A5039868033- Machine Learning in Healthcare
- Topic Modeling
- Artificial Intelligence in Healthcare and Education
- Biomedical Text Mining and Ontologies
- Genomics and Rare Diseases
- Natural Language Processing Techniques
- Trauma and Emergency Care Studies
- Artificial Intelligence in Healthcare
- Clinical Reasoning and Diagnostic Skills
- Data-Driven Disease Surveillance
- Injury Epidemiology and Prevention
- Cancer Genomics and Diagnostics
- Health, Environment, Cognitive Aging
- Radiomics and Machine Learning in Medical Imaging
- Maternal and fetal healthcare
- Advanced Graph Neural Networks
- Vaccine Coverage and Hesitancy
- Autopsy Techniques and Outcomes
- Emergency and Acute Care Studies
- Text Readability and Simplification
- Complex Network Analysis Techniques
- Meta-analysis and systematic reviews
- Healthcare Technology and Patient Monitoring
- Epilepsy research and treatment
- Advanced Statistical Process Monitoring
Brigham and Women's Hospital
2023-2025
Stanford University
2015-2025
Harvard University
2020-2025
Massachusetts Institute of Technology
2018-2022
Harvard University Press
2022
Harvard–MIT Division of Health Sciences and Technology
2021-2022
Uniformed Services University of the Health Sciences
2020
Johns Hopkins University
2020
United States Naval Medical Research Unit SOUTH
2020
Vanderbilt University Medical Center
2014
Contextual word embedding models such as ELMo and BERT have dramatically improved performance for many natural language processing (NLP) tasks in recent months. However, these been minimally explored on specialty corpora, clinical text; moreover, the domain, no publicly-available pre-trained yet exist. In this work, we address need by exploring releasing text: one generic text another discharge summaries specifically. We demonstrate that using a domain-specific model yields improvements 3/5...
Contextual word embedding models such as ELMo (Peters et al., 2018) and BERT (Devlin have dramatically improved performance for many natural language processing (NLP) tasks in recent months. However, these been minimally explored on specialty corpora, clinical text; moreover, the domain, no publicly-available pre-trained yet exist. In this work, we address need by exploring releasing text: one generic text another discharge summaries specifically. We demonstrate that using a domain-specific...
Large language models (LLMs) such as GPT-4 hold great promise transformative tools in health care, ranging from automating administrative tasks to augmenting clinical decision making. However, these also pose a danger of perpetuating biases and delivering incorrect medical diagnoses, which can have direct, harmful impact on care. We aimed assess whether encodes racial gender that its use
Prior research has shown that artificial intelligence (AI) systems often encode biases against minority subgroups. However, little work focused on ways to mitigate the harm discriminatory algorithms can cause in high-stakes settings such as medicine.In this study, we experimentally evaluated impact biased AI recommendations have emergency decisions, where participants respond mental health crises by calling for either medical or police assistance. We recruited 438 clinicians and 516...
Although recent advances in scaling large language models (LLMs) have resulted improvements on many NLP tasks, it remains unclear whether these trained primarily with general web text are the right tool highly specialized, safety critical domains such as clinical text. Recent results suggested that LLMs encode a surprising amount of medical knowledge. This raises an important question regarding utility smaller domain-specific models. With success general-domain LLMs, is there still need for...
Abstract Many areas of medicine would benefit from deeper, more accurate phenotyping, but there are limited approaches for phenotyping using clinical notes without substantial annotated data. Large language models (LLMs) have demonstrated immense potential to adapt novel tasks with no additional training by specifying task-specific instructions. Here we report the performance a publicly available LLM, Flan-T5, in patients postpartum hemorrhage (PPH) discharge electronic health records ( n =...
Abstract Background Large language models (LLMs) such as GPT-4 hold great promise transformative tools in healthcare, ranging from automating administrative tasks to augmenting clinical decision- making. However, these also pose a serious danger of perpetuating biases and delivering incorrect medical diagnoses, which can have direct, harmful impact on care. Methods Using the Azure OpenAI API, we tested whether encodes racial gender examined four potential applications LLMs domain—namely,...
Abstract Objectives Large language models (LLMs) are poised to change care delivery, but their impact on health equity is unclear. While marginalized populations have been historically excluded from early technology developments, LLMs present an opportunity our approach developing, evaluating, and implementing new technologies. In this perspective, we describe the role of in supporting equity. Materials Methods We apply National Institute Minority Health Disparities (NIMHD) research...
Medical licensing examinations, such as the United States Licensing Examination, have become default benchmarks for evaluating large language models (LLMs) in health care. Performance on these is frequently cited evidence of progress and used to justify deployment LLMs into clinical settings. However, we argue that are fundamentally limited signals assessing true utility.
Abstract Understanding reasons for treatment switching is of significant medical interest, but these factors are often only found in unstructured clinical notes and can be difficult to extract. We evaluated the zero-shot abilities GPT-4 eight other open-source large language models (LLMs) extract contraceptive information from 1964 derived UCSF Information Commons dataset. extracted contraceptives started stopped at each switch with microF1 scores 0.85 0.88, respectively, compared 0.81 0.88...
Deep learning methods for graphs achieve remarkable performance on many node-level and graph-level prediction tasks. However, despite the proliferation of their success, prevailing Graph Neural Networks (GNNs) neglect subgraphs, rendering subgraph tasks challenging to tackle in impactful applications. Further, present several unique challenges: subgraphs can have non-trivial internal topology, but also carry a notion position external connectivity information relative underlying graph which...
Griffin Adams, Emily Alsentzer, Mert Ketenci, Jason Zucker, Noémie Elhadad. Proceedings of the 2021 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2021.
The evaluation and management of first-time seizure-like events in children can be difficult because these episodes are not always directly observed might epileptic seizures or other conditions (seizure mimics). We aimed to evaluate whether machine learning models using real-world data could predict seizure recurrence after an initial event. This retrospective cohort study compared trained evaluated on two separate datasets between Jan 1, 2010, 2020: electronic medical records (EMRs) at...
Information overload in electronic health records (EHRs) hampers clinicians' ability to efficiently extract and synthesize critical information from a patient's longitudinal record, leading increased cognitive burden delays care. This study explores the potential of large language models (LLMs) address this challenge by generating problem-based admission summaries for patients admitted with heart failure, cause hospitalization worldwide. We developed an extract-then-abstract approach guided...
Patient summarization is essential for clinicians to provide coordinated care and practice effective communication. Automated has the potential save time, standardize notes, aid clinical decision making, reduce medical errors. Here we an upper bound on extractive of discharge notes develop LSTM model sequentially label topics history present illness notes. We achieve F1 score 0.876, which indicates that this can be employed create a dataset evaluation methods.
Prescription contraceptives play a critical role in supporting women's reproductive health. With nearly 50 million women the United States using contraceptives, understanding factors that drive selection and switching is of significant interest. However, many related to medication are often only captured unstructured clinical notes can be difficult extract. Here, we evaluate zero-shot abilities recently developed large language model, GPT-4 (via HIPAA-compliant Microsoft Azure API), identify...
Abstract Rare Mendelian disorders pose a major diagnostic challenge and collectively affect 300–400 million patients worldwide. Many automated tools aim to uncover causal genes in with suspected genetic disorders, but evaluation of these is limited due the lack comprehensive benchmark datasets that include previously unpublished conditions. Here, we present computational pipeline simulates realistic clinical address this deficit. Our framework jointly complex phenotypes challenging candidate...
Abstract There are more than 7,000 rare diseases, some affecting 3,500 or fewer patients in the US. Due to clinicians’ limited experience with such diseases and heterogeneity of clinical presentations, approximately 70% individuals seeking a diagnosis today remain undiagnosed. Deep learning has demonstrated success aiding common diseases. However, existing approaches require labeled datasets thousands diagnosed per disease. Here, we present SHEPHERD, few shot approach for multi-faceted...
Many areas of medicine would benefit from deeper, more accurate phenotyping, but there are limited approaches for phenotyping using clinical notes without substantial annotated data. Large language models (LLMs) have demonstrated immense potential to adapt novel tasks with no additional training by specifying task-specific i nstructions. We investigated the per-formance a publicly available LLM, Flan-T5, in patients postpartum hemorrhage (PPH) discharge electronic health records ( n...