Federated learning for cardiovascular disease prediction: a systematic review of clinical applications, validation, and translation readiness - Summary - MDSpire
Advertisement
Federated learning for cardiovascular disease prediction: a systematic review of clinical applications, validation, and translation readiness
To synthesize how federated learning has been implemented for cardiovascular disease prediction and to identify factors that determine clinical translation readiness.
Key Findings:
Twenty-two studies were included, covering early screening, clinical diagnosis, prognostic evaluation, and treatment-related decision support.
Most studies utilized horizontal federated learning with FedAvg baselines, addressing non-IID heterogeneity and personalization.
Federated learning performance often approached centralized training, but evidence was predominantly retrospective.
Inconsistent reporting on validation, calibration, subgroup performance, privacy safeguards, and system costs was noted.
Interpretation:
Limitations:
The majority of studies relied on public datasets or simulated client splits.
Inconsistent reporting on critical aspects such as held-out-site validation and privacy safeguards.
Conclusion:
Further validation and reporting improvements are needed.
Qualitative interviews identified four themes involving emergency challenges and response, teamwork, psychological stress and coping, and professional growth needs in trauma surgery.
