Measuring the impact of spatial perturbations on the relationship between data privacy and validity of descriptive statistics

Abstract Background Like many scientific fields, epidemiology is addressing issues of research reproducibility. Spatial epidemiology, which often uses the inherently identifiable variable participant address, must balance reproducibility with privacy. In this study, we assess impact several different data perturbation methods on key spatial statistics and patient Methods We analyzed patterns in full set address-level mortality from Lawrence, MA during period 1911 to 1913. The original death locations were perturbed using seven published approaches stochastic deterministic anonymization. Key descriptive calculated for each perturbation, including changes pattern center, Global Moran’s I, Local distance k-th nearest neighbors, L-function (a normalized form Ripley’s K). A spatially adapted k-anonymity was used measure privacy protection conferred by method, its compliance HIPAA GDPR standards. Results Random at 50 m, donut masking between 5 Voronoi maintain validity better than other perturbations. Grid center both 100 × 250 m cells led large statistics. None adhered standard that all points have a > 10. All employed had least 265 points, or over 6%, not adhering standard. Conclusions Using applied analysis, compliant de-identification compatible maintaining as measured our chosen summary Further should investigate alternate balancing tradeoffs preservation public health are medical importance.