CRIT AI 4.0: A Locally Deployable Artificial Intelligence System for Mortality Risk Stratification and Clinical...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• High intensive care unit (ICU) mortality in India necessitates improved risk stratification.
• Resource-limited settings face challenges with cloud-dependent AI decision support systems.
• CRIT AI 4.0 is developed as an offline, locally deployable solution for Indian ICUs.

Purpose of the Study:

• To describe CRIT AI 4.0, an offline AI-assisted decision support system.
• To evaluate the feasibility of AI in resource-constrained Indian ICUs.
• To provide a proof-of-concept for accessible AI healthcare solutions.

Main Methods:

• Developed a Python-based, offline Windows application.
• Trained a random forest classifier on a synthetic dataset (16,000 encounters, 17 disease categories, 37 variables).
• Integrated clinical scoring modules, fluid balance tracking, nursing documentation, and an on-device large language model (LLM) assistant.

Main Results:

• The model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.90 on a synthetic test set (n=3,200).
• Demonstrated high sensitivity (0.86) and specificity (0.77) in synthetic data analysis.
• All performance metrics are based solely on the synthetic test dataset.

Conclusions:

• CRIT AI 4.0 proves the technical feasibility of offline AI-augmented ICU decision support.
• The system is designed for resource-constrained environments.
• Prospective validation studies are currently in progress.