Privacy-Preserving Process Mining for Clinical Pathway Analysis and Classification

Process mining in healthcare offers significant benefits, including modeling patients' clinical pathways, identifying the load on healthcare infrastructure, and extracting useful statistics for analytics. However, these innovations pose privacy risks, such as the leakage or misuse of patient data, which could potentially disclose medical conditions and cause psychological distress or economic harm. In this context, privacy-preserving process mining is essential in order to maintain these advantages while safeguarding individual privacy. This paper proposes a framework that enables the analysis, classification and statistical study of patients' clinical pathways, while protecting against privacy threats. The framework supports various privacy-preserving techniques, such as introducing controlled noise through differential privacy, while integrating well-known process discovery and conformance checking algorithms for clinical pathways analysis and classification. We evaluate the effectiveness of the framework on an open-access dataset covering various diseases. The results demonstrate that we can achieve high classification accuracies for different clinical pathways while preventing malicious attempts to infer sensitive patient information, resulting in an effective trade-off between data utility and privacy guarantees.

Keywords: Healthcare, electronic health records, process mining, privacy-preserving techniques, differential privacy.