Forecasting HIV Treatment loss of follow up rate Using Deep Learning and Survival Analysis in Ho Chi Minh City

Maintaining antiretroviral therapy (ART) is critical for improving treatment outcomes and preventing thetransmission of HIV. However, dropout rates remain a significant challenge, particularly in urban centers like HoChi Minh City, which has the largest population of ART patients in Vietnam. This study aimed to predict ARTdropout rates using deep learning methods, specifically a one-dimensional convolutional neural network (1D CNN),in combination with survival analysis to identify factors affecting treatment retention. This approach leveragesadvanced machine learning techniques to analyze time-to-event data, offering deeper insights into patient retentiondynamics. The results demonstrate that the 1D CNN model significantly outperformed traditional methods, achievingan F1-score of 0.847 and a ROC AUC of 0.890, compared to significantly lower scores form traditional models. This highlights the model’s ability to efficiently process complex, high-dimensional datasets and identify high-risk patients for timely intervention. The model’s findings indicate retention rates of 52.09% after one year and 27.74% after 1.5 years, with factors such as treatment enrollment type and the presence of comorbidities being identified as significant factors reducing dropout rates. We recommend comprehensive care for patients at treatment facilities as a strategy to retain them in ART services, emphasizing the importance of addressing not only HIV treatment but also coexisting health conditions to enhance overall retention.