A research article entitled “Beyond Fine-Tuning: Leveraging Domain-Aware In-Context Learning with Large Language Models for Clinical Named Entity Recognition” has been published in the Journal of Biomedical Informatics (Volume 174, February 2026). The Journal of Biomedical Informatics is a leading international journal in the field of biomedical and clinical informatics that publish high-impact research at the intersection of medicine, data science, and artificial intelligence. Our study presents a comprehensive evaluation of large language model (LLM) based in-context learning (ICL) for clinical named entity recognition (NER) under realistic clinical constraints.

The study compared ICL with conventional encoder-based fine-tuning approaches using manually annotated real-world clinical notes. ICL configurations were optimized across task instructions, output formats, demonstration selection methods, sorting strategies, and demonstration pool sizes using the open-source LLaMA-3.3-70B model, while encoder-based fine-tuning was performed with RoBERTa-large as the strongest baseline.

The results showed that optimized demonstration selection played a major role in determining ICL performance, and improved macro F1 scores compared to random selection. In moderate-resource settings, ICL outperformed encoder-based fine-tuning and remained competitive across a wide range of labeled data sizes. Moreover, ICL demonstrated superior data efficiency and achieved substantial gains in cross-institutional transfer without requiring any parameter updates.

Overall, the study highlights the potential of domain-aware ICL as a flexible and data-efficient alternative to encoder-based fine-tuning for clinical NER. By enabling rapid adaptation through demonstration pools rather than retraining, the proposed approach offers a practical solution for real-world clinical natural language processing in dynamic and resource-limited healthcare environments.

This research was supported by grants from the MSIT (IITP), the Ministry of Health and Welfare (KHIDI), and the Ministry of Education (NRF) of the Republic of Korea.

On Friday, December 26, 2025, CCADD had the annual Year-End Workshop at Seoul National University, Seoul, South Korea. The workshop brought together all current CCADD members, part-time students and winter interns. Throughout a full day of presentations and discussions, CCADD members reflected on the year 2025 and looked ahead to 2026.

The workshop began with arrival and reception, followed by a series of presentations covering research progress, academic development, and practical lessons learned. As the first speaker, Seoyoon opened the program with a session titled “Looking Back on the Year 2025,” highlighting CCADD’s accomplishments in 2025 and celebrating the achievements of its members.

The yearly report was followed by Suhyun, who shared key takeaways from a year of full-time work at CCADD across two phases, from undergraduate research to the first semester of the master’s program. Then, Seeun presented “Preparation for the 1st Oral Presentation,” reflecting on her year-long research and walking through her conference-presentation preparation process. After that, Sunyoung delivered a presentation titled “Life in the Industry vs. Graduate School,” in which she introduced the role of a Clinical Project Leader at Daewoong Pharmaceutical Company highlighting key differences between life in industry and life as a master’s student. As the last session before lunch, David delivered a talk about the things he wished he had known early on, including having a daily habit of reading papers and making a usable personal template to take notes of what you learned from papers.

Following the morning sessions, participants took a break for lunch and then toured the main library at Seoul National University’s Gwanak Campus, exploring its study spaces, archives, and facilities. After the tour, members gathered for a photo time and captured group photos inspired by this year’s Pantone Color theme.

Soyoung presented “How I Learned Actually to Write” and explained what she learned while writing during the second half of 2025 under the supervision of Prof. Howard Lee. Then, Sungwook delivered a workshop talk titled “How I Became the Pre-crastinator at CCADD.” In the talk, he defined a “pre-crastinator” as someone who finishes tasks early and stays in the lab late to keep studying. It was followed by Chaeyoung, who shared her journey for the 2025 Google Solution Challenge, from ideation and development to presenting at the Asian Development Bank in the Philippines. She also reflected on pitching on an international stage and the lessons learned through receiving the award.

After a short break, the workshop continued with sessions focused on practical milestones and next-year planning. In the afternoon session, Seoyoon presented “First Grant Proposal: What I Learned and What I Wish I’d Known,” sharing key challenges and lessons from her first grant project management experience. In the tenth session, Yujin led “Looking Ahead to 2026 at CCADD,” presenting a vision for the center’s research trajectory in the coming year.  It also compiled target conferences for future submissions and concluded with a shared aspiration for deeper research in 2026.

Moreover, in the session, "AI 2027: Where we were, where we are, and where we will be", David went over the prospective report, "AI 2027," in which the authors made predictions on what will happen to the state of AI arms race between US and China. In addition to peeking into the future of AI, David emphasized the need for shifting gears toward using agentic AIs and understanding their inner workings by briefly discussing important developments in the field of agentic AI.

The most significant part of the workshop came in the closing talk, where Prof. Howard presented a paradigm shift from +AI to AI+. He contrasted a domain-first approach with an AI first approach and explaining how research strategy changes under an AI+ framework. He emphasized CCADD’s future direction toward AI-centered approach and proposed holding a “CCADD AI+ Conversion Meeting” in the early 2026 to brainstorm and define core research themes and strategies.

The workshop concluded with closing remarks from Prof. Howard, and everyone gathered for dinner. The 2025 CCADD Year-End Workshop provided members with an opportunity to exchange ideas, reflect on progress, and align on shared goals for the year ahead.

David Seung U Lee, a Ph.D. student at CCADD, and Seeun Park, an undergraduate researcher and the soon-to-be Master’s student at CCADD, received the Outstanding Presentation Awards at the 2025 Fall Conference of the Korean Society of Medical Informatics (KOSMI).

David presented his study titled "Improving Large Language Model-Based Drug-Drug Interaction Prediction Using Conformal Prediction." His research demonstrated that incorporating input features with coverage guarantees, verified through conformal prediction, significantly improves the performance of LLM-based DDI prediction in both in-context learning and fine-tuning settings.

Seeun Park, presented her study titled “Improving Synthetic Clinical Note Generation via Expert Prompting Reflecting Linguistic and Structural Characteristics.” Her research proposes a data augmentation framework that incorporates the knowledge of medical professionals into LLM-based prompting, coupled with a feedback-driven iterative generation loop. This approach enables the model to produce synthetic clinical notes that more closely reflect the characteristics of real-world clinical notes.

These awards highlight the growing importance of LLMs in the healthcare industry and recognizes the potential impact of our work in advancing AI-driven medical text generation.

CCADD members attended the 2025 Fall Conference of the Korean Society of Medical Informatics(KOSMI), held from November 23 to 24 at Songdo Convensia in Incheon. With the theme “Generative AI in Healthcare Systems: From Insight to Impact,” the conference showcased the latest developments, real-world applications, and future opportunities for generative AI in healthcare. Major sessions explored LLM evaluation, LLM-based clinical ation and validation, and AI-driven prediction of neurological disorders.

The conference featured keynote lectures from leading experts in Korea and Taiwan. Professor Sang-Hoon Jeon from Seoul National University Bundang Hospital discussed initiatives aimed at reducing global healthcare disparities, including the establishment of a Global Smart Hospital Network and the development of cloud-based platforms for clinician education and severe disease consultation. He emphasized that the key challenge is not merely how AI can reduce healthcare disparities, but how quickly high-quality medical services can reach those who are waiting for them. Professor Ming-Chin Lin from the College of Medicine at Taipei Medical University addressed the shared challenges facing both Korea and Taiwan—rapid population aging, declining birth rates, and increasing strain on healthcare professionals. He introduced Ambient Intelligence as the next evolution beyond ambient recording and traditional AI systems: a context-aware, ethically governed, sensor-integrated healthcare environment designed to support clinicians and empower patients.

More importantly, CCADD delivered five oral presentations.

David presented his work on “improving the performance of drug interaction prediction using conformal prediction”, which delved into a new method of selecting input for LLM-based prediction module using the statistical framework. 

Yujin presented her work on “Constructing a Domain-Specific Korean–English Parallel Corpus and Dictionary for Medical Translation Using a Large Language Model,” which aims to improve the reliability and consistency of LLM-based medical translation. By introducing a large-scale corpus construction framework, she provided foundational resources that can support future research in the field. During the session, she exchanged insights with reviewers and fellow researchers and explored ways to further expand and deepen the study.

Suhyun An presented her research on “Investigating LLM-based Reasoning and Training Strategies for automating causality assessment of adverse drug reactions”. Her study systematically explored optimal reasoning strategies and training methods to enable large language models to emulate the clinical decision-making process of medical experts.

Seoyoon Jang presented her work on "LLM-Based Extraction of Adverse Events from Real-World Clinical Notes", which leveraged open-source LLMs to build a privacy-preserving and robust framework capable of extracting adverse events from real-world clinical notes.

Seeun Park presented her work titled “Improving Synthetic Clinical Note Generation via Expert Prompting Reflecting Linguistic and Structural Characteristics”. Her research proposed a new synthetic data-augmentation framework that captures the characteristics of real-world clinical notes, demonstrating its potential to support large-scale, high-quality synthetic datasets for AI training.

Through this conference, CCADD members reaffirmed that their research aligns with the current interest in the generative AI, which will play a pivotal role in advancing the healthcare industry. The event strengthened our commitment to pursuing impactful research and contributing to the development of next-generation AI-driven healthcare systems.

Convergence Newsletter, published by the Graduate School of Convergence Science and Technology, Seoul National University, featured two congratulatory events from CCADD.

The first news article introduced Medicall, an AI-based emergency medical solution developed by Chaeyoung Chang, an undergraduate researcher at CCADD. Chaeyoung received the Most Societal Impact Award, one of the Top 3 honors at the 2025 Asia-Pacific Google Solution Challenge, co-hosted by the Asian Development Bank (ADB) and Google, which is a global competition that brings together student developers from Google Developer Groups on Campus to design AI based projects that contribute to the United Nations Sustainable Development Goals (UN SDGs). As a member of Team Atempo, Chaeyoung developed Medicall, an emergency medical service that helps paramedics quickly locate hospitals with available beds during patient transport.

Using Google’s Gemini, Medicall analyzes a patient’s condition and location in real time and recommends the most suitable emergency department within one minute. This enables patients to receive timely treatment within the critical golden hour, improving survival rates and enhancing the efficiency of emergency care systems. The project was praised for its creativity and practical impact, demonstrating how artificial intelligence can be applied to address real healthcare challenges. It was later featured in a news article introducing Medicall as an innovative AI solution to the emergency room crisis in Korea.

The second news article shared Chaeyoung’s experience during the 2025 Summer Convergence Research Program at CCADD, where she took part in two projects that used artificial intelligence to improve the efficiency of drug development. In the first project, she studied how GLP-1 receptor agonists (GLP-1RA) can help prevent nonalcoholic fatty liver disease (NAFLD) in patients with type 2 diabetes by analyzing real-world clinical data. In the second, she worked on developing a machine translation model for regulatory s in the pharmaceutical and medical device fields.

Through these projects, Chaeyoung gained practical experience in data analysis and AI model development. Reflecting on her internship, she said that the structured lectures, mentorship, and book reading sessions on Python and large language models helped her build a strong foundation in AI and strengthened her motivation to pursue graduate research in this field. She expressed her gratitude to the Graduate School of Convergence Science and Technology and to all CCADD members for their guidance and support throughout the program.

The research article entitled “Prephase Steroid Treatment before Remission Induction Chemotherapy in Patients with Pediatric Acute Lymphoblastic Leukemia Using Common Data Model-Based Real-World Data: A Retrospective Observational Study”¹, which was first-authored by the former CCADD graduate,Yoona Choi, and co-authored by the members of CCADD  has been cited in the lastest haematology guidelines on the diagnosis and managemet of tumour lysis syndrome² published by the British Society for Haematology. Moreover, the study was cited in a recent review article published in the New England Journal of Medicine (NEJM)³ detailing the epidemiology and pathophysiology of tumor lysis syndrome.

The British Society for Haematology produces up-to-date evidence-based guidelines on the diagnosis and treatment of haematological disease, while The New England Journal of Medicine is one of the prominent journals of new medical knowledge and clinical best practicies worldwide (Impact Factor 78.5).

This recognition highlights the clinical significance of our study in advancing evidence-based practice for pediatric acute lymphoblastic leukemia. CCADD takes pride in producing research that informs medical decision-making, and will continue efforts to deliver impactful studies.

1_Choi Y, Kim BK, Won JH, Yoo JW, Choi W, Jung S, Kim JY, Choi IY, Chung NG, Lee JW, Choi JY, Kang HJ, Lee H. Prephase steroid treatment before remission induction chemotherapy in patients with pediatric acute lymphoblastic leukemia using common data model-based real-world data: a retrospective observational study. Clin Epidemiol. 2024 Apr 22;16:293-304. doi:10.2147/CLEP.S454263. PMID:38681782; PMCID:PMC11049150.

2_Chan YLT, El-Sharkawi D, Shah P, Chanouzas D, O’Connor D, Marks SD, et al. British Society for Haematology updated guidelines for the diagnosis and management of tumour lysis syndrome in adults and children with haematological malignancies: a focus on patient safety. Br J Haematol. 2025;00:1-11. doi:10.1111/bjh.70092.

3_Bociek RG, Lunning M. Tumor lysis syndrome. N Engl J Med. 2025 Sep 18;393(11):1104-1116. doi:10.1056/NEJMra2300923. PMID:40961427.

Prof. Howard and his team (David and Yujin) presented their works at the 2025 American College of Clinical Pharmacology (ACCP) Annual Meeting, held in Phoenix, Arizona, from September 14–16, 2025. Under the theme “Leveraging Innovative Approaches to Transform Clinical Pharmacology,” the conference gathered experts to exchange insights on the evolving landscape in the field of clinical pharmacology.

The team delivered three poster presentations: 1) “Optimized In-Context Learning Outperforms Finetuning for Clinical Information Extraction across Diverse Healthcare Settings,” 2) “A Set of Inputs with Coverage Guarantee Enhances the Prediction of Drug Interactions,” and 3) “Constructing a Domain-Specific English–Korean Parallel Dictionary for Clinical Pharmacology and Regulatory Science Using a Large Language Model”. The presentations attracted considerable attention and prompted engaging exchanges among participants. The team’s research showed that in-context learning can serve as a data-efficient alternative to finetuning for clinical information extraction, offering practical guidance for applying large language models in healthcare settings with limited resources. They also demonstrated that incorporating input features with coverage guarantees, validated through conformal prediction, enhances the performance of LLM-based drug–drug interaction prediction. Lastly, the team showed that a domain-specific English–Korean parallel dictionary helps overcome language barriers and strengthens communication in clinical pharmacology and regulatory science.

As for the sessions in the ACCP conference, it began with a plenary session on optimizing patient outcomes through individualized pharmacotherapy. Subsequent symposia explored a wide range of critical topics, including diversity plans outlined by the U.S. Food and Drug Administration (FDA), therapeutic strategies for maternal and pediatric populations, innovations in dosing for preterm infants, therapeutic drug monitoring for biologics, and the application of artificial intelligence. Each meeting offered a clear picture of where clinical pharmacology is heading. The focus on individualized pharmacotherapy and model-informed drug development showed how artificial intelligence and mechanistic modeling are reshaping trial design and outcome prediction. At the same time, discussions on therapeutic drug monitoring for biologics and dosing in preterm infants revealed the field’s growing ability to tailor therapy for vulnerable populations.

Collectively, the conference underscored the expanding role of clinical pharmacology at the intersection of data science, patient diversity, and regulation. For the team, presenting at ACCP 2025 was not only an opportunity to share their own research but also a chance to situate their work within the broader trends shaping the field. The experience affirmed the value of their research trajectory and highlighted promising new directions for interdisciplinary and globally relevant contributions to clinical pharmacology.