Distributed data processing for public health surveillance

Lazarus, Ross; Yih, Katherine; Platt, Richard

Before describing our distributed system, we briefly review the more familiar model of centralized aggregation and processing of PHI for surveillance. In the more traditional type of system, individual patient records, often containing potentially identifiable information, such as date of birth and exact or approximate home address, are transferred, usually in electronic form, preferably through some secured method, to a central secured repository, where statistical tools can be used to develop and refine surveillance procedures. One of the main benefits of this data-processing model is that the software and statistical methods can be changed relatively easily to accommodate changes in requirements, because they only need to be changed at the one central location where analysis is taking place. As long as appropriate details have been captured for each individual encounter of interest, the raw data can be re-coded or manipulated in different ways. Only one suite of analysis code is needed, and because it is maintained at a single, central location, costs for upgrading and maintenance are small. Inadvertent disclosure of PHI is always a potential risk with centralized systems. Even where minimally identifiable data are stored in each record, the probability of being able to unambiguously identify an individual increases as multiple, potentially linkable records for that individual accrue over time.