Predicting malnutrition in PLWHIV using machine learning in gondar, Ethiopia

Area of Science:

• Artificial Intelligence in Healthcare
• Public Health Informatics
• Nutritional Epidemiology

Background:

• Human Immunodeficiency Virus (HIV) remains a global health crisis, disproportionately affecting Sub-Saharan Africa.
• Malnutrition is a common comorbidity in people living with HIV (PLWHIV), worsening immunosuppression and disease progression.
• This study investigates the use of machine learning (ML) to assess nutritional status and predict malnutrition risk in PLWHIV.

Purpose of the Study:

• To apply machine learning models for assessing nutritional status in PLWHIV.
• To predict the risk of malnutrition among PLWHIV using ML techniques.
• To evaluate the performance of different ML models in identifying malnutrition risk factors.

Main Methods:

• A quantitative, cross-sectional study was conducted with 4,152 PLWHIV attending ART clinics at the University of Gondar Comprehensive and Specialized Hospital, Ethiopia.
• Data included demographic, clinical, hematological, immunological, and treatment-related factors. Preprocessing involved imputation, encoding, and dimensionality reduction.
• ML models were trained (80:20 split) and evaluated using accuracy, precision, recall, F1 score, and AUC, with Synthetic Minority Oversampling Technique (SMOTE) applied to enhance performance.

Main Results:

• The majority of participants were aged 48-57 years (32.9%), female (59.5%), and urban dwellers (76.5%).
• Nutritional status: 62.8% normal BMI, 17.6% overweight, 15.2% underweight, 4.4% obese.
• A Support Vector Machine (SVM) model, enhanced with SMOTE, achieved the highest performance (80.1% accuracy, 80.4% precision, 80.1% recall, 79.4% F1 score, 0.92 AUC). Key predictors included ART duration, BMI, treatment adherence, and WHO stage.

Conclusions:

• Machine learning models provide a powerful method for predicting malnutrition risk in PLWHIV.
• Integrating ML tools into routine care, particularly electronic medical records, can significantly improve nutritional management in resource-limited settings.
• Further research is recommended to validate these findings and optimize the clinical deployment of ML models for malnutrition prediction.