Quantum-Aware Privacy Risk Evaluation for Differentially Private Federated Healthcare Analytics

Federated learning enables collaborative healthcare analysis without requiring centralized patient data; however, the updates to the machine learning model may contain private information. In this work, a privacy risk scoring framework considering quantum attacks is proposed for differentially private federated healthcare analytics by applying UCI Diabetes 130-US Hospitals dataset to predicting 30-day readmission. The evaluation approach utilizes differential privacy, secure aggregation, quantum communication simulation, membership inference attack, and Quantum-Aware Privacy Risk Score. From the results, conventional Federated ML had a High QPRS of 97.5000, while DP-FL, DP-FL with secure aggregation, and QADP-PQSA-based evaluation method obtained Low QPRS with equivalent F1-score. With fixed non-IID evaluation, when DP noise multiplier is set to 1.0, the metrics of accuracy, F1-score, MIA accuracy, and QPRS were 0.7101, 0.2676, 0.5022, and 3.0937, respectively.