Meliha Yetisgen

Wen-wai Yim, Yujuan Fu, Asma Ben Abacha et al. · uw

Recent immense breakthroughs in generative models such as in GPT4 have precipitated re-imagined ubiquitous usage of these models in all applications. One area that can benefit by improvements in artificial intelligence (AI) is healthcare. The note generation task from doctor-patient encounters, and its associated electronic medical record documentation, is one of the most arduous time-consuming tasks for physicians. It is also a natural prime potential beneficiary to advances in generative models. However with such advances, benchmarking is more critical than ever. Whether studying model weaknesses or developing new evaluation metrics, shared open datasets are an imperative part of understanding the current state-of-the-art. Unfortunately as clinic encounter conversations are not routinely recorded and are difficult to ethically share due to patient confidentiality, there are no sufficiently large clinic dialogue-note datasets to benchmark this task. Here we present the Ambient Clinical Intelligence Benchmark (ACI-BENCH) corpus, the largest dataset to date tackling the problem of AI-assisted note generation from visit dialogue. We also present the benchmark performances of several common state-of-the-art approaches.

Sihang Zeng, Matthew Thompson, Ruth Etzioni et al.

Modeling patient trajectories from longitudinal electronic health records (EHRs) requires reasoning over sparse, noisy, and long-context multimodal sequences. Existing LLM-based multi-agent systems address context length but process patients in isolation, failing to mirror how clinicians leverage accumulated experience from similar prior cases. We present Traj-Evolve, a self-evolving multi-agent system with two complementary evolving mechanisms. First, an Experience Pool (ExPool) acts as a non-parametric memory, indexing rejection-sampled reasoning traces to retrieve similar patients as few-shot contexts. Second, multi-agent reinforcement learning (MARL) via reward-ranked fine-tuning parametrically optimizes inter-agent and agent-memory collaboration. A leave-one-out cross-retrieval strategy unifies the two, aligning training- and inference-time behavior under retrieval augmentation. On a lung cancer prediction task utilizing up to five years of multimodal EHRs, Traj-Evolve outperforms 9 strong baselines on the overall population and a challenging never-smoker population. Analysis of the evolving dynamics highlights three key findings: (1) expanding the ExPool shifts optimal retrieval from diverse to specific samples; (2) under MARL, the manager agent's prediction loss converges quickly while the worker agents' temporal reasoning continues to benefit from more verified patients; and (3) the two mechanisms are complementary on the predicted risk, where ExPool improves specificity while MARL improves sensitivity.

Kevin Lybarger, Meliha Yetisgen, Özlem Uzuner

Objective: The n2c2/UW SDOH Challenge explores the extraction of social determinant of health (SDOH) information from clinical notes. The objectives include the advancement of natural language processing (NLP) information extraction techniques for SDOH and clinical information more broadly. This paper presents the shared task, data, participating teams, performance results, and considerations for future work. Materials and Methods: The task used the Social History Annotated Corpus (SHAC), which consists of clinical text with detailed event-based annotations for SDOH events such as alcohol, drug, tobacco, employment, and living situation. Each SDOH event is characterized through attributes related to status, extent, and temporality. The task includes three subtasks related to information extraction (Subtask A), generalizability (Subtask B), and learning transfer (Subtask C). In addressing this task, participants utilized a range of techniques, including rules, knowledge bases, n-grams, word embeddings, and pretrained language models (LM). Results: A total of 15 teams participated, and the top teams utilized pretrained deep learning LM. The top team across all subtasks used a sequence-to-sequence approach achieving 0.901 F1 for Subtask A, 0.774 F1 Subtask B, and 0.889 F1 for Subtask C. Conclusions: Similar to many NLP tasks and domains, pretrained LM yielded the best performance, including generalizability and learning transfer. An error analysis indicates extraction performance varies by SDOH, with lower performance achieved for conditions, like substance use and homelessness, that increase health risks (risk factors) and higher performance achieved for conditions, like substance abstinence and living with family, that reduce health risks (protective factors).

Nicholas J Dobbins, Bin Han, Weipeng Zhou et al.

Objective: Identifying study-eligible patients within clinical databases is a critical step in clinical research. However, accurate query design typically requires extensive technical and biomedical expertise. We sought to create a system capable of generating data model-agnostic queries while also providing novel logical reasoning capabilities for complex clinical trial eligibility criteria. Materials and Methods: The task of query creation from eligibility criteria requires solving several text-processing problems, including named entity recognition and relation extraction, sequence-to-sequence transformation, normalization, and reasoning. We incorporated hybrid deep learning and rule-based modules for these, as well as a knowledge base of the Unified Medical Language System (UMLS) and linked ontologies. To enable data-model agnostic query creation, we introduce a novel method for tagging database schema elements using UMLS concepts. To evaluate our system, called LeafAI, we compared the capability of LeafAI to a human database programmer to identify patients who had been enrolled in 8 clinical trials conducted at our institution. We measured performance by the number of actual enrolled patients matched by generated queries. Results: LeafAI matched a mean 43% of enrolled patients with 27,225 eligible across 8 clinical trials, compared to 27% matched and 14,587 eligible in queries by a human database programmer. The human programmer spent 26 total hours crafting queries compared to several minutes by LeafAI. Conclusions: Our work contributes a state-of-the-art data model-agnostic query generation system capable of conditional reasoning using a knowledge base. We demonstrate that LeafAI can rival an experienced human programmer in finding patients eligible for clinical trials.

Giridhar Kaushik Ramachandran, Yujuan Fu, Bin Han et al. · uw

Social determinants of health (SDOH) documented in the electronic health record through unstructured text are increasingly being studied to understand how SDOH impacts patient health outcomes. In this work, we utilize the Social History Annotation Corpus (SHAC), a multi-institutional corpus of de-identified social history sections annotated for SDOH, including substance use, employment, and living status information. We explore the automatic extraction of SDOH information with SHAC in both standoff and inline annotation formats using GPT-4 in a one-shot prompting setting. We compare GPT-4 extraction performance with a high-performing supervised approach and perform thorough error analyses. Our prompt-based GPT-4 method achieved an overall 0.652 F1 on the SHAC test set, similar to the 7th best-performing system among all teams in the n2c2 challenge with SHAC.

Nicholas J Dobbins, Tony Mullen, Ozlem Uzuner et al.

Identifying cohorts of patients based on eligibility criteria such as medical conditions, procedures, and medication use is critical to recruitment for clinical trials. Such criteria are often most naturally described in free-text, using language familiar to clinicians and researchers. In order to identify potential participants at scale, these criteria must first be translated into queries on clinical databases, which can be labor-intensive and error-prone. Natural language processing (NLP) methods offer a potential means of such conversion into database queries automatically. However they must first be trained and evaluated using corpora which capture clinical trials criteria in sufficient detail. In this paper, we introduce the Leaf Clinical Trials (LCT) corpus, a human-annotated corpus of over 1,000 clinical trial eligibility criteria descriptions using highly granular structured labels capturing a range of biomedical phenomena. We provide details of our schema, annotation process, corpus quality, and statistics. Additionally, we present baseline information extraction results on this corpus as benchmarks for future work.

Giridhar Kaushik Ramachandran, Kevin Lybarger, Yaya Liu et al.

An accurate and detailed account of patient medications, including medication changes within the patient timeline, is essential for healthcare providers to provide appropriate patient care. Healthcare providers or the patients themselves may initiate changes to patient medication. Medication changes take many forms, including prescribed medication and associated dosage modification. These changes provide information about the overall health of the patient and the rationale that led to the current care. Future care can then build on the resulting state of the patient. This work explores the automatic extraction of medication change information from free-text clinical notes. The Contextual Medication Event Dataset (CMED) is a corpus of clinical notes with annotations that characterize medication changes through multiple change-related attributes, including the type of change (start, stop, increase, etc.), initiator of the change, temporality, change likelihood, and negation. Using CMED, we identify medication mentions in clinical text and propose three novel high-performing BERT-based systems that resolve the annotated medication change characteristics. We demonstrate that our proposed systems improve medication change classification performance over the initial work exploring CMED.

Kevin Lybarger, Nicholas J Dobbins, Ritche Long et al.

Objective: Social determinants of health (SDOH) impact health outcomes and are documented in the electronic health record (EHR) through structured data and unstructured clinical notes. However, clinical notes often contain more comprehensive SDOH information, detailing aspects such as status, severity, and temporality. This work has two primary objectives: i) develop a natural language processing (NLP) information extraction model to capture detailed SDOH information and ii) evaluate the information gain achieved by applying the SDOH extractor to clinical narratives and combining the extracted representations with existing structured data. Materials and Methods: We developed a novel SDOH extractor using a deep learning entity and relation extraction architecture to characterize SDOH across various dimensions. In an EHR case study, we applied the SDOH extractor to a large clinical data set with 225,089 patients and 430,406 notes with social history sections and compared the extracted SDOH information with existing structured data. Results: The SDOH extractor achieved 0.86 F1 on a withheld test set. In the EHR case study, we found extracted SDOH information complements existing structured data with 32% of homeless patients, 19% of current tobacco users, and 10% of drug users only having these health risk factors documented in the clinical narrative. Conclusions: Utilizing EHR data to identify SDOH health risk factors and social needs may improve patient care and outcomes. Semantic representations of text-encoded SDOH information can augment existing structured data, and this more comprehensive SDOH representation can assist health systems in identifying and addressing these social needs.

Sitong Zhou, Meliha Yetisgen, Mari Ostendorf

This paper explores methods for extracting information from radiology reports that generalize across exam modalities to reduce requirements for annotated data. We demonstrate that multi-pass T5-based text-to-text generative models exhibit better generalization across exam modalities compared to approaches that employ BERT-based task-specific classification layers. We then develop methods that reduce the inference cost of the model, making large-scale corpus processing more feasible for clinical applications. Specifically, we introduce a generative technique that decomposes complex tasks into smaller subtask blocks, which improves a single-pass model when combined with multitask training. In addition, we leverage target-domain contexts during inference to enhance domain adaptation, enabling use of smaller models. Analyses offer insights into the benefits of different cost reduction strategies.

Sitong Zhou, Kevin Lybarger, Meliha Yetisgen et al.

Symptom information is primarily documented in free-text clinical notes and is not directly accessible for downstream applications. To address this challenge, information extraction approaches that can handle clinical language variation across different institutions and specialties are needed. In this paper, we present domain generalization for symptom extraction using pretraining and fine-tuning data that differs from the target domain in terms of institution and/or specialty and patient population. We extract symptom events using a transformer-based joint entity and relation extraction method. To reduce reliance on domain-specific features, we propose a domain generalization method that dynamically masks frequent symptoms words in the source domain. Additionally, we pretrain the transformer language model (LM) on task-related unlabeled texts for better representation. Our experiments indicate that masking and adaptive pretraining methods can significantly improve performance when the source domain is more distant from the target domain.

Namu Park, Giridhar Kaushik Ramachandran, Kevin Lybarger et al.

Large language models (LLMs) have shown considerable promise in clinical natural language processing, yet few domain-specific datasets exist to rigorously evaluate their performance on radiology tasks. In this work, we introduce an annotated corpus of 6,393 radiology reports from 586 patients, each labeled for follow-up imaging status, to support the development and benchmarking of follow-up adherence detection systems. Using this corpus, we systematically compared traditional machine-learning classifiers, including logistic regression (LR), support vector machines (SVM), Longformer, and a fully fine-tuned Llama3-8B-Instruct, with recent generative LLMs. To evaluate generative LLMs, we tested GPT-4o and the open-source GPT-OSS-20B under two configurations: a baseline (Base) and a task-optimized (Advanced) setting that focused inputs on metadata, recommendation sentences, and their surrounding context. A refined prompt for GPT-OSS-20B further improved reasoning accuracy. Performance was assessed using precision, recall, and F1 scores with 95% confidence intervals estimated via non-parametric bootstrapping. Inter-annotator agreement was high (F1 = 0.846). GPT-4o (Advanced) achieved the best performance (F1 = 0.832), followed closely by GPT-OSS-20B (Advanced; F1 = 0.828). LR and SVM also performed strongly (F1 = 0.776 and 0.775), underscoring that while LLMs approach human-level agreement through prompt optimization, interpretable and resource-efficient models remain valuable baselines.

Feng Chen, Justin Tauscher, Changye Li et al.

Speech monologues recorded in naturalistic settings provide opportunities to characterize mental illness phenomenology and detect symptom exacerbation. Large language models (LLMs) offer new possibilities for automating this process, as they require annotated data primarily for evaluation rather than training. In this paper, we present a novel automated, multi-agent LLM pipeline for the fine-grained, multi-label extraction of language suggestive of delusional beliefs, associated affective responses, and behavioral responses from transcripts of naturalistic audio diaries collected from people with moderate persecutory ideation. Evaluating an ensemble of three foundation models, we demonstrate that detailed diagnostic prompt instructions successfully reduce false positives for delusional theme classification, but also constrain the interpretation of affective or behavioral responses. Furthermore, comparing multi-agent adjudication frameworks shows that complex conversational debate between agents diminishes accuracy on clinically ambiguous text by inducing premature consensus. Instead, majority voting establishes robust performance (Micro F1 of 0.872 and 0.779 for delusion detection and classification respectively). This work provides a validated and scalable pipeline for the automated detection and characterization of content suggesting delusional beliefs in naturalistic speech.

Asma Ben Abacha, Wen-wai Yim, Yujuan Fu et al.

Several studies showed that Large Language Models (LLMs) can answer medical questions correctly, even outperforming the average human score in some medical exams. However, to our knowledge, no study has been conducted to assess the ability of language models to validate existing or generated medical text for correctness and consistency. In this paper, we introduce MEDEC (https://github.com/abachaa/MEDEC), the first publicly available benchmark for medical error detection and correction in clinical notes, covering five types of errors (Diagnosis, Management, Treatment, Pharmacotherapy, and Causal Organism). MEDEC consists of 3,848 clinical texts, including 488 clinical notes from three US hospital systems that were not previously seen by any LLM. The dataset has been used for the MEDIQA-CORR shared task to evaluate seventeen participating systems [Ben Abacha et al., 2024]. In this paper, we describe the data creation methods and we evaluate recent LLMs (e.g., o1-preview, GPT-4, Claude 3.5 Sonnet, and Gemini 2.0 Flash) for the tasks of detecting and correcting medical errors requiring both medical knowledge and reasoning capabilities. We also conducted a comparative study where two medical doctors performed the same task on the MEDEC test set. The results showed that MEDEC is a sufficiently challenging benchmark to assess the ability of models to validate existing or generated notes and to correct medical errors. We also found that although recent LLMs have a good performance in error detection and correction, they are still outperformed by medical doctors in these tasks. We discuss the potential factors behind this gap, the insights from our experiments, the limitations of current evaluation metrics, and share potential pointers for future research.

Tianmai M. Zhang, Zhaoyi Sun, Sihang Zeng et al.

The ChemoTimelines shared task benchmarks methods for constructing timelines of systemic anticancer treatment from electronic health records of cancer patients. This paper describes our methods, results, and findings for subtask 2 -- generating patient chemotherapy timelines from raw clinical notes. We evaluated strategies involving chain-of-thought thinking, supervised fine-tuning, direct preference optimization, and dictionary-based lookup to improve timeline extraction. All of our approaches followed a two-step workflow, wherein an LLM first extracted chemotherapy events from individual clinical notes, and then an algorithm normalized and aggregated events into patient-level timelines. Each specific method differed in how the associated LLM was utilized and trained. Multiple approaches yielded competitive performances on the test set leaderboard, with fine-tuned Qwen3-14B achieving the best official score of 0.678. Our results and analyses could provide useful insights for future attempts on this task as well as the design of similar tasks.

Wen-wai Yim, Yujuan Fu, Asma Ben Abacha et al.

Recent advances in dermatological image analysis have been driven by large-scale annotated datasets; however, most existing benchmarks focus on dermatoscopic images and lack patient-authored queries and clinical context, limiting their applicability to patient-centered care. To address this gap, we introduce DermaVQA-DAS, an extension of the DermaVQA dataset that supports two complementary tasks: closed-ended question answering (QA) and dermatological lesion segmentation. Central to this work is the Dermatology Assessment Schema (DAS), a novel expert-developed framework that systematically captures clinically meaningful dermatological features in a structured and standardized form. DAS comprises 36 high-level and 27 fine-grained assessment questions, with multiple-choice options in English and Chinese. Leveraging DAS, we provide expert-annotated datasets for both closed QA and segmentation and benchmark state-of-the-art multimodal models. For segmentation, we evaluate multiple prompting strategies and show that prompt design impacts performance: the default prompt achieves the best results under Mean-of-Max and Mean-of-Mean evaluation aggregation schemes, while an augmented prompt incorporating both patient query title and content yields the highest performance under majority-vote-based microscore evaluation, achieving a Jaccard index of 0.395 and a Dice score of 0.566 with BiomedParse. For closed-ended QA, overall performance is strong across models, with average accuracies ranging from 0.729 to 0.798; o3 achieves the best overall accuracy (0.798), closely followed by GPT-4.1 (0.796), while Gemini-1.5-Pro shows competitive performance within the Gemini family (0.783). We publicly release DermaVQA-DAS, the DAS schema, and evaluation protocols to support and accelerate future research in patient-centered dermatological vision-language modeling (https://osf.io/72rp3).

Yujuan Fu, Giridhar Kaushik Ramachandran, Ahmad Halwani et al.

Clinical notes contain unstructured representations of patient histories, including the relationships between medical problems and prescription drugs. To investigate the relationship between cancer drugs and their associated symptom burden, we extract structured, semantic representations of medical problem and drug information from the clinical narratives of oncology notes. We present Clinical Concept Annotations for Cancer Events and Relations (CACER), a novel corpus with fine-grained annotations for over 48,000 medical problems and drug events and 10,000 drug-problem and problem-problem relations. Leveraging CACER, we develop and evaluate transformer-based information extraction (IE) models such as BERT, Flan-T5, Llama3, and GPT-4 using fine-tuning and in-context learning (ICL). In event extraction, the fine-tuned BERT and Llama3 models achieved the highest performance at 88.2-88.0 F1, which is comparable to the inter-annotator agreement (IAA) of 88.4 F1. In relation extraction, the fine-tuned BERT, Flan-T5, and Llama3 achieved the highest performance at 61.8-65.3 F1. GPT-4 with ICL achieved the worst performance across both tasks. The fine-tuned models significantly outperformed GPT-4 in ICL, highlighting the importance of annotated training data and model optimization. Furthermore, the BERT models performed similarly to Llama3. For our task, LLMs offer no performance advantage over the smaller BERT models. The results emphasize the need for annotated training data to optimize models. Multiple fine-tuned transformer models achieved performance comparable to IAA for several extraction tasks.

Sihang Zeng, Young Won Kim, Wilson Lau et al.

Accurate estimation of cancer risk from longitudinal electronic health records (EHRs) could support earlier detection and improved care, but modeling such complex patient trajectories remains challenging. We present TrajOnco, a training-free, multi-agent large language model (LLM) framework designed for scalable multi-cancer early detection. Using a chain-of-agents architecture with long-term memory, TrajOnco performs temporal reasoning over sequential clinical events to generate patient-level summaries, evidence-linked rationales, and predicted risk scores. We evaluated TrajOnco on de-identified Truveta EHR data across 15 cancer types using matched case-control cohorts, predicting risk of cancer diagnosis at 1 year. In zero-shot evaluation, TrajOnco achieved AUROCs of 0.64-0.80, performing comparably to supervised machine learning in a lung cancer benchmark while demonstrating better temporal reasoning than single-agent LLMs. The multi-agent design also enabled effective temporal reasoning with smaller-capacity models such as GPT-4.1-mini. The fidelity of TrajOnco's output was validated through human evaluation. Furthermore, TrajOnco's interpretable reasoning outputs can be aggregated to reveal population-level risk patterns that align with established clinical knowledge. These findings highlight the potential of multi-agent LLMs to execute interpretable temporal reasoning over longitudinal EHRs, advancing both scalable multi-cancer early detection and clinical insight generation.

Yujuan Velvin Fu, Giridhar Kaushik Ramachandran, Namu Park et al.

Large language models (LLMs) such as ChatGPT are fine-tuned on large and diverse instruction-following corpora, and can generalize to new tasks. However, those instruction-tuned LLMs often perform poorly in specialized medical natural language understanding (NLU) tasks that require domain knowledge, granular text comprehension, and structured data extraction. To bridge the gap, we: (1) propose a unified prompting format for 7 important NLU tasks, (2) curate an instruction-tuning dataset, MNLU-Instruct, utilizing diverse existing open-source medical NLU corpora, and (3) develop BioMistral-NLU, a generalizable medical NLU model, through fine-tuning BioMistral on MNLU-Instruct. We evaluate BioMistral-NLU in a zero-shot setting, across 6 important NLU tasks, from two widely adopted medical NLU benchmarks: BLUE and BLURB. Our experiments show that our BioMistral-NLU outperforms the original BioMistral, as well as the proprietary LLMs - ChatGPT and GPT-4. Our dataset-agnostic prompting strategy and instruction tuning step over diverse NLU tasks enhance LLMs' generalizability across diverse medical NLU tasks. Our ablation experiments show that instruction-tuning on a wider variety of tasks, even when the total number of training instances remains constant, enhances downstream zero-shot generalization.

Sitong Zhou, Meliha Yetisgen, Mari Ostendorf

Tracking findings in longitudinal radiology reports is crucial for accurately identifying disease progression, and the time-consuming process would benefit from automatic summarization. This work introduces a structured summarization task, where we frame longitudinal report summarization as a timeline generation task, with dated findings organized in columns and temporally related findings grouped in rows. This structured summarization format enables straightforward comparison of findings across time and facilitates fact-checking against the associated reports. The timeline is generated using a 3-step LLM process of extracting findings, generating group names, and using the names to group the findings. To evaluate such systems, we create RadTimeline, a timeline dataset focused on tracking lung-related radiologic findings in chest-related imaging reports. Experiments on RadTimeline show tradeoffs of different-sized LLMs and prompting strategies. Our results highlight that group name generation as an intermediate step is critical for effective finding grouping. The best configuration has some irrelevant findings but very good recall, and grouping performance is comparable to human annotators.

Haritha Gangavarapu, Giridhar Kaushik Ramachandran, Kevin Lybarger et al.

A vast amount of medical knowledge is available for public use through online health forums, and question-answering platforms on social media. The majority of the population in the United States doesn't have the right amount of health literacy to make the best use of that information. Health literacy means the ability to obtain and comprehend the basic health information to make appropriate health decisions. To build the bridge between this gap, organizations advocate adapting this medical knowledge into plain language. Building robust systems to automate the adaptations helps both medical and non-medical professionals best leverage the available information online. The goal of the Plain Language Adaptation of Biomedical Abstracts (PLABA) track is to adapt the biomedical abstracts in English language extracted from PubMed based on the questions asked in MedlinePlus for the general public using plain language at the sentence level. As part of this track, we leveraged the best open-source Large Language Models suitable and fine-tuned for dialog use cases. We compare and present the results for all of our systems and our ranking among the other participants' submissions. Our top performing GPT-4 based model ranked first in the avg. simplicity measure and 3rd on the avg. accuracy measure.

Philip Chung, Christine T Fong, Andrew M Walters et al.

We investigate whether general-domain large language models such as GPT-4 Turbo can perform risk stratification and predict post-operative outcome measures using a description of the procedure and a patient's clinical notes derived from the electronic health record. We examine predictive performance on 8 different tasks: prediction of ASA Physical Status Classification, hospital admission, ICU admission, unplanned admission, hospital mortality, PACU Phase 1 duration, hospital duration, and ICU duration. Few-shot and chain-of-thought prompting improves predictive performance for several of the tasks. We achieve F1 scores of 0.50 for ASA Physical Status Classification, 0.81 for ICU admission, and 0.86 for hospital mortality. Performance on duration prediction tasks were universally poor across all prompt strategies. Current generation large language models can assist clinicians in perioperative risk stratification on classification tasks and produce high-quality natural language summaries and explanations.

Yujuan Fu, Ozlem Uzuner, Meliha Yetisgen et al.

Large language models (LLMs) have demonstrated great performance across various benchmarks, showing potential as general-purpose task solvers. However, as LLMs are typically trained on vast amounts of data, a significant concern in their evaluation is data contamination, where overlap between training data and evaluation datasets inflates performance assessments. Multiple approaches have been developed to identify data contamination. These approaches rely on specific assumptions that may not hold universally across different settings. To bridge this gap, we systematically review 50 papers on data contamination detection, categorize the underlying assumptions, and assess whether they have been rigorously validated. We identify and analyze eight categories of assumptions and test three of them as case studies. Our case studies focus on detecting direct, instance-level data contamination, which is also referred to as Membership Inference Attacks (MIA). Our analysis reveals that MIA approaches based on these three assumptions can have similar performance to random guessing, on datasets used in LLM pretraining, suggesting that current LLMs might learn data distributions rather than memorizing individual instances. Meanwhile, MIA can easily fail when there are data distribution shifts between the seen and unseen instances.

Yujuan Fu, Giridhar Kaushik Ramachandran, Nicholas J Dobbins et al.

Social determinants of health (SDoH) play a critical role in shaping health outcomes, particularly in pediatric populations where interventions can have long-term implications. SDoH are frequently studied in the Electronic Health Record (EHR), which provides a rich repository for diverse patient data. In this work, we present a novel annotated corpus, the Pediatric Social History Annotation Corpus (PedSHAC), and evaluate the automatic extraction of detailed SDoH representations using fine-tuned and in-context learning methods with Large Language Models (LLMs). PedSHAC comprises annotated social history sections from 1,260 clinical notes obtained from pediatric patients within the University of Washington (UW) hospital system. Employing an event-based annotation scheme, PedSHAC captures ten distinct health determinants to encompass living and economic stability, prior trauma, education access, substance use history, and mental health with an overall annotator agreement of 81.9 F1. Our proposed fine-tuning LLM-based extractors achieve high performance at 78.4 F1 for event arguments. In-context learning approaches with GPT-4 demonstrate promise for reliable SDoH extraction with limited annotated examples, with extraction performance at 82.3 F1 for event triggers.

Zhaoyi Sun, Wen-Wai Yim, Ozlem Uzuner et al.

Objective: This review aims to explore the potential and challenges of using Natural Language Processing (NLP) to detect, correct, and mitigate medically inaccurate information, including errors, misinformation, and hallucination. By unifying these concepts, the review emphasizes their shared methodological foundations and their distinct implications for healthcare. Our goal is to advance patient safety, improve public health communication, and support the development of more reliable and transparent NLP applications in healthcare. Methods: A scoping review was conducted following PRISMA guidelines, analyzing studies from 2020 to 2024 across five databases. Studies were selected based on their use of NLP to address medically inaccurate information and were categorized by topic, tasks, document types, datasets, models, and evaluation metrics. Results: NLP has shown potential in addressing medically inaccurate information on the following tasks: (1) error detection (2) error correction (3) misinformation detection (4) misinformation correction (5) hallucination detection (6) hallucination mitigation. However, challenges remain with data privacy, context dependency, and evaluation standards. Conclusion: This review highlights the advancements in applying NLP to tackle medically inaccurate information while underscoring the need to address persistent challenges. Future efforts should focus on developing real-world datasets, refining contextual methods, and improving hallucination management to ensure reliable and transparent healthcare applications.

Namu Park, Kevin Lybarger, Giridhar Kaushik Ramachandran et al.

Medical imaging is critical to the diagnosis, surveillance, and treatment of many health conditions, including oncological, neurological, cardiovascular, and musculoskeletal disorders, among others. Radiologists interpret these complex, unstructured images and articulate their assessments through narrative reports that remain largely unstructured. This unstructured narrative must be converted into a structured semantic representation to facilitate secondary applications such as retrospective analyses or clinical decision support. Here, we introduce the Corpus of Annotated Medical Imaging Reports (CAMIR), which includes 609 annotated radiology reports from three imaging modality types: Computed Tomography, Magnetic Resonance Imaging, and Positron Emission Tomography-Computed Tomography. Reports were annotated using an event-based schema that captures clinical indications, lesions, and medical problems. Each event consists of a trigger and multiple arguments, and a majority of the argument types, including anatomy, normalize the spans to pre-defined concepts to facilitate secondary use. CAMIR uniquely combines a granular event structure and concept normalization. To extract CAMIR events, we explored two BERT (Bi-directional Encoder Representation from Transformers)-based architectures, including an existing architecture (mSpERT) that jointly extracts all event information and a multi-step approach (PL-Marker++) that we augmented for the CAMIR schema.

Namu Park, Farzad Ahmed, Zhaoyi Sun et al.

Objective: To evaluate large language models (LLMs) against supervised baselines for fine-grained, lesion-level detection of incidentalomas requiring follow-up, addressing the limitations of current document-level classification systems. Methods: We utilized a dataset of 400 annotated radiology reports containing 1,623 verified lesion findings. We compared three supervised transformer-based encoders (BioClinicalModernBERT, ModernBERT, Clinical Longformer) against four generative LLM configurations (Llama 3.1-8B, GPT-4o, GPT-OSS-20b). We introduced a novel inference strategy using lesion-tagged inputs and anatomy-aware prompting to ground model reasoning. Performance was evaluated using class-specific F1-scores. Results: The anatomy-informed GPT-OSS-20b model achieved the highest performance, yielding an incidentaloma-positive macro-F1 of 0.79. This surpassed all supervised baselines (maximum macro-F1: 0.70) and closely matched the inter-annotator agreement of 0.76. Explicit anatomical grounding yielded statistically significant performance gains across GPT-based models (p < 0.05), while a majority-vote ensemble of the top systems further improved the macro-F1 to 0.90. Error analysis revealed that anatomy-aware LLMs demonstrated superior contextual reasoning in distinguishing actionable findings from benign lesions. Conclusion: Generative LLMs, when enhanced with structured lesion tagging and anatomical context, significantly outperform traditional supervised encoders and achieve performance comparable to human experts. This approach offers a reliable, interpretable pathway for automated incidental finding surveillance in radiology workflows.

Farzad Ahmed, Joniel Augustine Jerome, Meliha Yetisgen et al.

Objective: Clinical documentation contains factual, diagnostic, and management errors that can compromise patient safety. Large language models (LLMs) may help detect and correct such errors, but their behavior under different prompting strategies remains unclear. We evaluate zero-shot prompting, static prompting with random exemplars (SPR), and retrieval-augmented dynamic prompting (RDP) for three subtasks of medical error processing: error flag detection, error sentence detection, and error correction. Methods: Using the MEDEC dataset, we evaluated nine instruction-tuned LLMs (GPT, Claude, Gemini, and OpenAI o-series models). We measured performance using accuracy, recall, false-positive rate (FPR), and an aggregate score of ROUGE-1, BLEURT, and BERTScore for error correction. We also analyzed example outputs to identify failure modes and differences between LLM and clinician reasoning. Results: Zero-shot prompting showed low recall in both detection tasks, often missing abbreviation-heavy or atypical errors. SPR improved recall but increased FPR. Across all nine LLMs, RDP reduced FPR by about 15 percent, improved recall by 5 to 10 percent in error sentence detection, and generated more contextually accurate corrections. Conclusion: Across diverse LLMs, RDP outperforms zero-shot and SPR prompting. Using retrieved exemplars improves detection accuracy, reduces false positives, and enhances the reliability of medical error correction.

Sihang Zeng, Yujuan Fu, Sitong Zhou et al. · uw

Large language models (LLMs) offer a generalizable approach for modeling patient trajectories, but suffer from the long and noisy nature of electronic health records (EHR) data in temporal reasoning. To address these challenges, we introduce Traj-CoA, a multi-agent system involving chain-of-agents for patient trajectory modeling. Traj-CoA employs a chain of worker agents to process EHR data in manageable chunks sequentially, distilling critical events into a shared long-term memory module, EHRMem, to reduce noise and preserve a comprehensive timeline. A final manager agent synthesizes the worker agents' summary and the extracted timeline in EHRMem to make predictions. In a zero-shot one-year lung cancer risk prediction task based on five-year EHR data, Traj-CoA outperforms baselines of four categories. Analysis reveals that Traj-CoA exhibits clinically aligned temporal reasoning, establishing it as a promisingly robust and generalizable approach for modeling complex patient trajectories.

Wen-wai Yim, Asma Ben Abacha, Zixuan Yu et al.

Evaluating natural language generation (NLG) systems in the medical domain presents unique challenges due to the critical demands for accuracy, relevance, and domain-specific expertise. Traditional automatic evaluation metrics, such as BLEU, ROUGE, and BERTScore, often fall short in distinguishing between high-quality outputs, especially given the open-ended nature of medical question answering (QA) tasks where multiple valid responses may exist. In this work, we introduce MORQA (Medical Open-Response QA), a new multilingual benchmark designed to assess the effectiveness of NLG evaluation metrics across three medical visual and text-based QA datasets in English and Chinese. Unlike prior resources, our datasets feature 2-4+ gold-standard answers authored by medical professionals, along with expert human ratings for three English and Chinese subsets. We benchmark both traditional metrics and large language model (LLM)-based evaluators, such as GPT-4 and Gemini, finding that LLM-based approaches significantly outperform traditional metrics in correlating with expert judgments. We further analyze factors driving this improvement, including LLMs' sensitivity to semantic nuances and robustness to variability among reference answers. Our results provide the first comprehensive, multilingual qualitative study of NLG evaluation in the medical domain, highlighting the need for human-aligned evaluation methods. All datasets and annotations will be publicly released to support future research.

Sihang Zeng, Lucas Jing Liu, Jun Wen et al.

Trustworthy survival prediction is essential for clinical decision making. Longitudinal electronic health records (EHRs) provide a uniquely powerful opportunity for the prediction. However, it is challenging to accurately model the continuous clinical progression of patients underlying the irregularly sampled clinical features and to transparently link the progression to survival outcomes. To address these challenges, we develop TrajSurv, a model that learns continuous latent trajectories from longitudinal EHR data for trustworthy survival prediction. TrajSurv employs a neural controlled differential equation (NCDE) to extract continuous-time latent states from the irregularly sampled data, forming continuous latent trajectories. To ensure the latent trajectories reflect the clinical progression, TrajSurv aligns the latent state space with patient state space through a time-aware contrastive learning approach. To transparently link clinical progression to the survival outcome, TrajSurv uses latent trajectories in a two-step divide-and-conquer interpretation process. First, it explains how the changes in clinical features translate into the latent trajectory's evolution using a learned vector field. Second, it clusters these latent trajectories to identify key clinical progression patterns associated with different survival outcomes. Evaluations on two real-world medical datasets, MIMIC-III and eICU, show TrajSurv's competitive accuracy and superior transparency over existing deep learning methods.

Fardin Ahsan Sakib, Ziwei Zhu, Karen Trister Grace et al.

Social determinants of health (SDOH) extraction from clinical text is critical for downstream healthcare analytics. Although large language models (LLMs) have shown promise, they may rely on superficial cues leading to spurious predictions. Using the MIMIC portion of the SHAC (Social History Annotation Corpus) dataset and focusing on drug status extraction as a case study, we demonstrate that mentions of alcohol or smoking can falsely induce models to predict current/past drug use where none is present, while also uncovering concerning gender disparities in model performance. We further evaluate mitigation strategies - such as prompt engineering and chain-of-thought reasoning - to reduce these false positives, providing insights into enhancing LLM reliability in health domains.

Nic Dobbins, Christelle Xiong, Kristine Lan et al.

Objective: To demonstrate the capabilities of Large Language Models (LLMs) as autonomous agents to reproduce findings of published research studies using the same or similar dataset. Materials and Methods: We used the "Quick Access" dataset of the National Alzheimer's Coordinating Center (NACC). We identified highly cited published research manuscripts using NACC data and selected five studies that appeared reproducible using this dataset alone. Using GPT-4o, we created a simulated research team of LLM-based autonomous agents tasked with writing and executing code to dynamically reproduce the findings of each study, given only study Abstracts, Methods sections, and data dictionary descriptions of the dataset. Results: We extracted 35 key findings described in the Abstracts across 5 Alzheimer's studies. On average, LLM agents approximately reproduced 53.2% of findings per study. Numeric values and range-based findings often differed between studies and agents. The agents also applied statistical methods or parameters that varied from the originals, though overall trends and significance were sometimes similar. Discussion: In some cases, LLM-based agents replicated research techniques and findings. In others, they failed due to implementation flaws or missing methodological detail. These discrepancies show the current limits of LLMs in fully automating reproducibility assessments. Still, this early investigation highlights the potential of structured agent-based systems to provide scalable evaluation of scientific rigor. Conclusion: This exploratory work illustrates both the promise and limitations of LLMs as autonomous agents for automating reproducibility in biomedical research.

Wilson Lau, Kevin Lybarger, Martin L. Gunn et al.

Radiology reports contain a diverse and rich set of clinical abnormalities documented by radiologists during their interpretation of the images. Comprehensive semantic representations of radiological findings would enable a wide range of secondary use applications to support diagnosis, triage, outcomes prediction, and clinical research. In this paper, we present a new corpus of radiology reports annotated with clinical findings. Our annotation schema captures detailed representations of pathologic findings that are observable on imaging ("lesions") and other types of clinical problems ("medical problems"). The schema used an event-based representation to capture fine-grained details, including assertion, anatomy, characteristics, size, count, etc. Our gold standard corpus contained a total of 500 annotated computed tomography (CT) reports. We extracted triggers and argument entities using two state-of-the-art deep learning architectures, including BERT. We then predicted the linkages between trigger and argument entities (referred to as argument roles) using a BERT-based relation extraction model. We achieved the best extraction performance using a BERT model pre-trained on 3 million radiology reports from our institution: 90.9%-93.4% F1 for finding triggers 72.0%-85.6% F1 for arguments roles. To assess model generalizability, we used an external validation set randomly sampled from the MIMIC Chest X-ray (MIMIC-CXR) database. The extraction performance on this validation set was 95.6% for finding triggers and 79.1%-89.7% for argument roles, demonstrating that the model generalized well to the cross-institutional data with a different imaging modality. We extracted the finding events from all the radiology reports in the MIMIC-CXR database and provided the extractions to the research community.

Kevin Lybarger, Aashka Damani, Martin Gunn et al.

Medical imaging is critical to the diagnosis and treatment of numerous medical problems, including many forms of cancer. Medical imaging reports distill the findings and observations of radiologists, creating an unstructured textual representation of unstructured medical images. Large-scale use of this text-encoded information requires converting the unstructured text to a structured, semantic representation. We explore the extraction and normalization of anatomical information in radiology reports that is associated with radiological findings. We investigate this extraction and normalization task using a span-based relation extraction model that jointly extracts entities and relations using BERT. This work examines the factors that influence extraction and normalization performance, including the body part/organ system, frequency of occurrence, span length, and span diversity. It discusses approaches for improving performance and creating high-quality semantic representations of radiological phenomena.

Kevin Lybarger, Linzee Mabrey, Matthew Thau et al.

Acute respiratory distress syndrome (ARDS) is a life-threatening condition that is often undiagnosed or diagnosed late. ARDS is especially prominent in those infected with COVID-19. We explore the automatic identification of ARDS indicators and confounding factors in free-text chest radiograph reports. We present a new annotated corpus of chest radiograph reports and introduce the Hierarchical Attention Network with Sentence Objectives (HANSO) text classification framework. HANSO utilizes fine-grained annotations to improve document classification performance. HANSO can extract ARDS-related information with high performance by leveraging relation annotations, even if the annotated spans are noisy. Using annotated chest radiograph images as a gold standard, HANSO identifies bilateral infiltrates, an indicator of ARDS, in chest radiograph reports with performance (0.87 F1) comparable to human annotations (0.84 F1). This algorithm could facilitate more efficient and expeditious identification of ARDS by clinicians and researchers and contribute to the development of new therapies to improve patient care.

Nicholas Dobbins, David Wayne, Kahyun Lee et al.

Free-text clinical notes detail all aspects of patient care and have great potential to facilitate quality improvement and assurance initiatives as well as advance clinical research. However, concerns about patient privacy and confidentiality limit the use of clinical notes for research. As a result, the information documented in these notes remains unavailable for most researchers. De-identification (de-id), i.e., locating and removing personally identifying protected health information (PHI), is one way of improving access to clinical narratives. However, there are limited off-the-shelf de-identification systems able to consistently detect PHI across different data sources and medical specialties. In this abstract, we present the performance of a state-of-the art de-id system called NeuroNER1 on a diverse set of notes from University of Washington (UW) when the models are trained on data from an external institution (Partners Healthcare) vs. from the same institution (UW). We present results at the level of PHI and note types.

Kahyun Lee, Nicholas J. Dobbins, Bridget McInnes et al.

Objective: Neural network de-identification studies have focused on individual datasets. These studies assume the availability of a sufficient amount of human-annotated data to train models that can generalize to corresponding test data. In real-world situations, however, researchers often have limited or no in-house training data. Existing systems and external data can help jump-start de-identification on in-house data; however, the most efficient way of utilizing existing systems and external data is unclear. This article investigates the transferability of a state-of-the-art neural clinical de-identification system, NeuroNER, across a variety of datasets, when it is modified architecturally for domain generalization and when it is trained strategically for domain transfer. Methods and Materials: We conducted a comparative study of the transferability of NeuroNER using four clinical note corpora with multiple note types from two institutions. We modified NeuroNER architecturally to integrate two types of domain generalization approaches. We evaluated each architecture using three training strategies. We measured: transferability from external sources; transferability across note types; the contribution of external source data when in-domain training data are available; and transferability across institutions. Results and Conclusions: Transferability from a single external source gave inconsistent results. Using additional external sources consistently yielded an F1-score of approximately 80%. Fine-tuning emerged as a dominant transfer strategy, with or without domain generalization. We also found that external sources were useful even in cases where in-domain training data were available. Transferability across institutions differed by note type and annotation label but resulted in improved performance.

Paul Barry, Sam Henry, Meliha Yetisgen et al.

We hypothesize that explicit integration of contextual information into an Multi-task Learning framework would emphasize the significance of context for boosting performance in jointly learning Named Entity Recognition (NER) and Relation Extraction (RE). Our work proves this hypothesis by segmenting entities from their surrounding context and by building contextual representations using each independent segment. This relation representation allows for a joint NER/RE system that achieves near state-of-the-art (SOTA) performance on both NER and RE tasks while beating the SOTA RE system at end-to-end NER & RE with a 49.07 F1.

Kevin Lybarger, Mari Ostendorf, Matthew Thompson et al.

Coronavirus disease 2019 (COVID-19) is a global pandemic. Although much has been learned about the novel coronavirus since its emergence, there are many open questions related to tracking its spread, describing symptomology, predicting the severity of infection, and forecasting healthcare utilization. Free-text clinical notes contain critical information for resolving these questions. Data-driven, automatic information extraction models are needed to use this text-encoded information in large-scale studies. This work presents a new clinical corpus, referred to as the COVID-19 Annotated Clinical Text (CACT) Corpus, which comprises 1,472 notes with detailed annotations characterizing COVID-19 diagnoses, testing, and clinical presentation. We introduce a span-based event extraction model that jointly extracts all annotated phenomena, achieving high performance in identifying COVID-19 and symptom events with associated assertion values (0.83-0.97 F1 for events and 0.73-0.79 F1 for assertions). In a secondary use application, we explored the prediction of COVID-19 test results using structured patient data (e.g. vital signs and laboratory results) and automatically extracted symptom information. The automatically extracted symptoms improve prediction performance, beyond structured data alone.

Wilson Lau, Laura Aaltonen, Martin Gunn et al.

Selecting radiology examination protocol is a repetitive, and time-consuming process. In this paper, we present a deep learning approach to automatically assign protocols to computer tomography examinations, by pre-training a domain-specific BERT model ($BERT_{rad}$). To handle the high data imbalance across exam protocols, we used a knowledge distillation approach that up-sampled the minority classes through data augmentation. We compared classification performance of the described approach with the statistical n-gram models using Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and Random Forest (RF) classifiers, as well as the Google's $BERT_{base}$ model. SVM, GBM and RF achieved macro-averaged F1 scores of 0.45, 0.45, and 0.6 while $BERT_{base}$ and $BERT_{rad}$ achieved 0.61 and 0.63. Knowledge distillation improved overall performance on the minority classes, achieving a F1 score of 0.66.

Kevin Lybarger, Mari Ostendorf, Meliha Yetisgen

Social determinants of health (SDOH) affect health outcomes, and knowledge of SDOH can inform clinical decision-making. Automatically extracting SDOH information from clinical text requires data-driven information extraction models trained on annotated corpora that are heterogeneous and frequently include critical SDOH. This work presents a new corpus with SDOH annotations, a novel active learning framework, and the first extraction results on the new corpus. The Social History Annotation Corpus (SHAC) includes 4,480 social history sections with detailed annotation for 12 SDOH characterizing the status, extent, and temporal information of 18K distinct events. We introduce a novel active learning framework that selects samples for annotation using a surrogate text classification task as a proxy for a more complex event extraction task. The active learning framework successfully increases the frequency of health risk factors and improves automatic extraction of these events over undirected annotation. An event extraction model trained on SHAC achieves high extraction performance for substance use status (0.82-0.93 F1), employment status (0.81-0.86 F1), and living status type (0.81-0.93 F1) on data from three institutions.

Wilson Lau, Thomas H Payne, Ozlem Uzuner et al.

Communication of follow-up recommendations when abnormalities are identified on imaging studies is prone to error. In this paper, we present a natural language processing approach based on deep learning to automatically identify clinically important recommendations in radiology reports. Our approach first identifies the recommendation sentences and then extracts reason, test, and time frame of the identified recommendations. To train our extraction models, we created a corpus of 567 radiology reports annotated for recommendation information. Our extraction models achieved 0.92 f-score for recommendation sentence, 0.65 f-score for reason, 0.73 f-score for test, and 0.84 f-score for time frame. We applied the extraction models to a set of over 3.3 million radiology reports and analyzed the adherence of follow-up recommendations.