Development, Validation and Deployment of a Novel Prediction Model Utilizing Artificial Intelligence on Clinical and Proteomic Features to Predict Mortality among Patients with Acute-on-Chronic Liver Failure

Context & Background

Acute-on-Chronic Liver Failure (ACLF) is a rapid-onset clinical syndrome with a mortality rate ranging from 15% to 100%. Timely prediction of mortality is essential to allocate intensive care unit (ICU) resources, ventilators, and fast-track patients for liver transplants.

Problems to be Addressed

Existing clinical scores are based purely on basic vitals, neglecting proteomic and biomarkers that indicate organ failure. Clinicians lack automated bedside tools to predict patient mortality risk in real-time.

Aims and Objectives

1. Create a clinical and proteomic database of ACLF patients.
2. Develop machine learning models to predict mortality within 30 days.
3. Build a mobile/web application for real-time bedside risk prediction.

Methodology

The study prospectively collects clinical and plasma proteomic data from 100 patients, with retrospective records of 1000 patients. Altered proteins are identified and validated using ELISA assays. Supervised ML and deep learning algorithms are trained to calculate risk scores, which are integrated into a bedside web app.

Expected Outcomes

A bedside prognostic tool for liver disease mortality, clinical databases mapping ACLF proteomics in India, and peer-reviewed journals in hepatology.