AMIA Joint Summits on Translational Science proceedings. AMIA Joint Summits on Translational Science最新文献

The publication and sharing of clinical notes are crucial for healthcare research and innovation. However, privacy regulations such as HIPAA and GDPR pose significant challenges. While de-identification techniques aim to remove protected health information, they often fall short of achieving complete privacy protection. Similarly, the current state of synthetic clinical note generation can lack nuance and content coverage. To address these limitations, we propose an approach that combines de-identification, filtration, and synthetic clinical note generation. Variations of this approach currently retain 36%-61% of the original note's content and fill the remaining gaps using an LLM, ensuring high information coverage. We also evaluated the de-identification performance of the hybrid notes, demonstrating that they surpass or at least match the standalone de-identification methods. Our results show that hybrid notes can maintain patient privacy while preserving the richness of clinical data. This approach offers a promising solution for safe and effective data sharing, encouraging further research.

Clinical decision support systems (CDSS) are routinely employed in clinical settings to improve quality of care, ensure patient safety, and deliver consistent medical care. However, rule-based CDSS, currently available, do not feature reusable rules. In this study, we present CDSS with reusable rules. Our solution includes a common CDSS module, electronic medical record (EMR) specific adapters, CDSS rules written in the clinical quality language (CQL) (derived from CDC immunization recommendations), and patient records in fast healthcare interoperability resources (FHIR) format. The proposed CDSS is entirely browser-based and reachable within the user's EMR interface at the client-side. This helps to avoid the transmission ofpatient data and privacy breaches. Additionally, we propose to provide means of managing and maintaining CDSS rules to allow the end users to modify them independently. Successful implementation and deployment were achieved in OpenMRS and OpenEMR during initial testing.

We developed a simulated process to show a software implementation to facilitate an approach to integrate the Informed Consent Ontology, a reference ontology of informed consent information, to express implicit description and implement conceptual permission from informed consent life cycle. An early study introduced an experimental method to use Semantic Web Rule Language (SWRL) to describe and represent permissions to computational deduce more information from the Informed Consent Ontology (ICO), demonstrated by the use of the All of Us informed consent documents. We show how incomplete information in informed consent documents can be elucidated using a computational model of permissions toward health information technology that integrates ontologies. Future goals entail applying our computational approach for specific sub-domains of the informed consent life cycle, specifically for vaccine informed consent.

Oral and oropharyngeal cancers disproportionately affect Black Americans, contributing to significant healthcare disparities due to late-stage diagnoses and limited awareness. AI-powered chatbots have the potential to address these challenges by offering scalable, interactive, and personalized educational tools. This study evaluated the usability and accuracy of a Large Language Model-powered chatbot prototype under a Retrieval Augmented Generation framework designed to enhance oral cancer awareness, using a mixed-methods approach with six technical and clinical experts. Usability and accuracy were rated positively by 83.3% of the experts, with median scores of 6.65 and 7.67, respectively. Key areas for improvement included providing a clear introduction, simplifying the interface, addressing accessibility issues, and incorporating features like next-question suggestions, downloadable chats, and reference links. While content accuracy was well-received, gaps in conversational flow and technical term definitions were noted. These findings highlight the chatbot's potential to improve health literacy and reduce disparities.

Sharing protected health information (PHI) is critical for furthering biomedical research. Before data can be distributed, practitioners often perform deidentification to remove any PHI contained in the text. Contemporary deidentification methods are evaluated on highly saturated datasets (tools achieve near-perfect accuracy) which may not reflect the full variability or complexity of real-world clinical text and annotating them is resource intensive, which is a barrier to real-world applications. To address this gap, we developed an adversarial approach using a large language model (LLM) to re-identify the patient corresponding to a redacted clinical note and evaluated the performance with a novel De-Identification/Re-Identification (DIRI) method. Our method uses a large language model to reidentify the patient corresponding to a redacted clinical note. We demonstrate our method on medical data from Weill Cornell Medicine anonymized with three deidentification tools: rule-based Philter and two deep-learning-based models, BiLSTM-CRF and ClinicalBERT. Although ClinicalBERT was the most effective, masking all identified PII, our tool still reidentified 9% of clinical notes Our study highlights significant weaknesses in current deidentification technologies while providing a tool for iterative development and improvement.

Natural product-drug interactions (NPDIs) occurring due to concomitant exposure to botanical products and prescription drug therapies could lead to adverse events or reduced treatment efficacy. To better understand and address potential safety concerns, researchers investigate the underlying NPDI mechanisms using in vitro and clinical studies. Given that natural products are complex mixtures of compounds that are often not well characterized, it is important to advance computational methods for novel NPDI research. Biomedical knowledge graphs (KGs) can aid in identifying potential mechanisms to support such research efforts. We evaluated the ability of several KG embedding methods to improve NPDI prediction on NP-KG, a large-scale, heterogeneous, biomedical KG. We found that the ComplEx model outperformed other KG embedding approaches in both intrinsic and extrinsic evaluations. Future work will focus on utilizing the embeddings to identify underlying mechanisms of novel, potential NPDIs.

Developing and enforcing study protocols is crucial in medical research, especially as interactions with participants become more intricate. Traditional rules-based systems struggle to provide the automation and flexibility required for real-time, personalized data collection. We introduce SmartState, a state-based system designed to act as a personal agent for each participant, continuously managing and tracking their unique interactions. Unlike traditional reporting systems, SmartState enables real-time, automated data collection with minimal oversight. By integrating large language models to distill conversations into structured data, SmartState reduces errors and safeguards data integrity through built-in protocol and participant auditing. We demonstrate its utility in research trials involving time-dependent participant interactions, addressing the increasing need for reliable automation in complex clinical studies.

IgA Vasculitis (IgAV) is an immune-mediated condition with limited data on its long-term prognosis in adults. This study utilized the federated TriNetX network to evaluate the incidence of cardiac, renal, and vascular complications in adults diagnosed with IgAV. After propensity score matching, 12,506 patients with IgAV and 12,506 controls were analyzed. Over 6-month, 1-year, and 3-year periods, patients with IgAV had significantly higher risks of myocardial infarction, atrial fibrillation, stroke, pulmonary embolism, venous thromboembolism, chronic kidney disease, and end-stage renal disease compared to controls. Chronic kidney disease was the most common complication, with a 3-year risk of 8.6% and the highest absolute and relative risk difference. These findings suggest that adults with IgAV may be at increased risk for serious renal and cardiovascular complications, underscoring the need for further study to determine best practices for long term monitoring and management to mitigate morbidity associated with the disease.

There is an increasing demand for automatic mapping to standardized codes such as LOINC codes to create integrated medical databases across multiple facilities. However, natural language processing (NLP) in Japanese presents greater challenges than in English owing to a limited Japanese corpus for medical terms, such as test names. To address this limitation, we developed a machine learning-based method that maps in-house codes to LOINC codes by leveraging test result values without relying on test names that would require NLP. Using this approach, we achieved high mapping accuracy (70% or higher) for 80.4% of the analytes targeted in this study. The proposed method facilitates easier mapping to standardized codes in languages where NLP is challenging, ensuring accurate mapping to LOINC codes regardless of the source data language.

Safety event reporting forms a cornerstone of identifying and mitigating risks to patient and staff safety. However, variabilities in reporting and limited resources to analyze and classify event reports delay healthcare organizations' ability to rapidly identify safety event trends and to improve workplace safety. We demonstrated how large language models can classify safety event report narratives as workplace violence (F1: 0.80 for physical violence; F1: 0.94 for verbal abuse) and communication failures (F1: 0.94) as a first step toward enabling automated labeling of safety event reports and ultimately improving workplace safety.