Prognostic Modeling for Liver Cirrhosis Mortality Prediction and Real-Time Health Monitoring from Electronic Health...

Area of Science:

• Hepatology and Gastroenterology
• Artificial Intelligence in Medicine
• Biostatistics and Predictive Modeling

Background:

• Liver cirrhosis is a leading cause of mortality, affecting millions globally.
• Accurate prediction of mortality in cirrhosis patients is essential for improving treatment outcomes.
• Current prognostic tools, like the Model for End-Stage Liver Disease (MELD) score, have limitations.

Purpose of the Study:

• To investigate the correlation between laboratory test results, diagnoses, and mortality in liver cirrhosis patients.
• To evaluate the performance of a deep learning (artificial neural network) model for predicting cirrhosis mortality.
• To compare the predictive accuracy of the deep learning model against the existing MELD score.

Main Methods:

• An artificial neural network model was developed for liver disease classification.
• The model's performance was assessed using different training dataset sizes (70%, 80%, 90%).
• Evaluation metrics included precision, recall, F1-score, accuracy, Area Under the Curve (AUC) from ROC curves, and PR curves.

Main Results:

• Model performance generally improved with larger training datasets.
• The model trained with 80% of the data achieved the highest AUC, indicating superior classification ability.
• Precision-recall analysis showed an 80% training dataset offered a slightly better balance between precision and recall, despite inherent challenges in imbalanced medical datasets.

Conclusions:

• Deep learning models show potential for enhancing prognostic accuracy in liver cirrhosis.
• An 80% training dataset size appears optimal for this specific artificial neural network model.
• Further research is needed to refine models for predicting cirrhosis mortality, especially with imbalanced data.