CONVINCED -– Enabling privacy-preserving survival analyses using Multi-Party Computation

Integraal Kankercentrum Nederland (IKNL) maintains the Netherlands Cancer Registry (NCR). The NCR is a comprehensive, population-based registry with data on diagnosis and treatment. However, in an increasing amount of research projects, there is a need for additional data items that can be found in other databases. In particular the setting where the data is vertically partitioned is relevant when considering data that is distributed in the Netherlands: in this case, each party maintains a portion of the attributes, e.g. clinical, demographic and psycho-social variables, of the same group of patients. This is needed to generate a more complete picture of the patient, since not all data necessarily needs to be part of a cancer registry. However, these underlying attributes are privacy-sensitive and cannot be readily shared amongst involved parties.
In a joint research project, TNO and IKNL investigated the possibilities to analyze vertically partitioned data without compromising the patient’s privacy, and developed new solutions to do this using Multi-Party Computation (MPC). In particular this project focused on survival analysis: an often-used technique in oncology data research that can be used to indicate how likely someone is to be alive a few years after diagnosis. Additionally, it can give insight in which characteristics might relate to the chances of survival.
This report describes the results and conclusions of this project in more detail. In particular, it has been shown that MPC can indeed be employed to perform survival analysis on vertically-partitioned data, without having to reveal any sensitive data. For some survival analysis techniques (Kaplan-Meier and log rank) this can be done while keeping the performance of the secure solution high enough for practical application on realistic data sets. For more complex survival analysis techniques (Cox Proportional Hazards) additional research is needed to improve the performance before it can be applied in practice. Follow-up steps to bring the developed solutions towards operational applications have been identified.