Disease mapping models for data with weak spatial dependence or spatial discontinuities

Recent advances in the spatial epidemiology literature have extended traditional approaches by including determinant disease factors that allow for non-local smoothing and/or non-spatial smoothing. In this article, two of those approaches are compared and are further extended to areas of high interest from the public health perspective. These are a conditionally specified Gaussian random field model, using a similarity-based non-spatial weight matrix to facilitate non-spatial smoothing in Bayesian disease mapping; and a spatially adaptive conditional autoregressive prior model. The methods are specially design to handle cases when there is no evidence of positive spatial correlation or the appropriate mix between local and global smoothing is not constant across the region being study. Both approaches proposed in this article are producing results consistent with the published knowledge, and are increasing the accuracy to clearly determine areas of high- or low-risk.

Quantification of inequities in breast cancer incidence in New Mexico through Bayesian disease mapping.

e13617 Background: Breast cancer (BC) is the most prevalent cancer of women in the United States (US). The incidence rates for BC among American Indian and Alaska Native (AI/AN) women vary across the US. A novel application of Bayesian disease mapping was applied to quantify potential inequities in 10-year BC incidence in New Mexico (NM) in order to better inform health equity initiatives within the AI/AN at-risk population. Methods: Surveillance, Epidemiology, and End Results Program (SEER) data from 2005 to 2014 were used to identify new cases of BC within the 33 counties in NM. Initially, a Poisson-gamma model was applied to quantify the reduction in risk of BC within the at-risk AI/AN population compared with the general at-risk population. To account for spatial variation and to address the small area estimation problem inherent in these data by borrowing strength globally and locally in NM, we applied Bayesian disease mapping to the counts of county-level BC cases. We quantified the disparity effect, as measured by the rate ratio (95% credible interval [CI]), comparing the incidence of BC between at-risk AI/AN and non-AI/AN women, and assessed if the rate ratio differed between counties. Markov chain Monte Carlo sampling was used to estimate posterior quantities and the deviance information criterion was used for model selection. Results: In 2010, 1,041,758 women were at-risk for BC of which 107,656 (10.3%) were AI/AN women. During the 10-year study period, 12,974 new BC cases were recorded in the general at-risk population. In the at-risk AI/AN population, the expected number of new cases during the 10-year study period, therefore, was 1,340.74; however, 597 incidence cases of BC were diagnosed in the at-risk AI/AN population resulting in a posterior mean for the true relative risk of 0.445 (95% CI: 0.410, 0.482). Based on the selected model that accounted for over dispersion and spatial correlation among the 33 counties, the posterior mean of the overall adjusted rate ratio was 0.405 (95% CI: 0.336, 0.478). The adjusted rate of BC in AI/AN women was 0.40 times the corresponding adjusted rate for women who were non-AI/AN. Further, the adjusted rate ratios were similar for each county. Conclusions: The novel application of Bayesian disease mapping to these data provided substantial evidence of a significant overall disparity effect in BC incidence within NM between at-risk AI/AN and non-AI/AN women, which was more marked than previous reports. Targeted state-wide health equity initiatives may lead to reducing disparities in BC incidence in AI/AN at-risk women.