Yu-Chang YEH

Dr. Yeh is a clinician-scientist with research interests in critical care medicine, including microcirculation, sepsis, blood purification, extracorporeal life support, and the use of generative AI in clinical decision support. His clinical focus includes intensive care, sepsis, PADIS (Pain, Agitation, Delirium, Immobility, and Sleep), and hemodynamic monitoring, and palliative care. He has contributed to the development of the NTUH Center of Microcirculation Medical Research (NCMMR), the NTUH Smart Emergency and Critical Care group (NSECC), and Taiwan CORE, a national critical care database. He has authored over 90 peer-reviewed publications and shared his work at more than 175 national and international conferences.

20th September 2026 Sunday

Time Session
14:00
15:30
Chun-Fu Lai Moderator
Li-Kuo Kuo Moderator
  • Yu-Chang YEH Speaker Clinical Support Information with Generative AI: Content Generation and Evaluation Generative artificial intelligence is shifting from single-question answering toward structured clinical decision support at the bedside. This lecture presents a practical approach to generating and evaluating AI-derived clinical support information in critical care. On the generation side, we describe a multi-turn conversation and multi-task workflow in which structured patient data, a machine learning mortality prediction model, and SHAP-based explanations are passed sequentially to a large language model through five linked task prompts covering risk interpretation, syndrome identification, current status and diagnoses, recommended examinations, and management suggestions. Each turn inherits the context of the previous one, so the output accumulates into a coherent clinical narrative rather than a set of isolated answers. On the evaluation side, we introduce the IMPACT Framework, a six-domain, 21-item instrument developed through a multinational Delphi consensus involving 58 panelists from 12 countries. Its domains, Integration, Mastery, Precision, Applicability, Comprehensiveness, and Timeliness, allow both clinicians and automated judges to score generated content reproducibly. We share validation results, examples from an intensive care cohort, and lessons learned from iterative prompt refinement. Attendees will leave with a transferable method for building and auditing generative AI support tools in their own units.
  • Yu-Chen Chuang Speaker AI-Assisted Reference Management and Graphical Abstract CreationBy the end of this case-based learning session, participants will be able to apply the IMPACT framework to systematically evaluate the appropriateness, reliability, and potential risks of generative AI tools in ICU clinical contexts, while using AI-assisted workflows for reference management and graphical abstract creation to enhance research communication and citation potential.
Room 5
16:00
17:30
Yu-Chang YEH Moderator Clinical Support Information with Generative AI: Content Generation and Evaluation Generative artificial intelligence is shifting from single-question answering toward structured clinical decision support at the bedside. This lecture presents a practical approach to generating and evaluating AI-derived clinical support information in critical care. On the generation side, we describe a multi-turn conversation and multi-task workflow in which structured patient data, a machine learning mortality prediction model, and SHAP-based explanations are passed sequentially to a large language model through five linked task prompts covering risk interpretation, syndrome identification, current status and diagnoses, recommended examinations, and management suggestions. Each turn inherits the context of the previous one, so the output accumulates into a coherent clinical narrative rather than a set of isolated answers. On the evaluation side, we introduce the IMPACT Framework, a six-domain, 21-item instrument developed through a multinational Delphi consensus involving 58 panelists from 12 countries. Its domains, Integration, Mastery, Precision, Applicability, Comprehensiveness, and Timeliness, allow both clinicians and automated judges to score generated content reproducibly. We share validation results, examples from an intensive care cohort, and lessons learned from iterative prompt refinement. Attendees will leave with a transferable method for building and auditing generative AI support tools in their own units.
  • VinCent Wu Speaker AKD Care Redefined: Diabetes, Hypertension, and Strategies for Long-Term Health
  • Kyungho Lee Speaker
  • Nattachai Srisawat Speaker Precision Sepsis-AKI — Biomarkers, AI and Phenotyping in the Asia-PacificTiming of DialysisAcute PD vs Acute HD: Which Is the Right Choice?
  • Kent Doi Speaker Organ crosstalk in AKITargeted Polymyxin B Hemadsorption in Sepsis: Lessons from Japanese Experience and Patient Selection
  • John Prowle Speaker Trajectories of Critical Illness – Defining Endotypes from Routine DataCritical illness is a heterogeneous syndrome characterized by diverse clinical trajectories and variable responses to treatment. Defining biologically and clinically meaningful endotypes using routinely collected healthcare data offers an opportunity to improve risk stratification, prognostication, and personalized therapeutic strategies, enabling earlier intervention, enhanced clinical decision-making, and precision medicine in critical care. In particular the transition from acute to persistent critical illness represents a pivotal phase in intensive care, marked by a shift from the initial disease insult to prolonged organ dysfunction driven by complex host responses. Early identification of patients at risk of this transition remains a major clinical challenge and opportunity.Dynamic Prescription of CRRT Ready for Prime TimeContinuous renal replacement therapy (CRRT) has evolved from a standardized supportive therapy to a platform for precision critical care. Increasing evidence suggests that fixed prescriptions fail to accommodate the dynamic physiological changes encountered during critical illness, resulting in suboptimal solute clearance, fluid management, and metabolic control. Dynamic prescription integrates evolving haemodynamic status, fluid balance, acid–base homeostasis, electrolyte disturbances, and recovery trajectories to optimize therapy delivery while minimizing complications and treatment interruptions. This paradigm supports individualized kidney support rather than protocol-driven care. Emerging decision-support systems and data-driven algorithms have the potential to transform CRRT into a responsive, patient-centred intervention, improving kidney recovery, organ support, and outcomes in critically ill patients. Adapting CRRT for patients with Electrolyte and acid-base disordersContinuous renal replacement therapy (CRRT) is uniquely positioned to provide precise correction of complex electrolyte and acid–base disturbances in critically ill patients. Beyond kidney support, modern CRRT prescriptions can be individualized to account for the severity and evolution of dysnatremia, dyskalemia, calcium disorders, and metabolic acidosis or alkalosis. Customization of dialysate and replacement fluid composition, treatment dose, buffer delivery, and regional citrate anticoagulation enables controlled correction while minimizing the risks of metabolic complications. Dynamic prescription based on serial biochemical monitoring and changing clinical physiology can be embedded into treatment protocols, facilitating safe correction of sodium, potassium, calcium, and bicarbonate abnormalities while optimizing organ support and improving outcomes in critically ill patients. Extracorporeal Therapies in Trauma, Burns and Cerebral OedemaExtracorporeal therapies (ECTs) are an important adjunct in the management of critically ill patients with severe trauma, major burns, and brain injuries. Acute kidney injury, systemic inflammation, rhabdomyolysis, and profound fluid and electrolyte disturbances frequently complicate these conditions, necessitating timely organ support. Continuous kidney replacement therapy (CKRT) provides effective management of metabolic derangements while enabling precise fluid balance in hemodynamically unstable patients. In trauma and burns, ECTs may facilitate the clearance of myoglobin in severe rhabdomyolysis, support fluid management during resuscitation, and could enable removal of damage or pathogen associated molecular patterns mediating multiorgan failure. In patients with cerebral edema, CKRT offers gradual osmotic control and minimizes intracranial pressure fluctuations compared with intermittent dialysis, making it the preferred modality when kidney replacement therapy is required. We will review current evidence, practical considerations, and emerging extracorporeal strategies, highlighting their role in improving physiological stability and supporting recovery in critically ill patients.
Room 1