Optimizing hepatitis B virus screening in the United States using a simple demographics-based model.

Area of Science:

• Epidemiology
• Public Health
• Machine Learning in Healthcare

Background:

• Chronic hepatitis B (CHB) affects over 290 million globally, with only 10% diagnosed.
• A significant gap exists in identifying individuals with hepatitis B virus (HBV) infection.
• Accessible demographic data can be leveraged for improved HBV screening.

Purpose of the Study:

• To develop and validate logistic regression (LR) and machine learning (ML) models for identifying patients with HBV.
• To utilize easily obtainable demographic data for accurate HBV risk assessment.
• To compare the performance of ML models against traditional LR models in HBV detection.

Main Methods:

• Utilized data from 10 cycles (1999-2018) of the National Health and Nutrition Examination Survey.
• Developed and validated LR and random forest ML models using demographic factors (birth year, sex, race/ethnicity, birthplace).
• Compared model performance using area under the receiver operating characteristic (AUROC) values.

Main Results:

• ML model demonstrated superior performance over the LR model in the validation cohort (AUROC 0.83 vs. 0.75).
• Key demographic predictors for HBV infection included birth year, sex, race/ethnicity, and birthplace.
• The ML model effectively differentiated individuals at high and low risk for HBV infection.

Conclusions:

• A machine learning model using demographic data offers a simple and targeted approach for HBV screening.
• This method can help address the diagnostic gap in chronic hepatitis B.
• Demographic data-driven ML models can facilitate efficient HBV screening strategies.