A Geocomputational Process for Characterizing the Spatial Pattern of Lung Cancer Incidence in New Hampshire
Data that are detailed both spatially and demographically are increasingly available for disease incidence and background population at risk. Rapidly growing computer capability makes computational approaches that rely less on statistical assumptions and models increasingly feasible. Many previous methods for characterizing spatial variation of disease incidence and detecting clusters were designed for highly aggregate data and relied on modeling rather than computation. This article presents a geocomputational methodology designed to (1) take advantage of the detailed data, (2) quantify uncertainties in the analysis, and (3) reduce reliance on statistical assumptions. This methodology is an assemblage of several special methods for dealing with some critical issues in disease mapping with detailed data. These include (1) a kernel density calculation using a case-side adaptive bandwidth to address an uneven distribution of background population; (2) a restricted Monte Carlo process to maximize the utilization of imprecise location information and quantify the uncertainty in this utilization; and (3) an integration of the direct and indirect standardizations in epidemiology to maximize the utilization of the available demographic information and ease the bandwidth selection in the kernel density estimation. Using the New Hampshire State Cancer Registry data of individual lung cancer incidence cases, the LandScanTM data that detail the spatial distribution of the population, and demographic information from Census data, the methodology was able to create high-resolution maps presenting the spatial variation of lung cancer risk in New Hampshire and the associated uncertainties.