Identifying Pediatric Cancer Clusters in Florida Using Log-Linear Models and Generalized Lasso Penalties

Abstract

We discuss the identification of pediatric cancer clusters in Florida between 2000 and 2010 using a penalized generalized linear model. More specifically, we introduce a Poisson model for the observed number of cases on each of Florida's ZIP Code Tabulation Areas (ZCTA) and regularize the associated disease rate estimates using a generalized Lasso penalty. Our analysis suggests the presence of a number of pediatric cancer clusters during the period over study, with the largest ones being located around the cities of Jacksonville, Miami, Cape Coral/Fort Meyers and Palm Beach.

Modeling pediatric tumor risks in Florida with conditional autoregressive structures and identifying hot-spots

“…An estimated temporal autocorrelation parameter suggests that we need to incorporate temporal dependence in the model. Estimates of relative risks by WRSEL could capture hotspot areas which is similar to the results obtained by the other researchers (Amin et al, 2014;Wang and Rodriguez, 2014;Zhang et al, 2014). While hot-spot regions by the Poisson ICAR model under WRSEL substantially vary over the years, those by spatio-temporal model do not.…”

“…We adopt weighted ranks squared error loss to identify high risk regions which provides similar results with other researchers who have worked on the same data set (e.g. Zhang et al, 2014;Wang and Rodriguez, 2014). Our results, thus, provide additional statistical support on those identified high risk regions discovered by the other researchers.…”

“…Regions in further south from that cluster, which are located near Miami forms a cluster all years. Wang and Rodriguez (2014) also found that near Miami areas form a highly concentrated cluster when they map with covariate-adjusted relative risks. North Florida areas are clustered largely in [2001][2002][2003].…”

“…It becomes smaller and moves gradually over years after 2004. This phenomenon was also captured by Wang and Rodriguez (2014) by having high but not the highest in large areas of north Florida. Amin et al (2014) also found that their approach for each cancer type among brain cancer, leukemia and lymphoma makes a cluster in south Florida and includes Miami in common.…”

“…Two significant spatial clusters appear in south and central north regions when all cancer type were considered. Wang and Rodriguez (2014) adopted the penalized generalized linear model to find out a cluster. The data were fitted with Poisson regression to log-linear model having a conditionally autoregressive structure, which is commonly used in disease mapping.…”