AI-DRIVEN EARLY DETECTION OF UNDIAGNOSED HEPATITIS B IN LOW-INCOME COMMUNITIES

Abstract

Background: Hepatitis B virus (HBV) remains a significant global public health concern, particularly in socioeconomically marginalized populations where early diagnosis is often delayed or missed due to poor healthcare access and limited screening resources. Artificial intelligence (AI) offers promising tools for enhancing disease detection in such underserved settings.

Objective: To evaluate the performance of AI-based screening models in identifying previously undiagnosed chronic HBV cases in low-income communities in Pakistan.

Methods: A cross-sectional study was conducted over eight months in low-income urban communities of Lahore. A total of 500 adults aged 18–65 years, with no prior HBV diagnosis, were recruited using systematic sampling. Sociodemographic data and basic clinical indicators were collected, followed by HBsAg rapid tests and confirmatory ELISA testing. A gradient boosting machine learning model was developed and validated using structured data inputs. Model performance was assessed via sensitivity, specificity, predictive values, accuracy, and area under the receiver operating characteristic (ROC) curve. Logistic regression was used to identify significant predictors of infection.

Results: Among 500 participants, 39 tested positive for HBV (7.8%). The AI model achieved 89.7% sensitivity, 93.1% specificity, and an overall accuracy of 92.6%. The area under the ROC curve was 0.94. Lack of vaccination, low income, and household HBV exposure were significantly associated with undiagnosed HBV status. The model demonstrated strong potential as a community-level screening aid.

Conclusion: AI-based tools can effectively support early identification of undiagnosed HBV cases in low-resource populations. This approach may bridge diagnostic gaps and improve targeted screening in vulnerable settings.