Can Machine Learning Predict Hepatitis in Diabetic Patients?


A study utilizing this technology identified specific predictive factors for developing hepatitis in diabetic patients.

Photo credit: Steve Johnson, Unsplash

Photo credit: Steve Johnson, Unsplash

Hepatitis B and C infections represent a persistent public health challenge and are highly prevalent among individuals with diabetes. In the United States, the American Liver Foundation notes that 2.7-3.9 million people have chronic Hepatitis C (HCV) and there are roughly 17,000 new cases year.1 The Centers for Disease Control and Prevention (CDC) reported that in 2021, there were 2045 new Hepatitis B (HBV) acute infections and 13,300 estimated acute HBV infections.2 In terms of chronic HBV, the CDC reported that in 2021, there were 14,229 newly reported cases and 5.9 newly reported cases of chronic HBV per 100,000 people. 

Moreover, 73% of all acute HBV cases were persons aged 30-59 years. When we consider that 34.2 million individuals in the United States are affected by diabetes mellitus (DM), the impact of HBV and HCV can be highly significant How though, do we predict these infections in those with diabetes?

What You Need to Know

The provided information highlights a significant public health challenge with the prevalence of chronic Hepatitis B and C infections, particularly among individuals with diabetes. The statistics indicate a substantial number of cases in the United States, emphasizing the need for targeted interventions and preventive measures.

The research study discussed underscores the potential of machine learning in predicting hepatitis infections, especially in vulnerable populations such as diabetes patients. The use of various machine learning models, including the LASSO algorithm, demonstrated promising results in identifying potential cases.

The findings also emphasize the broader potential of machine learning and artificial intelligence tools in enhancing medical and public health applications, underscoring the importance of robust testing and evaluation for future deployment.

A new research study published in Scientific Reports sought to address this through machine learning.3 As hepatitis infections can often be asymptomatic, the goal was to try and reduce risk in certain vulnerable populations—specifically diabetes patients—by using various machine learning models to identify potential patients. The potential for machine learning and artificial intelligence (AI) tools to enhance public health and healthcare delivery has increasingly been highlighted as benefits and governance of such tools are evaluated. The authors of this study noted several key contexts—there’s been substantial studies showing accuracy in using machine learning to identify hepatitis infections and one analysis showed accuracy at 99.6% when predicting early diabetes risk through hybrid super ensemble learning models.

To address this question, they utilized several data points—demographics, lipids, questionnaires, etc, to identify a linkage between diabetes and the 12 risk factors for hepatitis. Utilizing the NHANES (National Health and Nutrition Examination Survey) dataset from 2013-2018 that included nearly 30,000 patients, a total of 3210 diabetes cases were identified. Following this, 1396 diabetic patients were included in the study, which used a synthetic minority oversampling technique (SMOTE) within the data, several models were then applied to try and train and then test the models. 

The authors then reported that, “We used a randomized search with ten iterations threefold cross-validation for each of the four models to predict hepatitis. Single machine learning models including RF, XGBoost, support vector machine (SVM), and least absolute shrinkage and selection operator (LASSO) algorithm and stacked ensemble model were created. Sixteen statistically significant clinical parameters were included within these models. The performance comparison results of the four machine-learning methods, both before and after hyperparameter tuning. The LASSO achieved the best accuracy value (0.978) after hyperparameter tuning. The sensitivity of all four models was generally low.”4

Ultimately, the study found that LASSO had the highest predictive performance and in terms of predictive factors for developing hepatitis in diabetic patients, the highest were illicit drug use, poverty, and race. This serves as a good indicator as to the potential applications for machine learning in medical and public health applications, drawing on the need for more quality testing for future deployment.

1.Hepatitis C Information Center. American Liver Foundation. August 18, 2023.–%203.9%20million,each%20year%20in%20the%20U.S.

2.Hepatitis B. Centers for Disease Control and Prevention. August 7, 2023.

3.Kim, S., Park, S. & Lee, H. Machine learning for predicting hepatitis B or C virus infection in diabetic patients. Sci Rep 13, 21518 (2023).

4. Ibid.

Related Videos
© 2024 MJH Life Sciences

All rights reserved.