Machine learning for early detection of HIV virological failure in resource-limited settings: A descriptive and predictive analysis in Uganda

1. To systematically review and meta-analyze predictors of HIV virological failure among people living with HIV in east africa; 

This objective aimed to consolidate existing research on factors contributing to HIV virological failure in East African populations. The region presents a unique epidemiological and healthcare context, with variations in ART regimens, adherence support mechanisms, and patient demographics. By conducting a systematic review of peer-reviewed literature, the project gathered and categorized known predictors of virological failure, such as baseline CD4 counts, lifestyle dynamics, baseline viral loads, treatment duration and presence of comorbidities. A subsequent meta-analysis quantified the strength of these predictors across different studies, shedding light on the most critical factors that clinicians and policymakers should prioritize. This evidence-based foundation informed both the descriptive and predictive components of the project, ensuring that the study’s analytical framework is grounded in robust, region-specific knowledge.

2. To descriptively characterize demographic and clinical variables of taso patients from 2014–2024;

The second objective focuses on leveraging a retrospective cohort dataset from TASO clinics in Uganda to create a detailed portrait of the patient population over a ten-year period, 2014 to 2024. Key demographic information (age, gender, socioeconomic status) and clinical indicators (CD4 counts, viral load measurements, ART regimen details, comorbidities) are being analyzed to identify trends and patterns in treatment outcomes. Descriptive statistics such as medians, frequencies, and interquartile ranges will inform the overall profile of individuals most susceptible to virological failure. By systematically mapping these variables, the project lays the groundwork for more advanced predictive modeling. Additionally, the descriptive phase may reveal potential data quality issues or missing information that can be addressed prior to model development, thereby strengthening the overall reliability of subsequent analyses.

3. To develop and validate predictive models for HIV virological failure using logistic regression and random forests;

Building on insights from the systematic review and descriptive analyses, the project’s third objective is to create robust predictive models capable of identifying patients at risk of virological failure. Logistic regression will serve as a clear, interpretable baseline, allowing for an understanding of how specific predictors like adherence measures, baseline CD4 counts affect the odds of treatment failure. In parallel, a random forest algorithm will be used to capture more complex, nonlinear interactions among variables. This ensemble method can potentially uncover interactions that a linear model might miss, offering a more fitted prediction of which patients are likely to experience virological failure. Model performance will be evaluated using metrics such as the area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and accuracy. By comparing the results of both models, the research will highlight the trade-offs between interpretability and predictive power. Once validated, the models could guide clinicians in proactively adjusting treatment plans and intensifying patient follow-up, ultimately mitigating the progression of virological failure.