Kassaye Yitbarek Yigzaw

Privacy preserving health data processing

The usage of electronic health data from different sources for statistical analysis requires a toolset where the legal, security and privacy concerns have been taken into consideration. The health data are typically located at different general practices and hospitals. The data analysis consists of local processing at these locations, and the locations become nodes in a computing graph. To support the legal, security and privacy concerns, the proposed toolset for statistical analysis of health data uses a combination of secure multi-party computation (SMC) algorithms, symmetric and public key encryption, and public key infrastructure (PKI) with certificates and a certificate authority (CA). The proposed toolset should cover a wide range of data analysis with different data distributions. To achieve this, large set of possible SMC algorithms and computing graphs have to be supported.

Towards privacy-preserving computing on distributed electronic health record data

The paper reports on work in progress towards construction of a peer-to-peer framework for privacy preserving computing on distributed electronic health data. The framework supports three different types of federated queries. For privacy-preserving computing, we proposed distributed secure multi-party computation (SMC), where each peer is only involved in secure computations with some of the peers. We hypothesize distributed SMC could enable to achieve more efficient and scalable computing solutions. The architecture of the framework is also described.

The snow system: A decentralized medical data processing system.

Systems for large-scale reuse of electronic health record data is claimed to have the potential to transform the current health care delivery system. In principle three alternative solutions for reuse exist: centralized, data warehouse, and decentralized solutions. This chapter focuses on the decentralized system alternative. Decentralized systems may be categorized into approaches that move data to enable computations or move computations to the where data is located to enable computations. We describe a system that moves computations to where the data is located. Only this kind of decentralized solution has the capabilities to become ideal systems for reuse as the decentralized alternative enables computation and reuse of electronic health record data without moving or exposing the information to outsiders. This chapter describes the Snow system, which is a decentralized medical data processing system, its components and how it has been used. It also describes the requirements this kind of systems need to support to become sustainable and successful in recruiting voluntary participation from health institutions.

Evaluation of secure multi-party computation for reuse of distributed electronic health data

There has been an increasing need for reuse of health data (i.e. research, quality assurance, public health, and commercial applications). However, privacy and legal issues have limited the reuse. Several privacy-preserving techniques (both centralized and distributed) have been developed to allow reuse of health data while preserving privacy. The distributed techniques enable institutions to jointly compute on their private data while preserving the privacy of their data. However, the centralize approach applies perturbation or anonymization technique on the private data before giving out the data for computation. This paper presents criteria, such as privacy level, linkability support, efficiency and scalability, to evaluate distributed privacy preserving techniques.

A communicable disease prediction benchmarking platform

The paper presents a platform for benchmarking disease prediction algorithms and mathematical models. The platform is applied to compare Bayesian and compartmental disease prediction models using. We used weekly aggregated cases of various diseases collected from a microbiology laboratory that covers northern Norway. The platform enables integration and benchmarking of various disease prediction models. Our benchmark shows that the Bayesian model was better on predicting the number of cases on a weekly basis. Normalized root mean square error (NRMSE) for the Bayesian prediction was within the range 0.072-0.1498 for weekly predictions, 0.171-0.254 for monthly. The compartmental SIR(S) model achieved a NRMSE of 0.133 for the weekly prediction against Influenza A data. Disease prediction models benchmarking platforms can help to improve the status of disease prediction systems, investment and time of development. It can speeds up mathematical modeling through its integrated environment for testing and evaluation.