Geospatial Imprecision With Constraints for Precision Public Health: Algorithm Development and Validation

Abstract

Background: Location and environmental social determinants of health are increasingly important factors in both an individual's health and the monitoring of community-level public health issues.

Objective: We aimed to measure the extent to which location obfuscation techniques, designed to protect an individual's privacy, can unintentionally shift geographical coordinates into neighborhoods with significantly different socioeconomic demographics, which limits the precision of findings for public health stakeholders.

Methods: Point obfuscation techniques intentionally blur geographic coordinates to conceal the original location. The pinwheel obfuscation method is an existing technique in which a point is moved along a pinwheel-like path given a randomly chosen angle and a maximum radius; we evaluate the impact of this technique using 2 data sets by comparing the demographics of the original point and the resulting shifted point by cross-referencing data from the United States Census Bureau.

Results: Using poverty measures showed that points from regions of low poverty may be shifted to regions of high poverty; similarly, points in regions with high poverty may be shifted into regions of low poverty. We varied the maximum allowable obfuscation radius; the mean difference in poverty rate before and after obfuscation ranged from 6.5% to 11.7%. Additionally, obfuscation inadvertently caused false hot spots for deaths by suicide in Cook County, Illinois.

Conclusions: Privacy concerns require patient locations to be imprecise to protect against risk of identification; precision public health requires accuracy. We propose a modified obfuscation technique that is constrained to generate a new point within a specified census-designated region to preserve both privacy and analytical accuracy by avoiding demographic shifts.

Keywords: confidentiality; geocoding; geographic; geography; geospatial; location; locations; low income; obfuscation; poverty; precision; privacy; security; social determinants of health; spatial.

Figure 1

Pinwheel obfuscation at theta 45 degrees (A) and 15 degrees (B).

Figure 2

Poverty and point obfuscation of original points (orange) to obfuscated points (blue) using a pinwheel (A) and a geographically constrained pinwheel (B).

Figure 3

Hot spot analysis of suicides using original data (A) and obfuscated data (B).