1062: Pediatric Critical Care Pattern Explorer: A Fuzzy C Mean-Based Framework for Mortality Prediction

Introduction: Pediatric Intensive Care Units (PICUs) face challenges in early risk identification due to patient heterogeneity and data complexity. Traditional clustering methods often fail to accommodate overlapping clinical features. This study introduces a novel data-driven framework—Pediatric Care Pattern Explorer —based on Fuzzy C-Means (FCM) clustering to analyze high-dimensional ICU data and stratify mortality risk in pediatric patients. Methods: This study utilized 8.5 million data points from 1,586 pediatric patients admitted to a tertiary care PICU (2018–2022). After expert-guided feature selection, 131 variables, including vital signs, lab tests, and demographics were reduced to 24 features. FCM clustering was applied to identify patient subgroups with overlapping risk profiles. SMOTE oversampling addressed class imbalance for mortality prediction. Clusters were validated using partition metrics and silhouette scores. Enhanced time-series visualizations and fuzzy membership-based risk scores were developed to support real-time decision-making. Results: Patients were stratified into High, Moderate, and Low-risk clusters based on age-specific indicators. SPO2, heart rate, and blood pressure emerged as universal predictors, while lab markers like WBC, BUN, and Creatinine varied by age group. Time-series plots enabled early detection of deterioration through visual trends and threshold breaches. Fuzzy membership values enabled the estimation of probabilistic mortality risk, enhancing personalized care strategies, by categorization into 3 categories with High risk at higher mortality and low risk with lowest mortality, i.e. patient with near mortality will show higher percentage of it falling into high risk and very low in low risk. Conclusions: FCM clustering offers a flexible and interpretable approach for mortality risk stratification in pediatric ICUs. By incorporating real-time data, probabilistic classification, and age-stratified insights, this framework improves clinical decision-making and resource allocation. Further validation across diverse settings may support its integration into ICU monitoring systems and national pediatric care registries.