JMIR Publications

ABSTRACT

Background:

By predicting PLHIV mortality, medical professionals can take prompt, focused, and proactive steps to reduce risk factors, enhance treatment compliance, and avoid complications. Using real-world data, this study assesses the effectiveness of several algorithms, finds important factors, and investigates the use of machine learning models to forecast death in individuals with HIV.

Objective:

This study aimed to predict mortality among people living with human immunodeficiency virus receiving antiretroviral therapy via machine learning in public health facilities within Gondar City Administration, Northwest Ethiopia, 2024 G.C.

Methods:

This study used data from the electronic medical records of the antiretroviral therapy database. A weighted sample of 12,871 patients was included in the study. The data were cleaned via Excel and SPSS version 27. Python 3.12 was used for machine learning analysis. Furthermore, eight supervised machine learning algorithms were employed to predict and identify the most important predictors of mortality among people living with human immunodeficiency virus.

Results:

The results of extreme gradient boosting were as follows: 99.6% accuracy, 99.9% area under the curve, 99.0% precision, 99.0% recall, and 99.0% F1 score. Random forest and gradient boosting also showed competitive performance. SHAP analysis revealed that the duration of antiretroviral therapy was the most significant predictor of mortality, followed by tuberculosis preventive therapy initiation, body mass index, and adherence to ART.

Conclusions:

This study highlights the potential of machine learning models, particularly ensemble algorithms, for mortality prediction in people living with human immunodeficiency virus. Incorporating these models into clinical practice could enhance patient management and resource allocation. The findings show the need for targeted interventions addressing critical factors such as ART duration, tuberculosis preventive therapy initiation, and adherence to improve survival outcomes. Implement ML models in healthcare, like XGBoost, for mortality prediction among People Living with HIV (PLHIV). This will enhance early detection and timely intervention to reduce HIV-related mortality.