AMIA ... Annual Symposium proceedings. AMIA Symposium最新文献

The increase in utilization of patient portal messages has imposed a considerable burden on healthcare providers, contributing to an increased incidence of provider burnout. This study introduces a framework for leveraging Large Language Models (LLMs) and Chain-of-Thought (CoT) prompting in order to automatically categorize and route messages to their appropriate location. The modeling framework, which utilizes gold standard annotations from triage nurses, not only facilitates the dynamic adaptation of the model to evolving healthcare workflows and emerging edge-case scenarios, but also significantly improves the model's classification accuracy compared to traditional zero-shot methods. In addition, the framework allows for flexibility in its task and continuous improvement via annotation of exemplar messages. The model is able to accurately categorize messages in an automated fashion, which has potential to dramatically ease the burden on providers and provide faster and safer responses to patients. This framework can also be readily extended to work in a variety of clinical and documentation settings.

Machine learning (ML) algorithms play a crucial role in the early and accurate diagnosis of Alzheimer's Disease (AD), which is essential for effective treatment planning. However, existing methods are not well-suited for identifying Mild Cognitive Impairment (MCI), a critical transitional stage between normal aging and AD. This inadequacy is primarily due to label imbalance and bias from different sensitve attributes in MCI classification. To overcome these challenges, we have designed an end-to-end fairness-aware approach for label-imbalanced classification, tailored specifically for neuroimaging data. This method, built on the recently developed FACIMS framework, integrates into STREAMLINE, an automated ML environment. We evaluated our approach against nine other ML algorithms and found that it achieves comparable balanced accuracy to other methods while prioritizing fairness in classifications with five different sensitive attributes. This analysis contributes to the development of equitable and reliable ML diagnostics for MCI detection.

Systemic lupus erythematosus (SLE) is a complex heterogeneous disease with many manifestational facets. We propose a data-driven approach to discover probabilistic independent sources from multimodal imperfect EHR data. These sources represent exogenous variables in the data generation process causal graph that estimate latent root causes of the presence of SLE in the health record. We objectively evaluated the sources against the original variables from which they were discovered by training supervised models to discriminate SLE from negative health records using a reduced set of labelled instances. We found 19 predictive sources with high clinical validity and whose EHR signatures define independent factors of SLE heterogeneity. Using the sources as input patient data representation enables models to provide with rich explanations that better capture the clinical reasons why a particular record is (not) an SLE case. Providers may be willing to trade patient-level interpretability for discrimination especially in challenging cases.

Young scientists, including postdocs and assistant professors, need access to grant writing resources for training and proposal development. To assist in this, we developed a web-based research guide providing centralized access to curated tools throughout the research funding process- finding funding, preparing proposals, managing awards, etc. Using consumer informatics principles, we enhanced the research grant repository's effectiveness, with lessons learned and insights generalizable to other institutions. Six faculty members completed nine tasks to explore the guide's ten sections. Participants found the guide highly usable, with an excellent System Usability Scale (SUS) score of 89.2. Suggestions included improving navigation, content organization and providing education on award management processes. Liked features were the chronological organization of information, samples from successful grants, pre-populated templates, and mechanisms for ongoing feedback. These findings underscore the importance of usability in developing resources that effectively support faculty in grant writing and proposal development.

Gene expression profiles obtained through DNA microarray have proven successful in providing critical information for cancer detection classifiers. However, the limited number of samples in these datasets poses a challenge to employ complex methodologies such as deep neural networks for sophisticated analysis. To address this "small data" dilemma, Meta-Learning has been introduced as a solution to enhance the optimization of machine learning models by utilizing similar datasets, thereby facilitating a quicker adaptation to target datasets without the requirement of sufficient samples. In this study, we present a meta-learning-based approach for predicting lung cancer from gene expression profiles. We apply this framework to well-established deep learning methodologies and employ four distinct datasets for the meta-learning tasks, where one as the target dataset and the rest as source datasets. Our approach is evaluated against both traditional and deep learning methodologies, and the results show the superior performance of meta-learning on augmented source data compared to the baselines trained on single datasets. Moreover, we conduct the comparative analysis between meta-learning and transfer learning methodologies to highlight the efficiency of the proposed approach in addressing the challenges associated with limited sample sizes. Finally, we incorporate the explainability study to illustrate the distinctiveness of decisions made by meta-learning.

Blood pressure variability (BPV) plays a critical role in vascular diseases, particularly in acute ischemic stroke patients in intensive care units (ICUs), where higher BPV correlates with increased mortality rates. Current interventions lack effective methods for controlling BPV across consecutive time windows. To addressing this gap, we propose an offline deep reinforcement learning approach with supervised guidance to regulate systolic BPV in the following consecutive time windows by optimizing intravenous nicardipine infusion rates for intracerebral hemorrhage patients. Using clinically inspired reward functions, our method aims to tailor antihypertensive medication management within the critical 24-hour recovery window. Compared to human performance, our best method showed 57.52% and 126.01% improvements over the human baseline for maintaining BP within the desired range for the next time window and across two consecutive time windows. This research promises streamlined antihypertensive medication dosing, offering potential just-in-time adaptive interventions through automated pumps during stroke patients' ICU stays.

Nowadays, healthcare systems increasingly utilize automated surveillance of electronic medical record (EMR) data to detect adverse events with specific patterns. Despite these technological advances, the early identification of adverse events remains challenging due to the absence of clearly defined prodromal sequences that could signal the onset of such events. Achieving clinically meaningful and interpretable prediction outcomes necessitates a framework that is capable of (i) deducing temporal relationships among various time series features within EMR data (e.g., laboratory test results, vital signs), and (ii) identifying specific patterns that herald the occurrence of an adverse event (e.g., acute kidney injury (AKI)). This study employs a time series forecasting approach to undertake neural Granger causal analysis, and further enhance it by integrating a personalized PageRank algorithm to analyze the critical causal derangements among ICU-acquired AKIpatients. An experimental analysis based on the proposed methodology was conducted using a dataset from MIMIC-IV.Finally, a Granger causality (GC) graph, which revealed several interpretable GC chains that could be used to predict the occurrence ofAKI in ICU settings, was generated. The GC graph and GC chains identified in this study have the potential to aid ICU physicians in providing timely interventions and may help improve patient outcomes.

Recent advancements in Large Language Models (LLMs) have ushered in a new era for knowledge extraction in the domains of biological and clinical natural language processing (NLP). In this research, we present a novel approach to understanding the regulatory effects of genes and medications on biological processes central to wound healing. Utilizing the capabilities of Generative Pre-trained Transformer (GPT) models by OpenAI, specifically GPT-3.5 and GPT-4, we developed a comprehensive pipeline for the identification and grounding of biological processes and the extraction of such regulatory relations. The performances of both GPTs were rigorously evaluated against a manually annotated corpus of 104 PubMed titles, focusing on their ability to accurately identify and ground biological process concepts and extract relevant regulatory relationships from the text. Our findings demonstrate that GPT-4, in particular, exhibits superior performance in all the tasks, showcasing its potential to facilitate significant advancements in biomedical research without requiring model fine-tuning.

Physical activity is crucial for children's healthy growth and development. In the US, most states have physical education standards. California implemented the mandated School-based Physical Fitness Testing (SB-PFT) over two decades ago. Despite the substantial effort in collecting the SB-PFT data, its research reuse has been limited due to the lack of readily accessible analytical tools. We developed a web application utilizing GeoServer, ArcGIS, and AWS to visualize the SB-PFT data. Education administrators and policymakers can leverage this user-friendly platform to gain insights into children's physical fitness trend, and identify schools and districts with successful programs to gauge the success of new physical education programs. The application also includes a custom mapping tool that allows users to compare external datasets with SB-PFT. We conclude that by incorporating advanced analytical capabilities through an informatics-based user-facing tool, this platform has great potential to encourage a broader engagement in enhancing children's physical fitness.

Symptoms, or subjective experiences of patients which can indicate underlying pathology, are important for guiding clinician decision-making and revealing patient wellbeing. However, they are difficult to study because information is primarily found in clinical free text, not in structured electronic health record fields. This study finds that large language models (LLMs) can extract several common symptom concepts from clinical narratives, using an approach of including clarifying information in the prompt, few-shot examples, and chain-of-thought-prompting. This approach is compared to symptom-specific machine learning classifiers based on clinical concepts mapped from free text. For most symptom concepts, the LLM performs better and achieves a higher F1-score, likely by leveraging context important for the symptom normalization task. Unlocking information about symptom concepts from clinical narratives has potential to improve healthcare workflows and facilitate a broad range of research agendas.