Abstract:
|
A learning healthcare system is powered by experience of patients and care providers collected as part of routine healthcare delivery. The real-world experience may come from distinct individuals whose information is collected in various sources (e.g., patients from different health plans or delivery systems; a setting known as horizontally partitioned data), from the same individuals whose information is available in multiple sources (e.g., patients' insurance claims and electronic health records; a setting known as vertically partitioned data), or both. Sharing information across delivery systems or organizations has proven to be challenging in practice due to concerns about privacy, confidentiality, security, and proprietary interests. Privacy-protecting distributed analytic methods are developed to facilitate multi-database analysis. The goal is to allow sophisticated and complex analysis without sharing potentially identifiable information. This presentation will describe recent development in privacy-protecting analytic and data-sharing methods, including summary score-based methods and distributed regression, and their performance in simulations and empirical studies.
|