Environment/Climate Change
Impact of population size denominators for census-tract level cancer incidence hotspot identification Johnnie Newton* Johnnie Newton Natalie DuPré
Background: Getis-Ord Gi* was used to map hotspots of cancer rates at the county-level. The impact of population size changes over time on this method is not clear. We investigate the sensitivity of population size denominator selection for this hotspot methodology at a smaller geographical unit (census tract) to identify cancer hotspots.
Methods: To calculate age-standardized colorectal cancer (CRC) and lung cancer (LC) rates in KY’s 1,115 census tracts, we used KY Cancer Registry case counts from 1995-2021 and age-specific population-size denominators from the 2018 American Community Survey (Method 1) and by summing annual tract estimates from the National Historical GIS 2000-2019 (Method 2). We excluded tracts designated as special use (n=10). The Getis-Ord Gi* statistic identified census-tract level hotspots of high CRC and LC rates. The Kappa (K) statistic assessed method agreement.
Results: The CRC analysis detected 1050 concordant census tracts; 44 hotspots were identified by both Methods, Method 1 uniquely identified 42 hotspots, and 13 were unique to Method 2. For the LC analysis, 1052 tracts were concordant non-hotspot tracts; 61 hotspots were identified by both methods, 43 hotspots were unique to Method 1, and 10 were unique to Method 2. CRC agreement was moderate (K= 0.59; 95% CI 0.49, 0.69) and LC agreement was substantial (K= 0.67; 95%CI 0.59, 0.75). Visually, the hotspots were in the same general geographic areas regardless of method. Method 1 detected predominantly urban hotspots (CRC 85% urban; LC 70% urban, 30% rural). Method 2 detected fewer hotspots with a higher proportion being rural than Method 1 (CRC 68% urban; LC 59% urban, 41% rural).
Conclusion: Single year population size denominators identified more hotspot census tracts, particularly in urban areas, than using estimated population sizes summed across nearly two decades. Public health practitioners should use both methods to best detect the extent of cancer incidence at small geographies.