Deep Learning for HIV Screening Using Laboratory and Demographic Data

Abstract

In this work, laboratory and demographic data were integrated to create a deep learning model for HIV screening. The rising incidence of HIV in Indonesia necessitates the development of more effective and precise screening techniques for early identification. The created methodology improves the accuracy of HIV status prediction by integrating many laboratory indicators, including total blood count, viral load, CD4 count, and patient demographic information. For the years 2020–2024, 5,847 patient samples from different Indonesian hospitals made up the dataset. A Deep Neural Network (DNN) architecture with Grid Search hyperparameter optimization was employed in this investigation. According to the evaluation results, the model obtained an F1 score of 93.5%, a sensitivity of 92.8%, a specificity of 95.1%, and an accuracy of 94.2%. When compared to using only laboratory data, the model's performance increased by 3.7% when demographic data was included. This methodology can lessen laboratory burden while assisting medical staff in doing HIV screening more quickly and accurately. An external validation plan has been created with a testing strategy using a separate dataset from ten referral hospitals that were not part of the model training process in order to guarantee the model's dependability in clinical application. To boost the confidence of medical staff, a workable implementation has been created in the form of an API and web application that can be included into the hospital's current information systems and provide an explanation of the prediction results. To help healthcare facilities with different resource levels embrace this technology, technical and clinical implementation recommendations are offered. In order to assess how well the model works to increase HIV detection rates and clinical workflow efficiency, a post-implementation impact evaluation is planned. The efficiency of HIV prevention and control initiatives in Indonesia might be greatly increased by incorporating this paradigm into the healthcare system.