THE MYTH OF 1000 METER BUFFERS: SPATIAL SCALE EFFECTS IN ENVIRONMENTAL RISK-FACTOR MODELING FOR DISEASES

Studies attempting to identify environmental risk factors for diseases can be seen to extract candidate variables from remotely sensed datasets, often using a single buffer-zone surrounding locations where disease statuses are recorded. A retrospective case-control study using leptospirosis infection in dogs was conducted to verify any effects of changing buffer-zones (spatial extents) on the risk factors derived. The case-control study included 94 case dogs predominantly selected based on positive polymerase chain reaction (PCR) test for leptospires in urine, and 185 control dogs based on negative PCR. Land cover features from National Land Cover Dataset (NLCD) and Kansas Gap Analysis Program (KS GAP) around geocoded addresses of cases/controls were extracted using multiple buffers at every 500 m up to 5,000 m, and multivariable logistic models were used to estimate the risk of different land cover variables to dogs. The types and statistical significance of risk factors identified changed with an increase in spatial extent in both datasets. Leptospirosis status in dogs was significantly associated with developed high-intensity areas in models that used variables extracted from spatial extents of 500-2000 m, developed medium-intensity areas beyond 2,000 m and up to 3,000 m, and evergreen forests beyond 3,500 m and up to 5,000 m in individual models in the NLCD. Significant associations were seen in urban areas in models that used variables extracted from spatial extents of 500-2,500 m and forest/woodland areas beyond 2,500 m and up to 5,000 m in individual models in Kansas gap analysis program datasets. The use of ad hoc spatial extents can be misleading or wrong, and the determination of an appropriate spatial extent is critical when extracting environmental variables for studies.