- Home
- 2025 Annual Symposium Gallery
- Contextual Phenotyping of Pediatric Sepsis Cohort Using Large Language Models
Custom CSS
double-click to edit, do not edit in source
11/17/2025 |
2:00 PM – 3:15 PM |
Room 8
S39: The Fast and the Febrile: Racing Against Sepsis
Presentation Type: Oral Presentations
Earlier ICU Transfer after CONCERN Early Warning System Score Escalation Reduced Sepsis-related Mortality: Results from a Multi-site Pragmatic Cluster Randomized Controlled Trial
Presentation Time: 02:00 PM - 02:15 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Nursing Informatics, Patient Safety, Real-World Evidence Generation, Machine Learning, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early recognition and timely escalation of care are critical for improving sepsis outcomes. This post-hoc analysis of a multi-site clinical trial examined whether the timing of ICU transfer following CONCERN Early Warning System (EWS) score escalation was associated with in-hospital mortality among patients later diagnosed with sepsis. Among 54 patients with score escalations prior to unanticipated ICU transfer, shorter score-change-to-ICU-transfer time intervals were significantly associated with lower odds of in-hospital death. A 36-hour threshold emerged as a potential inflection point; all patients transferred after this interval died. No significant differences were observed in the ICU-arrival-to-sepsis time interval between early and late transfers. These findings highlight the importance of acting promptly on early warning systems and suggest that CONCERN EWS may offer a meaningful lead time for intervention improving outcomes for sepsis patients.
Speaker:
Rachel Lee, PhD, RN
Columbia University
Authors:
Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia; PATRICIA C DYKES, PhD, MA, RN - Brigham and Women's Hospital/Harvard Medical School; Graham Lowenthal, BA - Brigham and Women's Hospital; Sandy Cho, MPH BSN - Newton-Wellesley Hospital; Haomiao Jia, PhD - Columbia University; Temiloluwa Daramola, BA - Columbia University Irving Medical Center - Department of Biomedical Informatics; Sachleen Tuteja, BS in Data Science and Statistics - Northwestern University; Sarah Rossetti, RN, PhD - Columbia University Department of Biomedical Informatics;
Presentation Time: 02:00 PM - 02:15 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Nursing Informatics, Patient Safety, Real-World Evidence Generation, Machine Learning, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early recognition and timely escalation of care are critical for improving sepsis outcomes. This post-hoc analysis of a multi-site clinical trial examined whether the timing of ICU transfer following CONCERN Early Warning System (EWS) score escalation was associated with in-hospital mortality among patients later diagnosed with sepsis. Among 54 patients with score escalations prior to unanticipated ICU transfer, shorter score-change-to-ICU-transfer time intervals were significantly associated with lower odds of in-hospital death. A 36-hour threshold emerged as a potential inflection point; all patients transferred after this interval died. No significant differences were observed in the ICU-arrival-to-sepsis time interval between early and late transfers. These findings highlight the importance of acting promptly on early warning systems and suggest that CONCERN EWS may offer a meaningful lead time for intervention improving outcomes for sepsis patients.
Speaker:
Rachel Lee, PhD, RN
Columbia University
Authors:
Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia; PATRICIA C DYKES, PhD, MA, RN - Brigham and Women's Hospital/Harvard Medical School; Graham Lowenthal, BA - Brigham and Women's Hospital; Sandy Cho, MPH BSN - Newton-Wellesley Hospital; Haomiao Jia, PhD - Columbia University; Temiloluwa Daramola, BA - Columbia University Irving Medical Center - Department of Biomedical Informatics; Sachleen Tuteja, BS in Data Science and Statistics - Northwestern University; Sarah Rossetti, RN, PhD - Columbia University Department of Biomedical Informatics;
Rachel
Lee,
PhD, RN - Columbia University
Subphenotyping Pediatric Sepsis Patients using Early Dynamics of Cardiovascular Status
Presentation Time: 02:15 PM - 02:30 PM
Abstract Keywords: Pediatrics, Precision Medicine, Artificial Intelligence, Critical Care, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Sepsis is a life-threatening condition with substantial clinical heterogeneity, requiring tailored treatment and early interventions. Using cardiovascular trajectories adjusted for the vasoactive medication use during the first 8 hours, we stratified pediatric sepsis patients into three subgroups. Patients with consistently decreasing trends had significantly higher hospital mortality at 24 hours, 3 days, and 7 days. Early identification of this high-risk subgroup provides valuable guidance toward precision medicine strategies in pediatric sepsis care.
Speaker:
Cheol Min Lee, PharmD
Northwestern University
Author:
L. Nelson Sanchez-Pinto, MD, MBI, FAMIA - Northwestern University Feinberg School of Medicine and Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL;
Presentation Time: 02:15 PM - 02:30 PM
Abstract Keywords: Pediatrics, Precision Medicine, Artificial Intelligence, Critical Care, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Sepsis is a life-threatening condition with substantial clinical heterogeneity, requiring tailored treatment and early interventions. Using cardiovascular trajectories adjusted for the vasoactive medication use during the first 8 hours, we stratified pediatric sepsis patients into three subgroups. Patients with consistently decreasing trends had significantly higher hospital mortality at 24 hours, 3 days, and 7 days. Early identification of this high-risk subgroup provides valuable guidance toward precision medicine strategies in pediatric sepsis care.
Speaker:
Cheol Min Lee, PharmD
Northwestern University
Author:
L. Nelson Sanchez-Pinto, MD, MBI, FAMIA - Northwestern University Feinberg School of Medicine and Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL;
Cheol Min
Lee,
PharmD - Northwestern University
External Validation of the Epic Early Detection of Sepsis Model on Standard Sepsis Outcome Definitions
Presentation Time: 02:30 PM - 02:45 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Prior evaluations of the Epic Early Detection of Sepsis model (ESM) have shown poor performance. We performed a prospective external validation of Epic's newest locally trained ESM on three standardly used sepsis outcome definitions in seven acute care hospitals. The model performs moderately well, but has a high false positive rate.
Speaker:
Sayon Dutta
Mass General Hospital
Authors:
Reid McMurry, MS - Boston University School of Medicine; Reid Mcmurry, MD - Boston University School of Medicine; Lisette Dunham, MPS, MSPH - Mass General Brigham Digital; Timothy Stump, BS - Mass General Brigham Digital; Michael Filbin, MD - Mass General Hospital; Chanu Rhee, MD, PhD - Brigham and Women's Hospital;
Presentation Time: 02:30 PM - 02:45 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Prior evaluations of the Epic Early Detection of Sepsis model (ESM) have shown poor performance. We performed a prospective external validation of Epic's newest locally trained ESM on three standardly used sepsis outcome definitions in seven acute care hospitals. The model performs moderately well, but has a high false positive rate.
Speaker:
Sayon Dutta
Mass General Hospital
Authors:
Reid McMurry, MS - Boston University School of Medicine; Reid Mcmurry, MD - Boston University School of Medicine; Lisette Dunham, MPS, MSPH - Mass General Brigham Digital; Timothy Stump, BS - Mass General Brigham Digital; Michael Filbin, MD - Mass General Hospital; Chanu Rhee, MD, PhD - Brigham and Women's Hospital;
Sayon
Dutta - Mass General Hospital
Contextual Phenotyping of Pediatric Sepsis Cohort Using Large Language Models
Presentation Time: 02:45 PM - 03:00 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Pediatrics, Critical Care, Clinical Decision Support, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Clustering patient subgroups is critical for personalized care and efficient resource allocation. Traditional methods falter with high-dimensional, heterogeneous healthcare data lacking contextual nuance. This study compares Large Language Model (LLM)-based clustering to classical techniques using a pediatric sepsis dataset from a low-income country, comprising 2,686 records with 28 numerical and 119 categorical variables. Patient records were serialized into text both with and without a clustering objective. Embeddings were generated using quantized LLAMA 3.1 8B, DeepSeek-R1-Distill-Llama-8B (with LoRA), and Stella-En-400M-V5, then clustered via K-means. Classical comparisons involved K-Medoids on UMAP and FAMD-reduced data. Silhouette scores and statistical tests assessed cluster quality. Notably, Stella-En-400M-V5 achieved the highest score (0.86), while LLAMA 3.1 8B with the clustering objective better distinguished subgroups with unique nutritional, clinical, and socioeconomic profiles. LLM-based methods outperformed classical techniques, capturing richer context and prioritizing key features. These promising results warrant exploration. The approach shows strong promise for future applications.
Speaker:
Aditya Nagori, PhD
Duke
Authors:
Aditya Nagori, PhD - Duke; Aditya Nagori, PhD - , Department of surgery, Duke university school of medicine, Durham NC, USA, Department of Anesthesiology, Duke university school of medicine, Durham NC, USA; Ayush Gautam, Btech - Indian Institute of Technology, Goa, India; Matthew Wiens, PhD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada 5 Department of Anesthesia, Pharmacology & Therapeutics, University of British Columbia, Vancouver, BC, Canada 6 BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada; Vuong Nguyen, PhD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada; Nathan Kenya Mugisha, MPH - Walimu, Kampala, Uganda; Jerome Kabakyenga Kabakyenga, PhD - Department of Community Health, Mbarara University of Science and Technology, Mbarara, Uganda 10 Maternal Newborn and Child Health Institute, Mbarara University of Science and Technology, Mbarara, Uganda; Niranjan Kissoon, MD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada, BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada 7 Department of Pediatrics, University of British Columbia, Vancouver, BC, Canada; J. Mark Ansermino, MD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada 5 Department of Anesthesia, Pharmacology & Therapeutics, University of British Columbia, Vancouver, BC, Canada 6 BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada; Rishikesan Kamaleswaran, PhD - Department of surgery, Duke university school of medicine, Durham NC, USA 2Department of Anesthesiology, Duke university school of medicine, Durham NC, USA;
Presentation Time: 02:45 PM - 03:00 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Pediatrics, Critical Care, Clinical Decision Support, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Clustering patient subgroups is critical for personalized care and efficient resource allocation. Traditional methods falter with high-dimensional, heterogeneous healthcare data lacking contextual nuance. This study compares Large Language Model (LLM)-based clustering to classical techniques using a pediatric sepsis dataset from a low-income country, comprising 2,686 records with 28 numerical and 119 categorical variables. Patient records were serialized into text both with and without a clustering objective. Embeddings were generated using quantized LLAMA 3.1 8B, DeepSeek-R1-Distill-Llama-8B (with LoRA), and Stella-En-400M-V5, then clustered via K-means. Classical comparisons involved K-Medoids on UMAP and FAMD-reduced data. Silhouette scores and statistical tests assessed cluster quality. Notably, Stella-En-400M-V5 achieved the highest score (0.86), while LLAMA 3.1 8B with the clustering objective better distinguished subgroups with unique nutritional, clinical, and socioeconomic profiles. LLM-based methods outperformed classical techniques, capturing richer context and prioritizing key features. These promising results warrant exploration. The approach shows strong promise for future applications.
Speaker:
Aditya Nagori, PhD
Duke
Authors:
Aditya Nagori, PhD - Duke; Aditya Nagori, PhD - , Department of surgery, Duke university school of medicine, Durham NC, USA, Department of Anesthesiology, Duke university school of medicine, Durham NC, USA; Ayush Gautam, Btech - Indian Institute of Technology, Goa, India; Matthew Wiens, PhD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada 5 Department of Anesthesia, Pharmacology & Therapeutics, University of British Columbia, Vancouver, BC, Canada 6 BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada; Vuong Nguyen, PhD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada; Nathan Kenya Mugisha, MPH - Walimu, Kampala, Uganda; Jerome Kabakyenga Kabakyenga, PhD - Department of Community Health, Mbarara University of Science and Technology, Mbarara, Uganda 10 Maternal Newborn and Child Health Institute, Mbarara University of Science and Technology, Mbarara, Uganda; Niranjan Kissoon, MD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada, BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada 7 Department of Pediatrics, University of British Columbia, Vancouver, BC, Canada; J. Mark Ansermino, MD - Institute for Global Health, BC Children’s Hospital and BC Women’s Hospital + Health Centre, Vancouver, BC, Canada 5 Department of Anesthesia, Pharmacology & Therapeutics, University of British Columbia, Vancouver, BC, Canada 6 BC Children’s Hospital Research Institute, BC Children’s Hospital, Vancouver, BC, Canada; Rishikesan Kamaleswaran, PhD - Department of surgery, Duke university school of medicine, Durham NC, USA 2Department of Anesthesiology, Duke university school of medicine, Durham NC, USA;
Aditya
Nagori,
PhD - Duke
Emergency Sepsis Risk Prediction Score for Early and Rapid Identification in the Emergency Department
Presentation Time: 03:00 PM - 03:15 PM
Abstract Keywords: Critical Care, Clinical Decision Support, Machine Learning, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early sepsis detection in the emergency department (ED) is vital for timely intervention. Existing models often require numerous variables or lack real-time applicability. Using the AutoScore framework and MIMIC-IV-ED data, we developed the Emergency Sepsis Risk Prediction (ESRP) Score—a nine-variable, point-based tool with superior performance and interpretability. ESRP enhances rapid sepsis identification, aiding timely and informed clinical decisions in the ED.
Speaker:
Yanwei Jin, Bachelor of Science in Nursing, Peking University
University of Minnesota, Twin Cities
Authors:
Yanwei Jin, Bachelor of Science in Nursing, Peking University - University of Minnesota, Twin Cities; Yinzhao Wang, Masters - University of Minnesota; Michael Puskarich, MD MSCR - University of Minnesota; Feng Xie, PhD - University of Minnesota;
Presentation Time: 03:00 PM - 03:15 PM
Abstract Keywords: Critical Care, Clinical Decision Support, Machine Learning, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early sepsis detection in the emergency department (ED) is vital for timely intervention. Existing models often require numerous variables or lack real-time applicability. Using the AutoScore framework and MIMIC-IV-ED data, we developed the Emergency Sepsis Risk Prediction (ESRP) Score—a nine-variable, point-based tool with superior performance and interpretability. ESRP enhances rapid sepsis identification, aiding timely and informed clinical decisions in the ED.
Speaker:
Yanwei Jin, Bachelor of Science in Nursing, Peking University
University of Minnesota, Twin Cities
Authors:
Yanwei Jin, Bachelor of Science in Nursing, Peking University - University of Minnesota, Twin Cities; Yinzhao Wang, Masters - University of Minnesota; Michael Puskarich, MD MSCR - University of Minnesota; Feng Xie, PhD - University of Minnesota;
Yanwei
Jin,
Bachelor of Science in Nursing, Peking University - University of Minnesota, Twin Cities
Contextual Phenotyping of Pediatric Sepsis Cohort Using Large Language Models
Category
Paper - Regular
Description
Custom CSS
double-click to edit, do not edit in source
11/17/2025 03:15 PM (Eastern Time (US & Canada))