Privacy-preserving statistical analysis methods and their applications on health research

Privacy consideration in health data usually prevents researchers and other data users from conducting their research. Also, data is distributed through various health organizations such as hospitals, thus gathering distributed health information becomes impractical. Various approaches have been proposed to preserve the patients privacy, whilst allowing researchers to perform mathematical operations and statistical analysis methods on health data, such as anonymization and secure computation. Data anonymization reduces the accuracy of the original data; hence the final result would not be precise enough. In addition, there are several known attacks on anonymized data, such as using public information and background knowledge to re-identify the original data. On the other hand, secure computation is more precise and the risk of data re-identification is zero; however, it is computationally less efficient than data anonymization. In this thesis, we implemented a web-based secure computation framework and propose new secure statistical analysis methods. Using the proposed web application, researchers and other data users would be able to perform popular statistical analysis methods on distributed data. They will be able to perform mathematical operations and statistical analysis methods as queries through different data owners, and receive the final result without revealing any sensitive information. Digital Epidemiology Chronic Disease Tool (DEPICT) database, which contains real patients information, will be used to demonstrate the applicability of the web application