Insights

The healthcare AI
research brief.

A weekly format for evidence on preventive care, chronic disease, FHIR infrastructure, privacy-preserving learning, social determinants of health, and explainable models.

Research archive

FHIR Infrastructure Is Maturing: While Shared Medical Representations Create New Privacy Risks

FHIR and CQL moved closer to supporting executable cross-institution quality measurement, while new studies demonstrated culturally contextual prediction and earlier chronic-disease risk detection. The strongest privacy warning came from medical-image foundation-model representations that could link patients across datasets with greater than 90% accuracy.

Read the brief

Federated Health AI Is Advancing: But Local Data Is Not the Same as Privacy

Federated clinical models are becoming more operationally practical, but keeping records inside each institution does not prevent sensitive information from leaking through model updates. The strongest preventive-care evidence also shows that transparent models can perform as well as more complex systems while making subgroup failures easier to detect.

Read the brief

Split-Federated ECG Monitoring Advances – But Its Intermediate Representations Can Reveal Patient Signals

New evidence showed that split-federated ECG models can operate on constrained devices while still leaking reconstructable physiological signals. A randomized maternal-health trial also found that access to a generative-AI chatbot did not improve anxiety, depression, or healthcare outcomes.

Read the brief

Federated Health AI Is Improving: But Accuracy Still Hides Serious Trust Problems

Federated health AI advanced this week through new methods for measuring explanation consistency and protecting rare classes during decentralized training. The evidence also exposed persistent weaknesses: simulated institutions, incompatible clinical outcomes, unproven privacy claims, and no demonstrated improvement in chronic-disease outcomes.

Read the brief