Enterocutaneous Fistula–Associated Sepsis and Mortality: Development and Validation of a Multimodal Artificial Intelligence Prediction Model - Report - MDSpire
Advertisement
Enterocutaneous Fistula–Associated Sepsis and Mortality: Development and Validation of a Multimodal Artificial Intelligence Prediction Model
Multimodal AI Model Predicts Sepsis and Mortality in Enterocutaneous Fistula Patients
Overview
This study developed and validated a multimodal artificial intelligence framework integrating clinical data, imaging, and transcriptomics to predict sepsis risk and 28-day mortality in patients with enterocutaneous fistula (ECF). The model demonstrated enhanced predictive accuracy and interpretability compared to existing approaches, identifying key immune dysregulation features and potential therapeutic targets.
Background
Enterocutaneous fistula is a severe complication characterized by intestinal content leakage, often leading to complicated intraabdominal infections and sepsis with mortality rates up to 50%. Current clinical protocols lack effective tools for early risk assessment, resulting in delayed intervention. Immune dysregulation plays a central role in disease progression, but existing predictive models rely mainly on clinical scores without integrating molecular or imaging data. Advances in artificial intelligence, particularly multimodal fusion models, offer promising avenues to improve prediction and mechanistic understanding in complex diseases like ECF-associated sepsis.
Data Highlights
Data Modality
Algorithm
Primary Strengths
Limitations
Relevance to Study
Clinical, RNA expression, CT/MRI imaging
XGBoost, CNN, VAE, Transformer fusion
High predictive accuracy, mechanistic insight, interpretability
Requires multimodal data, transcriptomics may be unavailable in some settings
Customized for ECF-related sepsis; modular design allows dual-modal use
Single-modality clinical or transcriptomic data (prior studies)
Various machine learning models
Simple implementation
Limited accuracy, lack of mechanistic integration
Baseline for comparison; highlights need for multimodal approach
Key Findings
The multimodal AI model effectively integrates longitudinal clinical data, abdominal imaging, and transcriptomic profiles to predict sepsis and 28-day mortality in ECF patients.
Incorporation of immune-related transcriptomic features reveals immune dysregulation patterns linked to sepsis progression and mortality risk.
The model’s modular architecture allows a dual-modal variant (clinical + imaging) to maintain predictive performance when transcriptomic data are unavailable.
Compared to existing sepsis prediction models, this framework offers superior specificity for ECF-related infections and improved early risk stratification.
Clinical Implications
This multimodal AI framework provides clinicians with a robust tool for early identification of ECF patients at high risk of sepsis and mortality, enabling timely and targeted interventions. Its adaptable design facilitates integration into diverse clinical settings, even when molecular data are limited. Furthermore, the identification of immune dysregulation signatures and druggable targets may guide personalized treatment approaches and improve patient outcomes.
Conclusion
The study presents a novel, interpretable multimodal AI model that advances prediction and mechanistic insight into sepsis and mortality in enterocutaneous fistula patients. This approach holds promise for enhancing clinical decision-making and informing precision therapies in complex intraabdominal infections.
References
Predicting Sepsis and Mortality in Enterocutaneous Fistula Patients: A Multimodal Artificial Intelligence Model Development and Validation
