Real-Time Clinical Decision Support Systems (CDSS) for Decompensation Risk Prediction in Cirrhosis: Machine Learning Modeling and Interface Optimization for Point-of-Care Integration

• Billy Elly

  Lautech

  Author

Cirrhosis affects millions worldwide, with acute decompensation events significantly increasing mortality risk. Traditional prognostic scoring systems such as MELD and Child-Pugh, while valuable, demonstrate limited predictive accuracy for decompensation and fail to integrate seamlessly into clinical workflows. This research addresses the gap between advanced machine learning predictive capabilities and practical clinical implementation by developing and validating a real-time CDSS framework for decompensation risk prediction. Using retrospective data from 983 patients, we developed ensemble machine learning models incorporating gradient boosting, random forest, and neural network architectures, achieving an AUC of 0.89 (95% CI: 0.84-0.94) for 6-month decompensation prediction—significantly outperforming MELD (AUC 0.77) and Child-Pugh (AUC 0.78) (p<0.001). The system was designed using human-centered design principles and implemented as a SMART-on-FHIR application to enable seamless EHR integration. Key predictors identified include albumin, bilirubin, INR, sodium, ascites status, and diuretic use. The framework demonstrates that routine clinical variables, when analyzed through ensemble machine learning approaches, can provide superior risk stratification while maintaining interpretability. This research contributes a validated, replicable CDSS architecture that bridges predictive analytics and clinical practice, with implications for personalized cirrhosis management and reduced preventable hospitalizations.

• pdf

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.