Evaluating the Efficacy of Federated Scoring Systems with Heterogeneous Electronic Health Records

Qiming Wu, Siqi Li, Di Miao, Yuqing Shang, Xin Li, Nan Liu

Keywords: federated learning, interpretable machine learning, electronic health records, privacy-preserving

Abstract: Federated learning in healthcare research has primarily focused on black-box models, leaving a notable gap in interpretability crucial for clinical decision-making. While scoring systems, acknowledged for their transparency, are widely employed in clinical science, there are notably limited privacy-preserving solutions for scoring system generators. FedScore, an example of such a solution, has been demonstrated using artificially partitioned data. In this study, we further improve FedScore and conduct empirical experiments utilizing real-world heterogeneous clinical data.

Submission Number: 67