Privacy-Preserving Predictive Models for Lung Cancer Survival Analysis

Privacy-preserving data mining (PPDM) is a recent emergent research area that deals with the incorporation of privacy preserving concerns to data mining techniques. We consider a real clinical setting where the data is horizontally distributed among different institutions. Each one of the medical institutions involved in this work provides a database containing a subset of patients. There is recent work that shows the potential of the PPDM approach in medical applications. However, there is few work in developing/implementing PPDM for predictive personalized medicine. In this paper we use real data from several institutions across Europe to build models for survival prediction for non-small-cell lung cancer patients while addressing the potential privacy preserving issues that may arise when sharing data across institutions located in different countries. Our experiments in a real clinical setting show that the privacy preserving approach may result in improved models while avoiding the burdens of traditional data sharing (legal and/or anonymization expenses).