STop Clock for Automated Tracking (STAT) during Time-Critical Medical Work: Evaluating the Accuracy and Usability of an AI-Driven Automated Stop Clock

Presentation Time: 09:45 AM - 09:57 AM

Abstract Keywords: Human-computer Interaction, Qualitative Methods, Artificial Intelligence, Clinical Decision Support, Quantitative Methods, Usability
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Delays and process inefficiencies during trauma resuscitation, a critical, initial phase of care for injured patients, can contribute to adverse patient outcomes. While tracking elapsed time may improve the trauma team’s temporal awareness and reduce delays, reliance on manual activation of stop clocks can introduce variability. To address this limitation, we implemented a computer vision-powered automatic stop clock designed to activate upon patient arrival without requiring manual input. We conducted a retrospective video review of 50 trauma resuscitation cases to assess how the clock was used in practice, followed by semi-structured interviews with nine trauma team members to elicit their feedback and perceptions. This study contributes to the broader discussion on AI-assisted clinical tools, highlighting how automation can support trauma teams by reducing variability in time tracking and improving process efficiency.

Speaker:
Katherine Zellner, MPH
Drexel University

Authors:
Katherine Zellner, MPH - Drexel University; Sifan Yuan, BS - Rutgers University; Emily Ernst, BA Student - Drexel University; Dylan Arkowitz, BA - Children's National Hospital; Aaron Mun, BS - Children's National Hospital; Mary Kim, MD - Children's National Hospital; Ivan Marsic, PhD - Rutgers University; Randall Burd, MD, PhD - Children's National Hospital; Aleksandra Sarcevic, PhD - Drexel University;

---

Failure Modes of Time Series Interpretability Algorithms for Critical Care Applications and Potential Solutions

Presentation Time: 09:57 AM - 10:09 AM

Abstract Keywords: Critical Care, Deep Learning, Machine Learning, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Interpretability plays a vital role in aligning and deploying deep learning models in critical care, especially in constantly evolving conditions that influence patient survival. However, common interpretability algorithms face unique challenges when applied to dynamic prediction tasks, where patient trajectories evolve over time. Gradient, Occlusion, and Permutation-based methods often struggle with time-varying target dependency and temporal smoothness. This work systematically analyzes these failure modes and supports learnable mask-based interpretability frameworks as alternatives, which can incorporate temporal continuity and label consistency constraints to learn feature importance over time. Here, we propose that learnable mask-based approaches for dynamic time-series prediction problems provide more reliable and consistent interpretations for applications in critical care and similar domains.

Speaker:
Shashank Yadav, PhD Student
University of Arizona

Author:
Vignesh Subbian, PhD - University of Arizona;

---

Validation of the pCREST Machine Learning Model for Predicting Code Events Among Critically Ill Children

Presentation Time: 10:09 AM - 10:21 AM

Abstract Keywords: Critical Care, Pediatrics, Machine Learning, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

pCREST is a validated machine learning model that continuously predicts the risk of deterioration in hospitalized children. The objective of this study was to evaluate the performance of pCREST in predicting code events among critically ill children. We compared the performance of pCREST to the baseline model, pSOFA. pCREST performs well (AUROC 0.80 [0.80, 0.81]) in predicting the outcome and outperforms the baseline model, pSOFA, at several time points before the outcome.

Speaker:
Colleen Badke, MD, MPH
Ann & Robert H. Lurie Children's Hospital of Chicago

Authors:
Sierra Strutz, PhD Student in Biomedical Data Science - University of Wisconsin - Madison; Austin Wang, MD - Johns Hopkins; 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; Anoop Mayampurath, PhD - University of Wisconsin - Madison;

---

Racial and Ethnic Disparities in Identification of Cyanosis in ICU Settings

Presentation Time: 10:21 AM - 10:33 AM

Abstract Keywords: Fairness and Elimination of Bias, Data Mining, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Cyanosis is a discoloration of the skin arising from deoxygenated hemoglobin in the blood, caused by heart, lung,
and blood diseases and treated with interventions including supplemental oxygen therapy. Cyanosis presents as a
bluish discoloration in light-skinned patients, but as a gray or white discoloration in dark-skinned patients. While
prior work hints at the under-identification of cyanosis for people with black and brown skin, in this study, we quantify differences in cyanosis identification rates and associated clinical treatments by race/ethnicity. Leveraging EHR
datasets from two hospital systems, we extract cyanosis mentions from clinical notes and compare cyanosis documentation rates by documented race/ethnicity. Cyanosis documentation was significantly less frequent for Black patients than White patients after adjusting for confounders. We measure impacts of cyanosis identification on provision of oxygen, vasopressors, and fluids. Adjusting for severity of a patient’s condition, documentation of cyanosis was associated with faster provision of oxygen.

Speaker:
Izzy Chaiken, BA
University of Washington

Authors:
Izzy Chaiken, BA - University of Washington; Neha A. Sathe, MD, MSc - University of Washington; Lucy Lu Wang, PhD - University of Washington; Mark M. Wurfel, MD, PhD - University of Washington Medicine;

---

AKI-Detector: A Multi-Agent Framework by Integrating Machine Learning and Large Language Models for Early Prediction of Acute Kidney Injury in ICU

Presentation Time: 10:33 AM - 10:45 AM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Large Language Models (LLMs), Critical Care, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Acute kidney injury (AKI) is a severe condition in ICUs, where early prediction is crucial for timely intervention and prevention. Traditional machine learning (ML) models lack interpretability, limiting real-world applicability. We propose AKI-Detector, a novel multi-agent framework that integrates ML, large language models (LLMs), and retrieval-augmented generation (RAG) to enhance accuracy and interpretability of AKI prediction. The framework combines structured EHR-based ML models, LLMs, and RAG to enhance clinical reasoning and explanations. The proposed AKI-Detector mitigates LLM hallucinations and bridges the gap between algorithmic output and clinically interpretable reports. Evaluated on ICU patient data from MIMIC-IV, AKI-Detector outperformed traditional models such as CatBoost and GRU, especially for patients with much uncertainties. Case studies demonstrate its ability to produce transparent, evidence-based reports that align with real-world diagnosis logic. This work highlights the promise of real-world big data and LLM-powered multi-agent systems to support trustworthy and explainable AI for clinical prediction.

Speaker:
Guilan Kong, PhD
National Institute of Health Data Science, Peking University

Authors:
Tongyue Shi, MS - Peking University; Meirong Xiao, PhD - Peking University; Haowei Xu, MS - Peking University; Huiying Zhao, MD - Peking University People's Hospital; Guilan Kong, PhD - National Institute of Health Data Science, Peking University;

---

Development of a High-Resolution Data Reporting Tool for Mechanical Ventilation Best-Practices in the Pediatric ICU

Presentation Time: 10:45 AM - 10:57 AM

Abstract Keywords: Critical Care, Pediatrics, Data Modernization, Data Standards, Real-World Evidence Generation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

In order to examine mechanical ventilation practices and compliance with guidelines, we developed a series of high-resolution queries that allowed us to generate new insights into unit operations and areas for improvement.

Speaker:
Blake Vander Wood, MD
Emory University

Authors:
Evan Orenstein, MD - Children's Healthcare of Atlanta; Mark Mai, MD, MHS - Children's Healthcare of Atlanta;

---