Online Program

Important factors to predict Anemia during the treatment of Malaria in HIV-infected population (309840)

Victor Mwapasa, University of Malawi, College of Medicine. 
*Yein Jeon, Georgetown University 

Keywords: Machine learning, Public health, HIV, Malaria, Epidemiology

There are geographical areas where high malaria prevalence and HIV prevalence overlaps. In these areas, HIV-positive patients, who are already receiving Antiretroviral Therapy(ART), take Artemisinin-based Combination Therapies (ACTs) for malaria treatment. With these circumstances, the assessment of the drug-drug interactions is necessary due to toxicities. This study focuses on the risk of anemia after taking both ART and ACTs treatment. The objective of this study is to identify the factors which need special attention during the treatment and prevent Anemia. For this study, the data is obtained from the previous clinical trial: the pre-screening survey, enrollment data, and the laboratory results. From the clinical trial data, the anemia emergence criteria are; a) The laboratory result of Hemoglobin below 8.5g/dL between day 3 and day 63 of ACTs treatment b) Hemoglobin above 8.5g/dL on day 0 of ACTs treatment. Due to the imbalance of the number of anemia cases and normal cases, the over-sampling method is added to better train the model. In this study, Synthetic Minority Oversampling (SMOTE), which selects the minority data randomly and duplicates them by calculating the k-nearest neighbors (K-NN), is used. Additional data points of minority, which is the anemia case in this study, are created having similar characteristics with the existing case data points. Among the different machine learning methods, the Naive Bayes classifier showed notable results with 71.97% accuracy and 0.7176 F1 score. The permutation Importance is calculated for additional feature engineering. Based on the result, the adjusted model showed improved accuracy and the F1 score: 73.48% and 0.7482. By using SHapley Additive exPlanations (SHAP), the feature importance of each feature is calculated. Eventually, it is found that the factors like Eosinophils level and Platelet count need the careful attention of clinicians and physicians when treating Malaria in HIV-infected population.