August 22, 2022: Privacy-preserving computation of matching groups

MAster assignment

privacy-preserving computation of matching groups

TYPE : MASTER M-CS

Period : Jan 2023 - 

Student : Gansel, A.X.G. (Antoine, Student M-CS)

Date Final project : 

Thesis t.b.a.

Supervisors:

Description:

The comparison of two similar (but disjoint) groups, with similar features except for a single, differentiating characteristic, is an important tool for retrospectively determining whether exposure to the characteristic has an effect on the outcome[1]. For example, consider a hospital that collects several features about their patients, such as age, sex, gotten treatments, and treatment effectiveness. The hospital would like to measure whether receiving a COVID vaccination has impact on the effectiveness of a treatment. This is measured by comparing the effectiveness of clients that have gotten their vaccination, with a control group (a matching cohort) of different clients with similar features that did not get their vaccination.

However, the COVID vaccination records in the Netherlands are stored by the RIVM, and since the vaccination record of a client is highly sensitive data, this means that the hospital does not have access to it (for analytics). Similarly, the data that the hospital collects about their clients is privacy-sensitive data. Therefore, these two parties cannot simply share their databases with each other to select the two groups (cohorts) for comparison without violating the privacy of their clients.

This project aims to design and implement a solution that is sub-quadratic in the database size. A solution quadratic in the database size has already been developed by the SCS group for this problem, and the running prototype for the quadratic solution can be used as starting point for this project. During this research we will explore solutions based on oblivious sorting, but you are free to come up with solutions of your own. Further, in the project we will investigate the computation-theoretic and empiric performance of the solution(s).

Requirements:

Related reading and references:

 [1] https://emj.bmj.com/content/emermed/20/1/54.full.pdf