhood cancer could benefit by incorporating histotype covariance into environmental modeling.
Authors: Angela D Liese; Andrew Lawson; Hae-Ryoung Song; James D Hibbert; Dwayne E Porter; Michele Nichols; Archana P Lamichhane; Dana Dabelea; Elizabeth J Mayer-Davis; Debra Standiford; Lenna Liu; Richard F Hamman; Ralph B D'Agostino Journal: Health Place Date: 2010-01-15 Impact factor: 4.078
Authors: Garyfallos Konstantinoudis; Dominic Schuhmacher; Roland A Ammann; Tamara Diesch; Claudia E Kuehni; Ben D Spycher Journal: Int J Health Geogr Date: 2020-04-17 Impact factor: 3.918