Landscape risk factors for Lyme disease in the eastern broadleaf forest province of the Hudson River valley and the effect of explanatory data classification resolution. (January 2015)

• Record Type: Journal Article
• Authors: Messier, Kyle P.; Jackson, Laura E.; White, Jennifer L.; Hilborn, Elizabeth D.
• Is Part Of: Spatial and spatio-temporal epidemiology. Volume 12(2015)
• Page Start: 9
• Page End: 17
• Publication Date: 2015-01
• Subjects: AIC Akaike Information Criteria -- NLCD National Landcover Database -- LD Lyme disease -- NB2 Negative Binomial Regression Model Parameterization; Land use -- Landcover -- Lyme disease -- Landscape design -- Negative binomial regression -- New York; Epidemiology -- Statistical methods -- Periodicals; Epidemiology -- Periodicals; 614.4072
• DOI: 10.1016/j.sste.2014.10.002
• Languages: English
• ISSNs: 1877-5845

Abstract

Highlights: We model Lyme disease in Hudson River valley of New York. We assess the effect of National Landcover Database (NLCD) explanatory data classification resolution on model outcomes. Lyme disease incidence model is robust to NLCD explanatory data classification resolution. Abstract: This study assessed how landcover classification affects associations between landscape characteristics and Lyme disease rate. Landscape variables were derived from the National Land Cover Database (NLCD), including native classes (e.g., deciduous forest, developed low intensity) and aggregate classes (e.g., forest, developed). Percent of each landcover type, median income, and centroid coordinates were calculated by census tract. Regression results from individual and aggregate variable models were compared with the dispersion parameter-based R 2 ( R α 2 ) and AIC. The maximum R α 2 was 0.82 and 0.83 for the best aggregate and individual model, respectively. The AICs for the best models differed by less than 0.5%. The aggregate model variables included forest, developed, agriculture, agriculture-squared, y -coordinate, y -coordinate-squared, income and income-squared. The individual model variables included deciduous forest, deciduous forest-squared, developed low intensity, pasture, y -coordinate, y -coordinate-squared, income, and income-squared. Results indicate that regional landscape models for Lyme disease rate are robust to NLCD landcover classification resolution.