Toward a More Accurate Accrual to Clinical Trials: Joint Cohort Discovery Using Bloom Filters and Homomorphic Encryption

AMIA Jt Summits Transl Sci Proc. 2020 May 30:2020:162-170. eCollection 2020.

Abstract

Reliable cohort discovery is an essential early part of clinical study design. Indeed, it is the defining feature of many clinical research networks, including the recently launched Accrual to Clinical Trials (ACT) network. As currently deployed, however, the ACT network only allows cohort queries in isolated silos, rendering cohort discovery across sites unreliable. Here we demonstrate a novel protocol to provide network participants access to more accurate combined cohort estimates (union cardinality) with other sites. A two-party Elgamal protocol is implemented to ensure privacy and security imperatives, and a special attribute of Bloom filters is exploited for accurate and fast cardinality estimates. To emulate mandatory privacy protecting obfuscation factors (like those applied to the counts reported for individual sites by ACT), we configure the Bloom filter based on the individual site cohort sizes, striking an appropriate balance between accuracy and privacy. Finally, we discuss additional approval and data governance steps required to incorporate our protocol in the current ACT infrastructure.