Photo by Bill Branson
Several statistical analyses have revealed higher-than-expected rates of pediatric cancers in 2 regions of Florida—the Miami metro area and an area west of the Everglades.
The anomalous rates were detected by 5 different research teams—each using different epidemiological and statistical methodology—on a data set spanning the period from 2000 to 2010 that was provided by the Florida Association of Pediatric Tumor Programs (FAPTP).
Lance A. Waller, PhD, of Emory University in Atlanta, Georgia, reviewed the different analyses and described his findings in Statistics and Public Policy.
The research groups applied different analytical approaches to achieve the same goal: detect spatio-temporal pediatric cancer clusters. The analyses used familiar methods—scan statistics, classification, and hierarchical Bayesian modeling—as well as some ideas new to disease clustering: wombling and machine learning.
During their respective analyses of the FAPTP data, the research groups found several suggestive results. For instance, each approach identified local areas in which the observed pediatric cancer rate is significantly higher than the rate expected, given the number of people at risk.
While the precise areas of high reported risk differ between methods, the groups identified a few common results that overlap but are not identical.
For example, all 5 teams identified significantly elevated rates of pediatric cancers in an urban area within collections of ZIP code tabulation areas (ZCTAs) in the Miami metro area and in an area just west of the Everglades. (ZCTAs are geographic areas defined by the US Census Bureau to provide a link between census geography—blocks, block groups, and tracts—and US Postal Service ZIP code areas.)
One analysis suggested the local increase west of the Everglades is based on 2 cases, both classified as “other” race, while another analysis indicated that this cluster is limited to the year 2000.
The observed elevated rates near Miami involved a much larger population size and many more cases, factors that complicate the identification of any shared characteristics common to cases in the cluster.
The analyses also revealed other patterns in the data. Dr Waller said an analysis that revealed a statewide increase in the baseline pediatric cancer incidence rate occurring between 2005 and 2006 merits a closer look to see whether this result represents an overall increase in risk or a change in reporting, because the statistical analysis does not reveal potential reasons for the change.
There also are subtle differences between the specific clusters identified by the various analytical approaches. Comparisons across analyses revealed characteristics of the detected patterns, including the number of cases (2), types of cancer (leukemia or brain/central nervous system cancer), and the racial composition and timing of the cluster west of the Everglades.
As the methods the researchers used don’t completely agree on the precise location, boundaries, and make-up of the detected clusters, the findings suggest a single method may not prove sufficient for such analyses, Dr Waller said.
He added that the identified clusters are geographically quite large and therefore unlikely to provide clear links between particular environmental exposures to local risks.
“While the results do not identify a ‘smoking gun’ in the form of a shared environmental exposure in high-incidence areas, the results do provide epidemiologic insight into the local demographics of the incidence of pediatric cancer cases and suggest more detailed assessment of migration patterns in the Miami area,” Dr Waller said.
“Policy-wise, the results point to responsibly responsive next steps of detailed description of the cases and the at-risk population in the detected areas to summarize local features in the data, particularly the race of cases west of the Everglades and demographic descriptors of any shifts in the at-risk population in the Miami area during the study period.”
Policy responses by local and state health officials may involve more detailed follow-up, including additional data collection, exposure surveys, or in-depth investigation of case histories within a reported cluster.
Dr Waller added that estimated cancer rates consist of the local number of cases (reported by the FAPTP) and the local number of children at risk (reported by the decennial census). Higher-than-expected rates can result from unusually high numbers of reported cases or low numbers of reported local residents.
Since Miami, like many urban areas, often experiences rapid changes in population size and composition between decennial censuses, it is important to assess the accuracy of both data components. Dr Waller suggested, as a first step, assessing the accuracy of the case counts and the inter-census population projections defining the local rates.
“State and local health departments and public health agencies regularly respond to cluster reports from the public,” he said. “Typically, a responsive and effective response is not based on a detailed new epidemiologic study but, rather, is based on education, assessments of local concentrations of demographic risk factors associated with the reported cluster, and an assessment of the distribution of numbers of cases expected given the local demographics.”
