Medical News Observer

Key Points:

• A machine learning model that significantly improves the prediction of hepatocellular carcinoma (HCC) risk in patients with metabolic dysfunction-associated steatotic liver disease (MASLD)
• The Gradient-Boosted Trees model was trained to accurately predict HCC risk, using factors like liver fibrosis, cholesterol levels, and hypertension.
• The research indicates a significant step forward in early HCC risk assessment, offering a foundation for selective screening, risk mitigation strategies, and enhanced patient care efficiency.

A collaborative effort by UC Davis Health clinicians and data scientists has led to developing a machine-learning model to improve the prediction of hepatocellular carcinoma (HCC) risk in patients with metabolic dysfunction-associated steatotic liver disease (MASLD).

The research explores the potential of predictive learning to enhance early risk assessment for hepatocellular carcinoma among MASLD patients.  This innovative approach offers hope for more personalized and proactive patient care.  The journal Gastro Hep Advances published this study.

Current Challenges

The study underscores the difficulty in identifying metabolic dysfunction-associated steatotic liver disease patients at higher risk for HCC due to the disease’s subtle presentation. It highlights the importance of a targeted screening process that could facilitate the early detection and treatment of HCC.

Machine Learning Model for Predicting Hepatocellular Cancer Risk

The team’s work involved training machine-learning algorithms on deidentified health data from MASLD patients, focusing on identifying those most at risk for developing HCC. Their study utilized nine open-source algorithms, eventually narrowing them down to five that showed promise in predicting HCC risk. Researchers used data from UC Davis for model training and an independent dataset from UC San Francisco for validation.

The Gradient-Boosted Trees algorithm emerged as the most accurate model. The results also show that advanced liver fibrosis, as determined by the Fibrosis-4 index (FIB-4) score, is a significant predictor. Other important hepatocellular cancer risk-predicting factors include high cholesterol, hypertension, bilirubin, and alkaline phosphatase levels. Using a validation cohort, the model demonstrated high accuracy (92.06%), specificity (98.34%), and 74.41% sensitivity in predicting HCC development among MASLD patients. 

Implications

This research supports the potential of ML models to facilitate early HCC risk assessment and inform selective screening and risk mitigation strategies. This ultimately contributes to more personalized and cost-effective patient care. In the future, researchers aim to use natural language processing and generative AI to extract data from electronic health records, incorporating it into this predictive model.

References

• Sarkar, Souvik, Aniket Alurwar, Carole Ly, Cindy Piao, Rajiv Donde, Christopher J. Wang, and Frederick J. Meyers. “A Machine Learning Model to Predict Risk for Hepatocellular Carcinoma in Patients with Metabolic Dysfunction-Associated Steatotic Liver Disease.” Gastro Hep Advances, January 23, 2024. https://doi.org/10.1016/j.gastha.2024.01.007.
• “Scikit-Learn: Machine Learning in Python — Scikit-Learn 1.4.1 Documentation.” Accessed March 8, 2024. https://scikit-learn.org/stable/.