Profile of HIV-Infected Women of Childbearing and Prevention of Vertical Transmissions of HIV/AIDS in sub-Saharan Africa: A Machine Learning Approach
No Thumbnail Available
Date
2024-11-25
Authors
Akinsola O.J
Afuwape O.
Akintan P.E.
Ogunsola F.T.
Journal Title
Journal ISSN
Volume Title
Publisher
MIRG
Abstract
It has been documented that 90% of paediatric HIV transmission is from mother-to-child. Despite the milestone achievements of Prevention Mother-to-child (PMTCT) of HIV program, Nigeria currently has the highest burden of mother-to-child transmission globally. Currently, the nation has the second highest burden of paediatric HIV which can be accounted for by the high rate of mother-to-child transmissions which may be attributed to dearth of novel techniques to halt the spread of the transmission of this pandemic disease. This study utilized an array of different top features machine learning techniques to investigate and predict HIV/AIDS transmission from mother-to-child to suppress the scourge of mortality and other associated outcomes.
An analytical retrospective cohort study of HIV-infected expectant mothers and their infants receiving care in the antiretroviral clinic of LUTH, Nigeria was conducted on 600 HIV-infected women in July 2023. The variables were captured in the National PMTCT maternal cohort register. This include using two polymerase chain reaction (PCR) test due to changing program protocol over the years. The proposed technique was assessed by utilizing the algorithms for the prediction of HIV/AIDS transmission from mother-to-child. Four machine learning algorithms which was splitted into ratio 80:20 for training and testing are: Logistic Regression (LR), Decision Tree (DT), Random Forest (RF) and Naïve Bayes (NB). Some fitting features were employed for the prevention of transmitting the HIV virus from the mothers to infants.
The mean age of women was 37.4±6.1 years, mean infant gestational age at birth was 38.1±0.5 weeks, mean infant weight at birth was 3.0±0.6 kg. Less than two-third have secondary education (61.2%) while close to half of them were married (47.5%). Infant antiretroviral therapy (χ2=41.367, p<0.001**) and type of delivery (χ2=5.304, p=0.021**) were found to be significantly associated with HIV outcome of infants. Of the top 17 similar features (variables) considered, Random Forest (RF) claims the overall model performance with accuracy (73.3%), precision (72.9%), recall (75.4%) and F1-score (76.7%) while Decision Tree has accuracy (70.0%), precision (72.2%), recall (60.2%) and F1-score (63.7%).
The RF model outperformed the most used learning algorithms for prediction of HIV/AIDS in mother-to-child transmission. This technique can also help with early HIV testing in expectant women and increase the precision of transmission risk evaluations of mother and child. Therefore, healthcare providers can utilize machine learning innovations to assist them make well-informed decisions and allocate scare resources efficiently in Nigeria and other sub-Saharan African countries.
Description
Scholarly article
Keywords
Citation
Akinsola O.J., Afuwape O., Akintan P.E. & Ogunsola F.T. (2024). “Profile of HIV-Infected Women of Childbearing and Prevention of Vertical Transmissions of HIV/AIDS in sub-Saharan Africa: A Machine Learning Approach". In Proceedings of the International Conference on Artificial Intelligence and Robotics (MIRG-ICAIR 2024), pp. 85-96, MIRG