Use of geocoding techniques for epidemiological surveillance in the Federal District, Brazil: a case study using dengue.

Abstract

This study aimed to compare different address geocoding services and their applicability to epidemiological surveillance using dengue as an example. We applied a cross-sectional, descriptive study based on case notifications in the Notifiable Diseases Information System (SINAN) for the Brazilian capital in 2014 that includes complete postal code (CEP) information identified in the National Address Database for Statistical Purposes (CNEFE), which is considered the 'gold standard' for accuracy analysis. For records without CEP, georeferencing was performed through linkage of the original database with four geocoding tools: Google Maps, CNEFE, OpenStreetMap (OSM) and ArcGIS. Variables used for georeferencing were 'street name', 'code for municipality/ city of residency' and 'State' using accuracy rate estimate and mean spatial error (MSE) of case locations. The two most accurate models were used for kernel density (KD) analysis which is valuable for identifying priority areas for intervention. There were 18,206 dengue cases, 109 (0.6%) of which had correct CEP information and geocoded using CNEFE bases. The linkage results showed that Google Maps application programming interface (API) had an accuracy of 17.6% (MSE: 178.89km), CNEFE 9.0% (MSE: 17.24km), OSM 7.1% (MSE: 564.19km), and ArcGIS 3.7% (MSE: 2001.33km). Although overall accuracy values were modest, the best two models proven to be effective for KD analysis revealed similar patterns between Google Maps and CNEFE results but choosing the preferable geocoding technique should also financial resources. This study recommends the use of Google Maps API for georeferencing, followed by CNEFE.