193 papers
This study demonstrates that Federated Learning effectively enables privacy-preserving, multi-site machine learning for HIV care across diverse international cohorts, achieving performance comparable to centralized models while significantly outperforming local site-specific approaches.
This study demonstrates that passive monitoring of wearable-derived data, specifically reduced step counts and elevated resting heart rates, can effectively identify heart failure patients at increased risk of hospitalization up to three days in advance.
This study developed and validated a multi-label BERT-based natural language processing application that successfully extracts and categorizes various forms of violence, patient roles, and contextual details from unstructured mental health records, achieving high performance on most features except temporal aspects.
This study demonstrates that Virtual Pooling (VP) enables accurate, end-to-end execution of multicenter clinical research and causal inference—including data harmonization, preprocessing, and statistical analysis—across institutions without centralized data sharing, while exactly replicating ground-truth results from a published diabetic eye disease study with minimal latency and no infrastructure changes.
This study demonstrates that an AI-derived ECG age gap, calculated using a foundation model on UK Biobank data, serves as a powerful non-invasive digital biomarker for stratifying cardiovascular risk by quantifying occult accelerated cardiac aging, where an overestimated age gap significantly predicts major adverse cardiovascular events.
The authors have withdrawn their manuscript regarding a high-precision model for predicting ICU admission from ED triage due to identified analytical issues requiring revision, and they request that the work not be cited.
This paper presents a regulatory-informed statistical framework that quantifies the semantic and internal repeatability and reproducibility of large language models, demonstrating that these metrics vary significantly based on prompting strategies and model configurations, are often independent of diagnostic accuracy, and are essential for systematically evaluating LLM reliability in biomedical applications.
This study of 442 adults demonstrates that wearable-derived cardiovascular fitness age is a valid metric that significantly correlates with independent lifestyle factors like sleep quality and physical activity, while also tracking physiological improvements over time.
This study presents a pragmatic evaluation of AIFYA, a human-supervised large language model-based clinical decision support system aligned with Kenya's national newborn protocols, demonstrating successful implementation, high expert-rated accuracy, and strong user adoption in low-resource public health facilities.
This study reveals that despite achieving high diagnostic accuracy on full histories, large language models exhibit critical safety failures—including discouraging essential investigations and inappropriately downgrading triage for life-threatening conditions, with significant gender disparities—when evaluated against 1,000 synthetic clinical transcripts that simulate real-world medical complexity.
This paper presents a methodology for implementing action plans to mitigate the negative impacts of Electronic Health Record (EHR) adoption, demonstrating that effective governance and multidisciplinary execution significantly resolve implementation issues and increase project success rates.
SleepJEPA is a foundational deep learning model trained on over 422,000 hours of at-home sleep data that effectively learns latent sleep representations to accurately predict long-term risks for ten major cardiovascular, metabolic, and neurological diseases, as well as estimate sleep stages and specific sleep disorders.
This study evaluates four major open health data repositories using a novel two-degree citation methodology to reveal that while open data consistently generates a ~10x indirect citation amplification across vastly different funding levels, significant disparities in global representation and persistent gender gaps in senior authorship highlight that data access alone cannot address structural inequities in research leadership.
This mixed-methods longitudinal audit reveals that health AI systems deployed in Nigeria significantly underperform compared to vendor claims, resulting in substantial preventable patient harm and exacerbated inequities, thereby underscoring the urgent need for mandatory independent post-deployment verification in low- and middle-income countries.
The paper introduces NiaAge, a clinically interpretable biological-age model derived from non-linear long-term mortality risk and normative aging trajectories, which outperforms chronological age and DNA-methylation clocks in predicting mortality, physical health, and cognitive health.
Analyzing data from nearly 260,000 participants in the All of Us Research Program, this study demonstrates that integrating social determinants of health with demographics significantly improves chronic disease risk prediction, revealing that mental health outcomes are primarily driven by experiential factors like stress and discrimination, whereas cardiometabolic conditions are more strongly influenced by structural neighborhood characteristics, thereby supporting the adoption of condition-specific social screening and targeted interventions to reduce health disparities.
This study validates the Intensive Documentation Index (IDI), a zero-burden mortality predictor derived from EHR documentation timestamps, demonstrating that its predictive accuracy is heavily dependent on temporal resolution, as evidenced by a significant performance gap between the high-latency MIMIC-IV cohort and the near-real-time HiRID cohort.
This study demonstrates that a social media-informed digital self-management platform, "Jang Geongang," significantly reduced population-level irritable bowel syndrome emergency department visits and unplanned hospitalizations in South Korea, particularly among younger adults and those with the diarrhea-predominant subtype, as evidenced by a controlled interrupted time series analysis of national health insurance data.
This study demonstrates that the sensitivity- and precision-maximizing RCT filter on Ovid MEDLINE significantly reduces the screening burden compared to the sensitivity-maximizing filter while retrieving nearly all eligible records, with any missed studies being recoverable through citation searching.
This study demonstrates that an LLM-based platform (Brim Analytics) achieves high accuracy and efficient processing for extracting cancer registry data from pathology reports, outperforming an ontology-driven system (DeepPhe) particularly in T stage classification across pancreatic and breast cancer cases.