This bibliography brings together research and guidance across fairness, synthetic data, clinical AI systems, and governance. These sources inform how we frame questions and evaluation design.
For Synset’s own published results and their scope, see Evidence for what current results establish.
Research on bias across clinical AI pipelines, subgroup reporting, and the relationship between fairness methods and health equity.
Maps clinical AI fairness research and gaps in bias attributes, fairness methods, and clinical translation.
DOI: 10.1038/s41746-025-01667-2Reviews bias across the medical AI pipeline and implications for clinical decisions and mitigation.
DOI: 10.1371/journal.pdig.0000651Discusses health-equity, transparency, and governance considerations, including systemic bias.
DOI: 10.1111/1468-0009.12712Examines demographic subgroup reporting in commercial radiology AI validation and gaps in fairness assessment.
DOI: 10.1007/s00330-026-12652-yTests demographic cues in LLM choices within synthetic clinical ethical scenarios.
DOI: 10.1007/s41666-025-00211-xSummarizes healthcare AI use and concerns regarding racial and ethnic disparities.
Read sourceWork on synthetic-data methods, digital twins, recursive generation, fidelity, and performance changes over time.
Maps synthetic-data research and the gap between research growth and reported operational use.
DOI: 10.64898/2026.09.12.26360876Proposes causal inference and digital twins for trial design and analysis alongside validation and regulatory constraints.
DOI: 10.1038/s41746-026-02871-4Reviews healthcare synthetic-data methods and open-source tools across modalities, privacy, and fidelity.
DOI: 10.1016/j.csbj.2024.07.005General machine-learning research on distribution loss when models are recursively trained on generated data.
DOI: 10.1038/s41586-024-07566-yAnalyzes performance degradation over time across industries, including healthcare operations.
DOI: 10.1038/s41598-022-15245-zDiscusses synthetic-data opportunities and fidelity, validation, and inherited-bias risks.
Read sourceMaps generative AI methods for biomedical time-series across text, imaging, waveforms, and multi-omics.
DOI: 10.2196/59792Research on clinical language models and the knowledge structures and retrieval methods used around them.
Reviews healthcare causal graph neural networks and validation, computation, and causal-washing risks.
DOI: 10.1038/s41551-026-01742-3Reviews technical implementations, clinical uses, and ethical considerations of medical RAG, including gaps in clinical validation.
DOI: 10.1016/j.xcrm.2026.102927An NIH challenge on temporal representation and reasoning over evolving biomedical knowledge.
Read sourceProposes permutation-invariant set-distance rewards for chest X-ray report generation, candidate selection, and generation pruning.
DOI: 10.48550/arXiv.2606.00440Guidance and analysis on responsible development, evaluation, oversight, and deployment of health AI.
Uses 2023 AHA survey data on hospital predictive-model adoption and local evaluation of accuracy and bias.
DOI: 10.1377/hlthaff.2024.00842Describes PCCP content for AI-enabled device software submissions, planned modifications, validation, and impact assessment.
Read sourceMaps the landscape of U.S. state and federal regulatory sandboxes for health AI.
Read source