Explainable Machine Learning for Classification of HIV Viral Load Suppression in Resource- Limited Settings

Explainable Machine Learning for Classification of HIV Viral Load Suppression in Resource- Limited Settings | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Explainable Machine Learning for Classification of HIV Viral Load Suppression in Resource- Limited Settings Abraham Keffale Mengistu, Aynadis Worku Shime, Muluken Belachew Mengistie, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8799312/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Effective viral load (VL) monitoring is crucial in the management of HIV care, but is difficult in resource-constrained settings due to limited access to laboratory examinations. Machine learning (ML) has a promising approach to viral load suppression (VLS) prediction using normal clinical information. This study aimed to develop and interpret an ML model for VLS classification among an Ethiopian cohort. Methods A retrospective analysis was undertaken with electronic medical records of 4,152 patients on antiretroviral therapy (ART) in the University of Gondar Comprehensive Specialized Hospital. Eight ML algorithms, namely Logistic Regression, Random Forest, and Gradient Boosting, were trained and optimized to classify a binary VLS outcome. Model performance was assessed based on accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). The best-performing model was interpreted with SHapley Additive exPlanations (SHAP) to identify the significant predictors and their sign of impact. Results The best-performing Gradient Boosting model performed the best with 76% accuracy, 0.74 F1-score, and 0.79 AUC-ROC. Baseline CD4 Category and Duration on ART in Months were identified as the most impactful predictors through feature importance evaluation. SHAP analysis supported that longer ART duration and larger baseline CD4 count were associated with increased odds of VLS, and that higher WHO clinical stage and male sex were associated with unsuppressed VL. The model's decision-making was further depicted for individual patients by waterfall plots, which enhanced clinical interpretability. Conclusion This work demonstrates that one can have an interpretable Gradient Boosting model to properly predict viral load suppression in a low-resource setting. The predictions of the model are made from clinically reasonable factors, linking algorithmic performance to corresponding clinical insight. The tool can potentially assist healthcare workers in identifying patients at risk of treatment failure, enabling the implementation of early interventions and optimizing HIV care management in settings where routine VL testing is not feasible. HIV Viral Load Suppression Machine Learning Predictive Modeling SHAP Low-Resource Settings Clinical Decision Support Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Background Viral load measurement is a critical parameter in the management of infectious diseases such as HIV, hepatitis, and COVID-19 (Coronavirus Disease 2019) [ 1 ], [ 2 ]. It quantifies the virus in a patient’s blood, aiding diagnosis, monitoring disease burden, and assessing treatment efficacy [ 3 , 4 ]. Consequently, the prediction of viral load is of high priority in low-resource settings where access to frequent laboratory testing may be poor [ 5 ], [ 6 ], [ 7 ]. In such contexts, predictive models can serve as valuable tools for estimating viral load levels, enabling healthcare providers to make appropriate treatment plans that improve patient outcomes [ 8 ], [ 9 ], [ 10 ]. HIV/AIDS (Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome) remains one of the most significant global health challenges, with an estimated 39 million people living with the disease worldwide as of 2022, including 1.5 million new infections annually [ 11 ]. Sub-Saharan Africa bears the most significant burden, accounting for approximately 67% of all people living with HIV and 72% of AIDS-related deaths globally [ 12 ]. In Ethiopia, over 500,000 people are living with HIV, and the epidemic remains a leading cause of morbidity and mortality despite expanded access to antiretroviral therapy [ 8 , 13 ]. These stark disparities highlight the urgency of optimizing strategies for viral load monitoring and treatment management in resource-limited settings. The global burden of HIV/AIDS, particularly in sub-Saharan Africa, which accounts for the majority of new infections and AIDS-related deaths, remains a major public health challenge[ 11 ], [ 14 ]. Ethiopia, one of the most affected countries, has made great progress in expanding access to ART through its health system. However, there are a lot of challenges, such as not enough health infrastructures, unequally distributed resources, and the impossibility of monitoring patients regularly [ 13 ]. In this context, novel approaches to the management of diseases, including the use of AI (Artificial Intelligence) and ML, could improve the efficiency and effectiveness of HIV care [ 15 – 16 ]. In recent years, ML has become a subarea of AI that has grown as one of the strongest predictive modeling tools in healthcare. The ML algorithms can identify patterns and relationships from large datasets that cannot easily be determined by traditional statistical methods [ 17 ], [ 18 ], [ 19 ]. This capability is particularly relevant in the context of viral load prediction, where multiple factors interact in complex ways [ 15 ]. However, the performance of ML models depends heavily on the quality of the data, the choice of algorithms, and the strategies employed to address common challenges such as class imbalance and missing data [ 17 ], [ 19 ]. In many LMICs (Low- and Middle-Income Countries), including Ethiopia, data quality challenges are pervasive and multifaceted[ 20 ], [ 21 ]. At the University of Gondar Comprehensive and Specialized Hospital, missing or incomplete data are common in both demographic and clinical fields. Examples include missing age or residence information, inconsistent documentation of ART initiation dates, and incomplete CD4 or viral load test results. Additionally, manual data entry errors and the lack of standardized formats in the EMR system contribute to inaccuracies in medical records. These issues hinder the robustness of ML models, which rely heavily on clean, complete datasets[ 22 ]. Therefore, careful preprocessing, imputation strategies, and quality control mechanisms are necessary to mitigate the effects of poor data quality and improve model reliability. Gondar University Comprehensive and Specialized Hospital is also a very active hub for the care and study of HIV-infected individuals located in the Amhara National Regional State. The ART Clinic, established in 2005, provides free services. The clinic started its work of enrollment in August 2005 and has accrued more than 15,000 patients, with more than 5,000 patients on active follow-up. This will be the source for developing and validating predictive models since a lot of data related to patient demographics, clinical history, immunological status, and treatment outcomes is stored in the EMR (Electronic Medical Records) system. Using this, researchers can derive knowledge concerning the factors associated with viral load suppression and develop tools to support clinical decision-making. Even though ML is a very promising tool for viral load prediction, there are only a few comprehensive studies that compare different algorithms and imbalance resolution techniques on the subject. Most of the current literature is from high-income settings, where data availability and healthcare infrastructure differ significantly from those in LMICs [ 15 , 16 ]. This literature gap underlines the necessity for studies investigating the performance of ML models under resource-constrained settings where the challenges of poor data quality and class imbalance are mostly heightened. This study will, therefore, seek to close this gap through a comparative performance analysis of the eight conventionally used machine learning algorithms, namely, Random Forest, Support Vector Machines, Gradient Boosting, Decision Tree, LightGBM, XGBoost, Logistic Regression, and k-nearest Neighbors on viral load prediction. The findings of this study have important implications for public health practice and policy. Accurate viral load prediction can support the timely identification of patients at risk of treatment failure, enabling healthcare providers to intervene early and prevent adverse outcomes [ 23 ]. Beyond this, the use of ML models reduces reliance on expensive and time-consuming laboratory tests, increasing access to and the sustainability of HIV care in resource-constrained settings [ 24 ]. The study will optimize AI models for viral load prediction to further advance the global effort to combat HIV/AIDS and improve the quality of life for individuals living with the virus. This study advances previous research by comprehensively comparing multiple machine learning models using a balanced dataset within a real-world LMIC setting. Unlike prior studies that may have been constrained by imbalanced data or an over-reliance on traditional statistical methods, the use of a balanced dataset in this work ensures a robust and unbiased evaluation of model performance. Furthermore, moving beyond a primary focus on accuracy, this research places significant emphasis on explainability through SHAP values, ensuring that the models are not only predictive but also interpretable and clinically relevant. In conclusion, the integration of AI and ML into healthcare represents a promising avenue for addressing some of the most pressing challenges in infectious disease management [ 25 ], [ 26 ]. This study further extends the literature by providing a state-of-the-art comparison of ML techniques in predicting viral load suppression status in a resource-limited setting. Using data from the University of Gondar Comprehensive and Specialized Hospital, we strive to develop a strong and generalizable model that will support clinical decision-making and add to the global fight against HIV/AIDS. Methods and Materials Study Design and Setting A quantitative research approach has been adopted in this study, where machine learning techniques will be employed to predict the viral load status among PLHIV (People Living with Human Immunodeficiency Virus). This research was conducted at the University of Gondar Comprehensive and Specialized Hospital, one of the leading healthcare institutions in Gondar City, Amhara National Regional State, Ethiopia. Gondar City, located about 748 km northwest of Addis Ababa, has a population of 457,938 and is one of the largest hubs for health service delivery and research. This ART clinic was established in March 2005 and provides free ART services to over 7 million people in Gondar province and surrounding regions. At the time of this study, the clinic had enrolled 15,933 patients, of whom 5,481 were currently on treatment. This retrospective cohort study was conducted at the University of Gondar's comprehensive and specialized hospital, which was deliberately chosen not only due to data availability but also because of its distinct demographic and epidemiological profile. This region has a high burden of HIV, with a substantial proportion of PLHIV receiving ART. Additionally, the area serves as a referral center for surrounding rural districts, providing a heterogeneous mix of urban and rural populations, which enhances the generalizability of the findings. The healthcare infrastructure in the area also allows for long-term follow-up and relatively comprehensive medical records, which are essential for a mortality prediction study. Data Source and Study Population The secondary data in this study were obtained from the EMR-ART at the University of Gondar Comprehensive and Specialized Hospital. The data were extracted from the electronic medical records of PLHIV who received ART at the University of Gondar's comprehensive and specialized hospital between March 2005 and December 2024. Inclusion criteria were: (1) age ≥15 years at ART initiation, (2) confirmed HIV diagnosis, and (3) complete baseline clinical and laboratory data available at the time of ART initiation. Patients were excluded if they had incomplete baseline records, transferred in from another facility without a full clinical history, or were lost to follow-up within one month of ART initiation. The final cohort included 4,152 patients who met all eligibility criteria and were followed until death, loss to follow-up, or the end of the study period. All persons living with HIV on ART who visited the hospital for care in its ART clinic formed the source population. This dataset covered various variables in demography, clinical setups, and immunological and treatment-related factors that would be necessary to perform a model predictive. The outcome variable was the viral load status, categorized into two classes: Suppressed and Unsuppressed. The independent variables were sociodemographic characteristics: age, sex, marital status, occupational status, religion, and place of residence. Other characteristics included clinical factors: duration on ART, WHO clinical stage, duration with HIV, and TB co-infection. Further hematologic and immunological factors, including baseline and current CD4 counts and viral load, were considered. Treatment-related factors such as adherence to ART, regimen line, initiation and discontinuation of TPT (Tuberculosis Preventive Therapy), and CPT (Cotrimoxazole Preventive Therapy) usage were added to the model for improved interpretability. Data Preprocessing The data was preprocessed, following a structured data preprocessing pipeline to ensure the quality and reliability of the data. Machine learning requires a high-quality dataset for prediction. Due to this, handling the missing data during the pre-processing of the dataset is a crucial phase. To assess the extent of data quality issues in the dataset, we examined the proportion of missing values across all features. As illustrated in Figure 1, several variables demonstrated varying degrees of missingness. Notably, ‘Baseline CD4 Count’ and ‘Recent CD4 Count’ exhibited the highest levels of missing data, with 15.2% and 15.0% of values missing, respectively. Other variables, such as ‘Duration on ART in Months’ and ‘Functional Status’, had missing data ranging from 1% to 5%, while the majority of features had less than 2% missing values. Overall, the dataset contained approximately 1.7% missing data. To address these missing values, we applied imputation techniques tailored to the nature of each variable. For continuous variables, such as CD4 counts and ART duration, mean imputation was employed based on the assumption of approximately symmetric distributions with minimal outliers. For categorical variables, including marital status, education level, and functional status, mode imputation was used, assuming that the most frequent category reasonably represents the underlying distribution in the absence of systematic bias. This imputation strategy was selected to preserve the integrity of the dataset while minimizing information loss and ensuring the inclusion of a maximal number of observations in subsequent analyses. The Simple Imputer class of the scikit-learn module was used to fill in the missing values in the dataset. Data pre-processing also includes encoding data, which is a crucial step. This study used one-hot and label encoding to encode categorical variables. Values with two or more category values are considered categorical if they are discrete and not continuous. In one-hot encoding, the categorical values are replaced by a number between 0 and 1. Feature Selection From the EMR, we extract all relevant features and do correlation analysis to determine if there is a strongly correlated feature that will affect the model performance and introduce biases. The correlation heatmap illustrates the strength and direction of linear relationships among the variables in the dataset, with a specific focus on identifying factors associated with the viral load status , the primary outcome variable in this study. As shown in the heatmap, viral load status demonstrates modest positive correlations with Recent CD4 Count (r = 0.35), Recent_CD4_Category (r = 0.35), and Baseline CD4 Count (r = 0.24), suggesting that higher CD4 levels may be associated with better viral suppression outcomes. Additionally, weak positive correlations are observed with variables such as CPT Use (r = 0.05) and Duration on ART in Months (r = 0.11), indicating potential clinical relevance. Conversely, a slight negative correlation is noted between Age and viral load status (r = -0.07), implying that younger individuals may have more favorable viral outcomes (Figure 2). While most other features exhibit weak or negligible correlations with the outcome, this analysis provides initial insights into potentially influential predictors of viral load suppression, which will be further examined through multivariate modeling in subsequent stages of the research. Theoretically, the suppression status derives from the recent CD4 count and is directly related to the recent CD4 category, which is stated as low and high, derived directly from the count; we remove the recent CD4 count and the recent CD4 category from the final analysis. Data Balancing In this dataset, the dependent variable (viral load status) was almost evenly distributed, with suppressed cases representing 50.4% and unsuppressed cases 49.6% (Fig. 3). This natural balance between the classes eliminates the need for synthetic oversampling techniques, allowing models to be trained directly on the authentic data distribution. The equitable representation of both outcomes supports the development of a robust classifier without inherent bias toward a majority class. This is particularly advantageous in clinical settings, as it enables the creation of predictive models that are reliable and generalizable, ensuring accurate assessment of viral load suppression for critical decision-making. Machine Learning Models For the estimation of viral load status, we evaluated eight ML algorithms, comprising both linear and ensemble-based models. Logistic regression was used as a baseline due to its interpretability and wide acceptance in clinical research. Ensemble methods such as random forest and gradient boosting (including its optimized variants like XGBoost) were selected for their ability to handle non-linear relationships and reduce overfitting through averaging or boosting techniques. Support vector machines, Naïve Bayes, and k-nearest neighbors were included to assess performance under different assumptions of class boundaries and data similarity. Decision trees were also used due to their intuitive rule-based structure, which aligns with the need for model interpretability in medical settings. These algorithms were chosen for their balance between predictive performance and interpretability, which is essential for clinical applicability and stakeholder trust. Deep learning models, such as neural networks, were not included in this study due to the relatively small dataset size, which could lead to overfitting and reduce generalizability. Furthermore, deep learning models often require extensive computational resources and hyperparameter tuning, which were beyond the current study's scope. Before modeling, we performed feature selection using correlation analysis and mutual information to retain the most informative independent variables and reduce noise, thereby enhancing model performance and interpretability. Model Training and Optimization The computational environment used for model training and optimization consisted of an Intel(R) Core (TM) i7-8650U CPU running at 1.90 GHz, with 16 GB of installed RAM on a 64-bit Windows operating system. On average, model training and hyperparameter tuning took approximately 3 to 8 minutes per algorithm, depending on the model complexity (e.g., logistic regression and decision tree models converged faster, while ensemble methods like XGBoost and Gradient Boosting required longer due to iterative boosting). The dataset was split into training and testing subsets, with 80% allocated for training and 20% for testing. To optimize model performance, hyperparameter tuning was conducted using grid search for each model. For logistic regression, tuning focused on regularization strength (C), explored over a logarithmic scale from 10 -4 to 10 3 and penalty type (both L1 and L2, which are for Lasso and Ridge regularization, respectively), while for Random Forest, these were several trees (between 50 to 500), maximum depth between 5 to 50, minimum samples per leaf from 1 to 10. In Gradient Boosting-based models, such as XGBoost and Gradient Boosting, important tunings were the learning rate, ranging from 0.001 to 0.3; estimators ranging from 50 to 500; and maximum depth between 3 to 15. For support vector machines (SVM), tuning was performed on kernel type (linear, polynomial, and RBF (Radial Basis Function), regularization parameter (C) ranging from 10 -3 to 10 2 , and gamma values between 10 -4 to 10 3 and decision trees were fine-tuned by varying max depth (3 to 20), minimum samples split (2 to 10), and evaluating both Gini impurity and entropy criteria. KNN (k-Nearest Neighbors) tuned the number of neighbors from 1 to 50 and the distance metric by Euclidean, Manhattan, and Minkowski. Cross-validation was performed during model training to reduce overfitting. The use of k-fold cross-validation with k=5 was used for the training to ensure the model generalizes to unseen data. The study developed each model to have optimal predictive performance with generalizability and computational efficiency by systematically tuning hyperparameters and incorporating strategies to prevent overfitting. Evaluation Metrics Model performance was evaluated using a suite of metrics selected for clinical relevance and robustness. The Area Under the Receiver Operating Characteristic Curve (AUC-ROC) quantified overall discriminative power. Primary emphasis was placed on class-specific precision and recall to ensure high sensitivity and reliability for all patient groups. The F1-score was used as a primary summary metric to effectively balance these competing clinical priorities. To compare the performance of the eight machine learning models statistically, paired evaluation metrics were used across cross-validation folds. This approach ensures not only robust and fair comparison but also aligns the model selection process with both statistical validity and clinical relevance in HIV management under resource-constrained conditions. Overall, the process involves collecting EMR data, preprocessing it, selecting and training models, optimizing performance, and evaluating results to identify the best model (Fig. 4). Implementation All of the models were implemented in the Python programming language using the machine learning libraries. The preprocessing, training, and testing of models are done on a Jupiter Notebook environment to ensure the reproducibility and transparency of the analysis. Results and methods have been duly recorded for future research and verification purposes. Ethical Considerations Ethical approval for this study was obtained from the Institutional Review Board, Debre Markos University. Since all data are anonymized before the analysis, patient confidentiality is strictly assured. Even though these data are secondary, access to the data set was restricted to authorized researchers only to comply with ethical guidelines and data protection policies. Results Descriptive Statistics The dataset consisted of 4,152 patients, with 50.4% having suppressed viral load and 49.6 % having unsuppressed viral load. The mean age of participants was 46.72 years (SD = 11.38), with the majority being female (59.51%) (Table 1). Table 1: Socio-demographic variables Variable Category Frequency Percentage Sex F 2471 59.51 M 1681 40.49 Marital Status Married 2700 65.03 Divorced 698 16.81 Never Married 437 10.53 Widowed 317 7.63 Education level Secondary Education 2145 51.66 No Education 805 19.39 Primary Education 789 19.00 Higher Education 413 9.95 Residence Urban 3177 76.52 Rural 975 23.48 Religion Orthodox 3910 94.17 Muslim 206 4.96 Protestant 26 0.63 Catholic 9 0.22 Other 1 0.02 Regarding clinical characteristics, the median duration of ART was approximately 12.7 years (152.5 months), and most patients were in WHO clinical stage 4. The dataset exhibited a balanced class, with suppressed viral load being the predominant class (50.4%), and unsuppressed (49.6 %) viral load categories (Table 2). Table 2:Clinical-related variables Variable Category Frequency Percentage BMI Normal 2607 62.79 Overweight 733 17.65 Underweight 630 15.17 Obese 182 4.38 TB screening Result No 4070 98.03 Yes 82 1.97 Functional Status Working 3861 92.99 Ambulatory 163 3.93 Bedridden 128 3.08 Viral Load Status Suppressed 2092 50.4 Unsuppressed 2060 49.6 Adherence Good 3611 86.97 Poor 355 8.55 Fair 186 4.48 Regimen Line First Line 3811 91.79 Second Line 319 7.68 Third Line 22 0.53 TPT Started Yes 2958 71.24 No 1194 28.76 CPT Use Yes 3313 79.79 CPT Use No 839 20.21 Model Training Results In this study, the comparative evaluation of machine learning models revealed clear distinctions in performance across the selected algorithms. Gradient Boosting and Random Forest emerged as the top-performing models, achieving accuracies of 0.73 and 0.72, respectively, with consistently balanced precision, recall, and F1 scores (all around 0.70). This highlights their robustness and reliability in classifying the data effectively. Naive Bayes stood out for its exceptionally high recall (0.84), suggesting a strong ability to identify positive cases, albeit at the expense of lower precision (0.59), which may increase false positives. Decision Tree and XGBoost offered stable mid-range performance, with accuracies of 0.67 and 0.69 and balanced metrics, making them reasonable alternatives. Logistic Regression yielded moderate results (accuracy = 0.62), while SVM and KNN underperformed, both with accuracies of 0.54 and comparatively lower overall scores (Table 3). Collectively, these findings indicate that ensemble-based methods, particularly Gradient Boosting and Random Forest, are the most effective approaches for this dataset, providing a promising direction for building accurate and reliable predictive models. Table 3: Model Training Results Model Accuracy Precision Recall F1 Score Logistic Regression 0.62 0.60 0.55 0.58 Random Forest 0.72 0.70 0.70 0.70 Gradient Boosting 0.73 0.70 0.70 0.70 Naive Bayes 0.66 0.59 0.84 0.69 SVM 0.54 0.51 0.55 0.53 KNN 0.54 0.50 0.58 0.54 Decision Tree 0.67 0.65 0.62 0.64 XGBoost 0.69 0.66 0.67 0.66 The ROC curves shown in Figure 5 illustrate the trade-off between the true positive rate and false positive rate for all models evaluated. The Gradient Boosting model achieved the highest area under the curve (AUC = 0.79), closely followed by Random Forest (AUC = 0.78) and XGBoost (AUC = 0.76), confirming their strong discriminative ability. Naive Bayes also performed well, with an AUC of 0.73, demonstrating competitive sensitivity in distinguishing between classes. In contrast, Logistic Regression (AUC = 0.66) and Decision Tree (AUC = 0.67) provided moderate results, while SVM (AUC = 0.57) and KNN (AUC = 0.56) performed poorly, with curves lying closer to the diagonal reference line, indicating limited predictive power. Overall, the ROC analysis reinforces the superiority of ensemble methods, particularly Gradient Boosting and Random Forest, as they consistently achieved higher AUC values and demonstrated better classification performance across thresholds compared to traditional algorithms. The Gradient Boosting model, identified as the best-performing algorithm in the initial evaluation, was further optimized through hyperparameter tuning using GridSearchCV . After tuning, the optimized Gradient Boosting classifier achieved an accuracy of 0.76 , with a precision of 0.73 , a recall of 0.75 , and an F1 score of 0.74 . The confusion matrix (Fig. 6) shows that the model correctly classified 339 suppressed cases and 211 unsuppressed cases, while misclassifying 108 suppressed and 113 unsuppressed instances. These results demonstrate a well-balanced performance between sensitivity and precision, reflecting the model’s reliability in distinguishing between suppressed and unsuppressed viral load statuses. Overall, hyperparameter tuning improved the Gradient Boosting model’s predictive capability, making it the most effective and robust approach among the models tested. Feature Importance The feature importance analysis from the tuned Gradient Boosting model (Fig. 7) highlights the variables that most strongly contributed to predicting viral load suppression status. Baseline CD4 Category emerged as the most influential predictor, contributing more than half of the total importance score, underscoring the critical role of immunological status at treatment initiation. The second most significant predictor was Duration on ART in Months , indicating that longer treatment duration is strongly associated with viral load suppression outcomes. Other features, such as sex and age, contributed moderately, while factors like regimen line , CPT use , and marital status had relatively smaller impacts. Variables including WHO stage, adherence, BMI, educational level, TPT initiation, and TB screening result showed minimal importance in the model. These findings suggest that immunological markers and treatment duration are the most decisive factors in determining viral load suppression, while sociodemographic and clinical background factors play a comparatively limited role in prediction. Model Explainability The interpretation of the model using SHAP revealed that the most important features driving predictions were the duration on ART, baseline CD4 category, and patient age. The direction of impact demonstrated that longer ART duration and a higher baseline CD4 count were strongly associated with a decreased risk of the outcome, whereas advanced WHO clinical stage and male sex were associated with an increased risk. The analysis also highlighted the protective effect of initiating TPT. This feature importance ranking and directional analysis align with established clinical understanding, thereby enhancing the credibility and interpretability of the model's decision-making process (Fig. 8). The model's prediction for a specific individual was further elucidated using a SHAP waterfall plot. The analysis begins with the baseline model output, E[f(X)] = -0.201, representing the average prediction across the dataset. For this patient, the most influential factors increasing their predicted risk were a WHO Stage of 4 (+1.3) and being male (Sex=1, +1.27). Their age of 53 also contributed significantly to a higher risk (+0.86). These factors were partially offset by a protective effect from a longer duration on ART (162 months) , which decreased the risk (-0.41)(Fig. 9). Contributions from other features, such as CPT Use and Baseline CD4 Category , had smaller positive effects. Ultimately, the cumulative effect of all these feature contributions shifted the model's prediction from the population average to a final, higher value for this individual, clearly illustrating how their specific clinical and demographic profile led to the elevated risk score. Discussion This study sought to develop and interpret a machine learning classifier for estimating HIV viral load suppression from regularly collected clinical information in a low-resource environment. The principal finding is that ensemble-based machine learning models, namely Gradient Boosting, can generate stable and clinically sound estimates of viral load status. The best-performing optimized Gradient Boosting model achieved a balanced accuracy of 76%, with superior performance in precision, recall, and F1-score measures. This performance, coupled with a model explainability evaluation via SHAP, is an actionable and interpretable tool consistent with existing clinical evidence, with great potential for the facilitation of HIV care management in resource-limited environments. Higher performance of tree-based ensemble methods like Gradient Boosting and Random Forest aligns with existing literature on healthcare predictive modeling [ 27 ], [ 28 ]. These models are specially adept at making inferences of complex, non-linear interactions between features, characteristic of clinical information, wherein patient outcomes are a function of a complex interaction between demographic, clinical, and treatment-related variables. Less fundamental models like Logistic Regression and KNN were found to perform poorly, as they may not be able to accurately represent such complex relationships. Naive Bayes' high recall, while intriguing, came with low precision, rendering it less desirable for the clinical setting where a false positive equates to wasteful resource use. A second foundation of this work's value is its emphasis on model explainability. Feature importance analysis and SHAP plots provide beyond a "black box" prediction, clinically interpretable information. The observation that Baseline CD4 Category and Duration on ART (in months) were the strongest predictors is well-aligned with established principles of virology and immunology[ 29 ], [ 30 ], [ 31 ]. Higher baseline CD4 levels are in agreement with an intact immune system upon initiation of treatment, and higher duration on ART is indicative of cumulative treatment efficacy and adherence, both of which are established predictors of viral suppression[ 32 ], [ 33 ], [ 34 ]. Moreover, the SHAP summary plot supported the protective direction of these associations: longer ART duration and higher CD4 levels always pushed model prediction away from the unsuppressed class. The findings of male gender and advanced WHO Stage (Stage 4) as risk factors for unsuppressed viral load are also supported by epidemiologic studies that report differential treatment outcomes by stage of disease and gender[ 35 ], [ 36 ]. The plot of a SHAP waterfall on one patient's data is a lovely way to demonstrate the utility of the model for personalized clinical decision-making. For the example given, the model quantified how much the patient's risk factors (WHO Stage 4 disease, male, old age) were offset by the beneficial effect of long ART duration. This granularity of explanation might help clinicians understand the "why" behind a model's risk score and facilitate appropriate interventions. For instance, a patient who has a high-risk score because of poor adherence (a modifiable factor) would be managed differently from one whose risk is largely owing to poor baseline CD4 count (a non-modifiable factor). Incorporation of such a model into EMR systems of low-resource clinic settings can be an early warning system. With automated risk score calculation during the encounter of the patient, the model has the potential to notify clinicians to flag patients at increased risk of viral load suppression for priority counseling regarding adherence, escalated follow-up, or enhanced clinical monitoring before treatment failure. This pre-emptive strategy may make the best of finite healthcare resources and also enhance the overall outcomes of therapy. That the model depends upon variables generally found in EMRs for ART clinics makes it most feasible for deployment without new, expensive testing. Limitations and Future Directions While having supportive results, this study has several limitations. First, the data are derived from a single tertiary hospital in Ethiopia and hence may have limited generalizability to other populations with their own unique demographic and epidemiological characteristics. External verification with datasets from other regions is required. Second, while the dataset was balanced for the outcome variable, it may not have contained all relevant predictors, e.g., certain socio-economic variables, psychosocial stressors, or genetic markers. Future research would be well-advised to incorporate these variables to further enhance predictive power. Third, the model's performance, while good, indicates that there remains some unexplained variance, and viral load suppression must therefore rely, at least in part, on factors yet to be captured in the current feature set. Future studies would need to focus on three areas: 1) Multi-center verification to test the strength and transportability of the model to a variety of sub-Saharan African health settings. 2) Prospective studies of implementation to quantify the real-world impact of deploying the model into clinical practice on patient outcomes and utilization of resources. 3) Exploration of temporal modeling techniques to estimate viral load suppression risk dynamically through time, rather than at a snapshot in time, providing yet another more persuasive tool for long-term management of patients. Conclusion In conclusion, this study successfully developed an interpretable machine learning model for predicting viral load suppression in a low-resource setting. The Gradient Boosting model not only demonstrated superior classification performance but also, through SHAP analysis, provided insights that are clinically coherent and actionable. By bridging the gap between predictive accuracy and interpretability, this research offers a practical and trustworthy decision-support tool that can empower healthcare providers to optimize HIV care, ultimately contributing to the global effort to achieve sustained virological suppression for all people living with HIV. Abbreviations AI: Artificial Intelligence ART: Antiretroviral Therapy AUC-ROC: Area Under the Receiver Operating Characteristic Curve BMI: Body Mass Index CPT: Cotrimoxazole Preventive Therapy EMR: Electronic Medical Record HIV: Human Immunodeficiency Virus KNN: k-Nearest Neighbors LMICs: Low- and Middle-Income Countries ML: Machine Learning PLHIV: People Living with HIV SHAP: SHapley Additive exPlanations SVM: Support Vector Machine TB: Tuberculosis TPT: Tuberculosis Preventive Therapy VL: Viral Load VLS: Viral Load Suppression WHO: World Health Organization XGBoost: Extreme Gradient Boosting Declarations Ethical approval and consent to participate The study was approved by the Institutional Review Board (IRB) of Debre Markos University College of Medicine and Health Sciences. The need for individual consent was waived by the IRB, as the research utilized anonymized and de-identified data from existing records, making it unfeasible to obtain consent from individuals. The study was conducted in accordance with the principles of the Helsinki Declaration and adhered to all relevant national regulations and ethical guidelines. All data were kept confidential, and no personal identifiers were included in the analysis or reporting. The research processes were performed and secured under the appropriate ethical and regulatory standards by national regulations and ethical guidelines. Clinical Trial Number Not applicable Consent for publication Not applicable Availability of data and materials Data will be available upon request from the corresponding author. The code and experiments used for this research are available on the following public GitHub repository. (https://github.com/abrahamekeffale/Viral_suppression_project.git) Funding Funding not applicable Competing interests The authors declare that they have no competing interests Authors’ contributions A.K.M. conceptualized the study, and A.E.G. was involved in design, analysis, interpretation, report, and manuscript writing. A.W.S. and M.B.M. edited the manuscript for clarity and correctness. All the authors read and approved the final manuscript . Acknowledgments We would like to express our heartfelt thanks to the University of Gondar Hospital for cooperating and permitting the use of the data. Moreover, we would like to give our special appreciation to Debre Markos University for its contribution. Finally, thanks to all who made significant contributions to the success of this study. References “Guterres A. (2023). Viral load: We need a new look at an old problem?. Journal of medical virology, 95(8), e29061. https://doi.org/10.1002/jmv.29061”. T. Marin, “The Importance of Viral Load in Human Body Fluids and its Role in Infectious Diseases,” 2023, doi: 10.35248/2161-0517.23.12.276. K. Lakshmanan and B. M. Liu, “Impact of Point-of-Care Testing on Diagnosis, Treatment, and Surveillance of Vaccine-Preventable Viral Infections,” Jan. 01, 2025, Multidisciplinary Digital Publishing Institute (MDPI) . doi: 10.3390/diagnostics15020123. F. Rouet and C. Rouzioux, “The measurement of HIV-1 viral load in resource-limited settings: How and where?,” Clin Lab , vol. 53, pp. 135–148, Feb. 2007. F. Rouet and C. Rouzioux, “The measurement of HIV-1 viral load in resource-limited settings: How and where?,” Clin Lab , vol. 53, pp. 135–148, Feb. 2007. E. A. Ochodo, E. E. Olwanda, J. J. Deeks, and S. Mallett, “Point-of-care viral load tests to detect high HIV viral load in people living with HIV/AIDS attending health facilities,” Mar. 10, 2022, John Wiley and Sons Ltd . doi: 10.1002/14651858.CD013208.pub2. J. Greig et al. , “Viral load testing in a resource-limited setting: Quality control is critical,” J Int AIDS Soc , vol. 14, no. 1, 2011, doi: 10.1186/1758-2652-14-23. D. N. Mamo et al. , “Machine learning to predict virological failure among HIV patients on antiretroviral therapy in the University of Gondar Comprehensive and Specialized Hospital, in Amhara Region, Ethiopia, 2022,” BMC Med Inform Decis Mak , vol. 23, no. 1, Dec. 2023, doi: 10.1186/s12911-023-02167-7. D. Dixon et al. , “Unveiling the Influence of AI Predictive Analytics on Patient Outcomes: A Comprehensive Narrative Review,” Cureus , May 2024, doi: 10.7759/cureus.59954. M. Maskew et al. , “Applying machine learning and predictive modeling to retention and viral suppression in South African HIV treatment cohorts,” Sci Rep , vol. 12, no. 1, Dec. 2022, doi: 10.1038/s41598-022-16062-0. S. Payagala and A. Pozniak, “The global burden of HIV,” Clin Dermatol , vol. 42, no. 2, pp. 119–127, 2024, doi: https://doi.org/10.1016/j.clindermatol.2024.02.001. A. Kippen, L. Nzimande, D. Gareta, and C. Iwuji, “The viral load monitoring cascade in HIV treatment programmes in sub-Saharan Africa: a systematic review,” BMC Public Health , vol. 24, no. 1, p. 2603, Dec. 2024, doi: 10.1186/s12889-024-20013-x. T. A. Kitaw and R. N. Haile, “Virological outcomes of antiretroviral therapy and its determinants among HIV patients in Ethiopia: Implications for achieving the 95–95–95 target,” PLoS One , vol. 20, no. 1, Jan. 2025, doi: 10.1371/journal.pone.0313481. E. Moyo, P. Moyo, G. Murewanhema, M. Mhango, I. Chitungo, and T. Dzinamarira, “Key populations and Sub-Saharan Africa’s HIV response,” 2023, Frontiers Media S.A. doi: 10.3389/fpubh.2023.1079990. J. L. Marcus, W. C. Sewell, L. B. Balzer, and D. S. Krakower, “Artificial Intelligence and Machine Learning for HIV Prevention: Emerging Approaches to Ending the Epidemic,” Jun. 01, 2020, Springer . doi: 10.1007/s11904-020-00490-6. Y. Xiang, J. Du, K. Fujimoto, F. Li, J. Schneider, and C. Tao, “Application of artificial intelligence and machine learning for HIV prevention interventions,” Lancet HIV , vol. 9, Nov. 2021, doi: 10.1016/S2352-3018(21)00247-2. S. Pal, “A Comparative Analysis of Machine Learning Algorithms for Predictive Analytics in Healthcare,” vol. 72, pp. 10–25, Mar. 2024. M. Javaid, A. Haleem, R. Pratap Singh, R. Suman, and S. Rab, “Significance of machine learning in healthcare: Features, pillars and applications,” International Journal of Intelligent Networks , vol. 3, pp. 58–73, 2022, doi: https://doi.org/10.1016/j.ijin.2022.05.002. S. A. Alowais et al. , “Revolutionizing healthcare: the role of artificial intelligence in clinical practice,” Dec. 01, 2023, BioMed Central Ltd . doi: 10.1186/s12909-023-04698-z. A. Chekol, A. Ketemaw, A. Endale, A. Aschale, B. Endalew, and M. A. Asemahagn, “Data quality and associated factors of routine health information system among health centers of West Gojjam Zone, northwest Ethiopia, 2021,” Frontiers in Health Services , vol. 3, 2023, doi: 10.3389/frhs.2023.1059611. L. G. Gebretsadik et al. , “Health information system in primary health care units of the Central Zone, Tigray, Northern Ethiopia,” BMC Med Inform Decis Mak , vol. 25, no. 1, Dec. 2025, doi: 10.1186/s12911-025-03078-5. H. Abebe et al. , “Health Management Information System Data Quality and its associated factors in Addis Ababa Public Hospitals, Ethiopia, 2022.A cross-sectional study,” Jan. 21, 2025. doi: 10.1101/2025.01.19.25320816. E. Tucker et al. , “Using a Health Information Exchange to Characterize Changes in HIV Viral Load Suppression and Disparities During the COVID-19 Pandemic in New York City,” Open Forum Infect Dis , vol. 10, no. 12, p. ofad584, Dec. 2023, doi: 10.1093/ofid/ofad584. Z. Xie et al. , “Prevention of adverse HIV treatment outcomes: machine learning to enable proactive support of people at risk of HIV care disengagement in Tanzania,” BMJ Open , vol. 14, no. 9, p. e088782, Sep. 2024, doi: 10.1136/bmjopen-2024-088782. V. Zuhair et al. , “Exploring the Impact of Artificial Intelligence on Global Health and Enhancing Healthcare in Developing Nations,” Jan. 01, 2024, SAGE Publications Inc. doi: 10.1177/21501319241245847. A. Z. Al Meslamani, I. Sobrino, and J. de la Fuente, “Machine learning in infectious diseases: potential applications and limitations,” 2024, Taylor and Francis Ltd. doi: 10.1080/07853890.2024.2362869. H. Wang, M. Zhang, L. Mai, X. Li, A. Bellou, and L. Wu, “An effective multi-step feature selection framework for clinical outcome prediction using electronic medical records,” BMC Med Inform Decis Mak , vol. 25, no. 1, Dec. 2025, doi: 10.1186/s12911-025-02922-y. I. Mienye and Y. Sun, “A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects,” IEEE Access , vol. PP, p. 1, Sep. 2022, doi: 10.1109/ACCESS.2022.3207287. G. G. Gebrerufael and Z. G. Asfaw, “Predictors of change in CD4 cell count over time for HIV/AIDS patients on ART follow-up in northern Ethiopia: a retrospective longitudinal study,” BMC Immunol , vol. 25, no. 1, Dec. 2024, doi: 10.1186/s12865-024-00659-3. T. Y. Birhan, L. D. Gezie, D. F. Teshome, and M. M. Sisay, “Predictors of CD4 count changes over time among children who initiated highly active antiretroviral therapy in Ethiopia,” Trop Med Health , vol. 48, no. 1, May 2020, doi: 10.1186/s41182-020-00224-9. A. Phillips et al. , “HIV Viral Load Response to Antiretroviral Therapy According to the Baseline CD4 Cell Count and Viral Load,” JAMA , vol. 286, pp. 2560–2567, Nov. 2001, doi: 10.1001/jama.286.20.2560. A. S. Tegegne, “Robustness of viral load over CD4 cell count in measuring the quality of life of people with HIV at second line regimen in Amhara region,” Sci Rep , vol. 15, no. 1, Dec. 2025, doi: 10.1038/s41598-025-93608-y. N. S. Muhie and A. S. Tegegne, “CD4 cell count and viral load count association and its joint risk factors among adult TB/HIV co-infected patients: a retrospective follow-up study,” BMC Res Notes , vol. 18, no. 1, Dec. 2025, doi: 10.1186/s13104-025-07428-4. S. Mukwenha et al. , “Predictors of Unsuppressed HIV Viral Load and Low CD4 Count Among ZIMPHIA 2020 Survey Participants,” Texila International Journal of Public Health , vol. 12, no. 4, Dec. 2024, doi: 10.21522/TIJPH.2013.12.04.Art062. D. Fernandez et al. , “Assessing sex differences in viral load suppression and reported deaths using routinely collected program data from PEPFAR-supported countries in sub-Saharan Africa,” BMC Public Health , vol. 23, no. 1, Dec. 2023, doi: 10.1186/s12889-023-16453-6. R. Wisaksana, Y. Hartantri, and E. Hutajulu, “Risk Factors Associated with Unsuppressed Viral Load in People Living with HIV Receiving Antiretroviral Treatment in Jawa Barat, Indonesia,” HIV/AIDS - Research and Palliative Care , vol. 16, pp. 1–7, 2024, doi: 10.2147/HIV.S407681. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8799312","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":597336667,"identity":"854a17c9-069d-4a2e-9642-0dab82bfdd48","order_by":0,"name":"Abraham Keffale Mengistu","email":"data:image/png;base64,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","orcid":"","institution":"Debre Markos University","correspondingAuthor":true,"prefix":"","firstName":"Abraham","middleName":"Keffale","lastName":"Mengistu","suffix":""},{"id":597336668,"identity":"6ff04992-9f0a-4fd0-b5f9-bf2c3b361731","order_by":1,"name":"Aynadis Worku Shime","email":"","orcid":"","institution":"Debre Markos University","correspondingAuthor":false,"prefix":"","firstName":"Aynadis","middleName":"Worku","lastName":"Shime","suffix":""},{"id":597336669,"identity":"7cc57d9e-0edd-482d-9ad2-dc553d62fa51","order_by":2,"name":"Muluken Belachew Mengistie","email":"","orcid":"","institution":"Debre Markos University","correspondingAuthor":false,"prefix":"","firstName":"Muluken","middleName":"Belachew","lastName":"Mengistie","suffix":""},{"id":597336670,"identity":"d7f88385-84f7-4a80-92fc-7110c16edf5f","order_by":3,"name":"Gizaw Hailiye Teferi","email":"","orcid":"","institution":"Debre Markos University","correspondingAuthor":false,"prefix":"","firstName":"Gizaw","middleName":"Hailiye","lastName":"Teferi","suffix":""}],"badges":[],"createdAt":"2026-02-05 16:24:38","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8799312/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8799312/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104398698,"identity":"300ddefa-110b-4f39-b1c7-186aa0be29a4","added_by":"auto","created_at":"2026-03-11 12:03:18","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":61511,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePercentage of Missing Data by Feature\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/ba9d010b30f152a84b8bd211.png"},{"id":104399441,"identity":"fe56d1c1-fa33-449f-951a-d20d175e589a","added_by":"auto","created_at":"2026-03-11 12:06:09","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":218690,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCorrelation Heatmap Analysis for Features\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/7875dd08a418f1fbffce5227.png"},{"id":104399276,"identity":"2249c6d7-8937-41a7-8241-3dcf755e570f","added_by":"auto","created_at":"2026-03-11 12:05:21","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":57395,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClass Distribution of Output Variable\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/f06cd8d569ea45f9a5163108.png"},{"id":103613563,"identity":"44354641-b77f-4598-87bb-fc4f7b28bbcb","added_by":"auto","created_at":"2026-02-27 16:18:12","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":384924,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWorkflow diagram\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/0750184b3b85c23e15d60918.jpeg"},{"id":104399463,"identity":"d82219e2-a7f8-4696-a3db-70c96d8fc797","added_by":"auto","created_at":"2026-03-11 12:06:14","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":111014,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAUC-ROC curve Results of Models\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/2804348da906bc4aa37e14bb.png"},{"id":104399442,"identity":"7dce6ff6-5685-49e9-a0e2-9f77b8f9e377","added_by":"auto","created_at":"2026-03-11 12:06:09","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":36473,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion Matrix Result of Tuned Gradient Boosting Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/079f756e8026af33c5a4315a.png"},{"id":104399156,"identity":"5c0b47e3-62e0-44a9-bf8c-e5d42908dfb1","added_by":"auto","created_at":"2026-03-11 12:04:53","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":44259,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature Importance of our Best Performing Model\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/5a1a88ad0b13a5b2cbf0f5d6.png"},{"id":104398667,"identity":"182281e9-0e33-4ce1-a0e5-7bdabd7d4bac","added_by":"auto","created_at":"2026-03-11 12:03:13","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":96172,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP analysis Results\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/86906fa1517d88c008306192.png"},{"id":104399174,"identity":"8886848a-afca-4ab1-9fbc-109eecc6e08e","added_by":"auto","created_at":"2026-03-11 12:04:58","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":92485,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInterpretation of an individual prediction using a SHAP waterfall plot\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/b1e07dc04ea73026b8b1fb8b.png"},{"id":107518201,"identity":"4bd23f06-3208-4389-8ccf-8c5fcdb8d448","added_by":"auto","created_at":"2026-04-22 08:43:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1855493,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8799312/v1/7660bb4e-7268-4a93-803f-b51bcc68d7b2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Explainable Machine Learning for Classification of HIV Viral Load Suppression in Resource- Limited Settings","fulltext":[{"header":"Background","content":"\u003cp\u003eViral load measurement is a critical parameter in the management of infectious diseases such as HIV, hepatitis, and COVID-19 (Coronavirus Disease 2019) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. It quantifies the virus in a patient\u0026rsquo;s blood, aiding diagnosis, monitoring disease burden, and assessing treatment efficacy [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Consequently, the prediction of viral load is of high priority in low-resource settings where access to frequent laboratory testing may be poor [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e], [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. In such contexts, predictive models can serve as valuable tools for estimating viral load levels, enabling healthcare providers to make appropriate treatment plans that improve patient outcomes [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHIV/AIDS (Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome) remains one of the most significant global health challenges, with an estimated 39\u0026nbsp;million people living with the disease worldwide as of 2022, including 1.5\u0026nbsp;million new infections annually [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Sub-Saharan Africa bears the most significant burden, accounting for approximately 67% of all people living with HIV and 72% of AIDS-related deaths globally [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. In Ethiopia, over 500,000 people are living with HIV, and the epidemic remains a leading cause of morbidity and mortality despite expanded access to antiretroviral therapy [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. These stark disparities highlight the urgency of optimizing strategies for viral load monitoring and treatment management in resource-limited settings.\u003c/p\u003e \u003cp\u003eThe global burden of HIV/AIDS, particularly in sub-Saharan Africa, which accounts for the majority of new infections and AIDS-related deaths, remains a major public health challenge[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e], [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Ethiopia, one of the most affected countries, has made great progress in expanding access to ART through its health system. However, there are a lot of challenges, such as not enough health infrastructures, unequally distributed resources, and the impossibility of monitoring patients regularly [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. In this context, novel approaches to the management of diseases, including the use of AI (Artificial Intelligence) and ML, could improve the efficiency and effectiveness of HIV care [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn recent years, ML has become a subarea of AI that has grown as one of the strongest predictive modeling tools in healthcare. The ML algorithms can identify patterns and relationships from large datasets that cannot easily be determined by traditional statistical methods [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. This capability is particularly relevant in the context of viral load prediction, where multiple factors interact in complex ways [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. However, the performance of ML models depends heavily on the quality of the data, the choice of algorithms, and the strategies employed to address common challenges such as class imbalance and missing data [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn many LMICs (Low- and Middle-Income Countries), including Ethiopia, data quality challenges are pervasive and multifaceted[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. At the University of Gondar Comprehensive and Specialized Hospital, missing or incomplete data are common in both demographic and clinical fields. Examples include missing age or residence information, inconsistent documentation of ART initiation dates, and incomplete CD4 or viral load test results. Additionally, manual data entry errors and the lack of standardized formats in the EMR system contribute to inaccuracies in medical records. These issues hinder the robustness of ML models, which rely heavily on clean, complete datasets[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Therefore, careful preprocessing, imputation strategies, and quality control mechanisms are necessary to mitigate the effects of poor data quality and improve model reliability.\u003c/p\u003e \u003cp\u003eGondar University Comprehensive and Specialized Hospital is also a very active hub for the care and study of HIV-infected individuals located in the Amhara National Regional State. The ART Clinic, established in 2005, provides free services. The clinic started its work of enrollment in August 2005 and has accrued more than 15,000 patients, with more than 5,000 patients on active follow-up. This will be the source for developing and validating predictive models since a lot of data related to patient demographics, clinical history, immunological status, and treatment outcomes is stored in the EMR (Electronic Medical Records) system. Using this, researchers can derive knowledge concerning the factors associated with viral load suppression and develop tools to support clinical decision-making.\u003c/p\u003e \u003cp\u003eEven though ML is a very promising tool for viral load prediction, there are only a few comprehensive studies that compare different algorithms and imbalance resolution techniques on the subject. Most of the current literature is from high-income settings, where data availability and healthcare infrastructure differ significantly from those in LMICs [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. This literature gap underlines the necessity for studies investigating the performance of ML models under resource-constrained settings where the challenges of poor data quality and class imbalance are mostly heightened.\u003c/p\u003e \u003cp\u003eThis study will, therefore, seek to close this gap through a comparative performance analysis of the eight conventionally used machine learning algorithms, namely, Random Forest, Support Vector Machines, Gradient Boosting, Decision Tree, LightGBM, XGBoost, Logistic Regression, and k-nearest Neighbors on viral load prediction.\u003c/p\u003e \u003cp\u003eThe findings of this study have important implications for public health practice and policy. Accurate viral load prediction can support the timely identification of patients at risk of treatment failure, enabling healthcare providers to intervene early and prevent adverse outcomes [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Beyond this, the use of ML models reduces reliance on expensive and time-consuming laboratory tests, increasing access to and the sustainability of HIV care in resource-constrained settings [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The study will optimize AI models for viral load prediction to further advance the global effort to combat HIV/AIDS and improve the quality of life for individuals living with the virus.\u003c/p\u003e \u003cp\u003eThis study advances previous research by comprehensively comparing multiple machine learning models using a balanced dataset within a real-world LMIC setting. Unlike prior studies that may have been constrained by imbalanced data or an over-reliance on traditional statistical methods, the use of a balanced dataset in this work ensures a robust and unbiased evaluation of model performance. Furthermore, moving beyond a primary focus on accuracy, this research places significant emphasis on explainability through SHAP values, ensuring that the models are not only predictive but also interpretable and clinically relevant.\u003c/p\u003e \u003cp\u003eIn conclusion, the integration of AI and ML into healthcare represents a promising avenue for addressing some of the most pressing challenges in infectious disease management [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. This study further extends the literature by providing a state-of-the-art comparison of ML techniques in predicting viral load suppression status in a resource-limited setting. Using data from the University of Gondar Comprehensive and Specialized Hospital, we strive to develop a strong and generalizable model that will support clinical decision-making and add to the global fight against HIV/AIDS.\u003c/p\u003e"},{"header":"Methods and Materials","content":"\u003ch2\u003e\u003cstrong\u003eStudy Design and Setting\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eA quantitative research approach has been adopted in this study, where machine learning techniques will be employed to predict the viral load status among PLHIV (People Living with Human Immunodeficiency Virus). This research was conducted at the University of Gondar Comprehensive and Specialized Hospital, one of the leading healthcare institutions in Gondar City, Amhara National Regional State, Ethiopia. Gondar City, located about 748\u0026thinsp;km northwest of Addis Ababa, has a population of 457,938 and is one of the largest hubs for health service delivery and research. This ART clinic was established in March 2005 and provides free ART services to over 7 million people in Gondar province and surrounding regions. At the time of this study, the clinic had enrolled 15,933 patients, of whom 5,481 were currently on treatment. This retrospective cohort study was conducted at the University of Gondar\u0026apos;s comprehensive and specialized hospital, which was deliberately chosen not only due to data availability but also because of its distinct demographic and epidemiological profile. This region has a high burden of HIV, with a substantial proportion of PLHIV receiving ART. Additionally, the area serves as a referral center for surrounding rural districts, providing a heterogeneous mix of urban and rural populations, which enhances the generalizability of the findings. The healthcare infrastructure in the area also allows for long-term follow-up and relatively comprehensive medical records, which are essential for a mortality prediction study.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eData Source and Study Population\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe secondary data in this study were obtained from the EMR-ART at the University of Gondar Comprehensive and Specialized Hospital. The data were extracted from the electronic medical records of PLHIV who received ART at the University of Gondar\u0026apos;s comprehensive and specialized hospital between March 2005 and December 2024. Inclusion criteria were: (1) age \u0026ge;15 years at ART initiation, (2) confirmed HIV diagnosis, and (3) complete baseline clinical and laboratory data available at the time of ART initiation. Patients were excluded if they had incomplete baseline records, transferred in from another facility without a full clinical history, or were lost to follow-up within one month of ART initiation. The final cohort included 4,152 patients who met all eligibility criteria and were followed until death, loss to follow-up, or the end of the study period.\u003c/p\u003e\n\u003cp\u003eAll persons living with HIV on ART who visited the hospital for care in its ART clinic formed the source population. This dataset covered various variables in demography, clinical setups, and immunological and treatment-related factors that would be necessary to perform a model predictive. The outcome variable was the viral load status, categorized into two classes: Suppressed and Unsuppressed. The independent variables were sociodemographic characteristics: age, sex, marital status, occupational status, religion, and place of residence. Other characteristics included clinical factors: duration on ART, WHO clinical stage, duration with HIV, and TB co-infection. Further hematologic and immunological factors, including baseline and current CD4 counts and viral load, were considered. Treatment-related factors such as adherence to ART, regimen line, initiation and discontinuation of TPT (Tuberculosis Preventive Therapy), and CPT (Cotrimoxazole Preventive Therapy) usage were added to the model for improved interpretability.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eData Preprocessing\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe data was preprocessed, following a structured data preprocessing pipeline to ensure the quality and reliability of the data. Machine learning requires a high-quality dataset for prediction. Due to this, handling the missing data during the pre-processing of the dataset is a crucial phase. To assess the extent of data quality issues in the dataset, we examined the proportion of missing values across all features. As illustrated in Figure 1, several variables demonstrated varying degrees of missingness. Notably, \u0026lsquo;Baseline CD4 Count\u0026rsquo; and \u0026lsquo;Recent CD4 Count\u0026rsquo; exhibited the highest levels of missing data, with 15.2% and 15.0% of values missing, respectively. Other variables, such as \u0026lsquo;Duration on ART in Months\u0026rsquo; and \u0026lsquo;Functional Status\u0026rsquo;, had missing data ranging from 1% to 5%, while the majority of features had less than 2% missing values. Overall, the dataset contained approximately 1.7% missing data. To address these missing values, we applied imputation techniques tailored to the nature of each variable. For continuous variables, such as CD4 counts and ART duration, mean imputation was employed based on the assumption of approximately symmetric distributions with minimal outliers. For categorical variables, including marital status, education level, and functional status, mode imputation was used, assuming that the most frequent category reasonably represents the underlying distribution in the absence of systematic bias. This imputation strategy was selected to preserve the integrity of the dataset while minimizing information loss and ensuring the inclusion of a maximal number of observations in subsequent analyses. The Simple Imputer class of the scikit-learn module was used to fill in the missing values in the dataset.\u003c/p\u003e\n\u003cp\u003eData pre-processing also includes encoding data, which is a crucial step. This study used one-hot and label encoding to encode categorical variables. Values with two or more category values are considered categorical if they are discrete and not continuous. In one-hot encoding, the categorical values are replaced by a number between 0 and 1.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eFeature Selection\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eFrom the EMR, we extract all relevant features and do correlation analysis to determine if there is a strongly correlated feature that will affect the model performance and introduce biases. The correlation heatmap illustrates the strength and direction of linear relationships among the variables in the dataset, with a specific focus on identifying factors associated with the \u003cstrong\u003eviral load status\u003c/strong\u003e, the primary outcome variable in this study. As shown in the heatmap, \u003cem\u003eviral load status\u003c/em\u003e demonstrates modest positive correlations with \u003cstrong\u003eRecent CD4 Count\u003c/strong\u003e (r = 0.35), \u003cstrong\u003eRecent_CD4_Category\u003c/strong\u003e (r = 0.35), and \u003cstrong\u003eBaseline CD4 Count\u003c/strong\u003e (r = 0.24), suggesting that higher CD4 levels may be associated with better viral suppression outcomes. Additionally, weak positive correlations are observed with variables such as \u003cstrong\u003eCPT Use\u003c/strong\u003e (r = 0.05) and \u003cstrong\u003eDuration on ART in Months\u003c/strong\u003e (r = 0.11), indicating potential clinical relevance. Conversely, a slight negative correlation is noted between \u003cstrong\u003eAge\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eand \u003cem\u003eviral load status\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e(r = -0.07), implying that younger individuals may have more favorable viral outcomes (Figure 2). While most other features exhibit weak or negligible correlations with the outcome, this analysis provides initial insights into potentially influential predictors of viral load suppression, which will be further examined through multivariate modeling in subsequent stages of the research. Theoretically, the suppression status derives from the recent CD4 count and is directly related to the recent CD4 category, which is stated as low and high, derived directly from the count; we remove the recent CD4 count and the recent CD4 category from the final analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Balancing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this dataset, the dependent variable (viral load status) was almost evenly distributed, with suppressed cases representing 50.4% and unsuppressed cases 49.6% (Fig. 3). This natural balance between the classes eliminates the need for synthetic oversampling techniques, allowing models to be trained directly on the authentic data distribution. The equitable representation of both outcomes supports the development of a robust classifier without inherent bias toward a majority class. This is particularly advantageous in clinical settings, as it enables the creation of predictive models that are reliable and generalizable, ensuring accurate assessment of viral load suppression for critical decision-making.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine Learning Models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor the estimation of viral load status, we evaluated eight ML algorithms, comprising both linear and ensemble-based models. Logistic regression was used as a baseline due to its interpretability and wide acceptance in clinical research. Ensemble methods such as random forest and gradient boosting (including its optimized variants like XGBoost) were selected for their ability to handle non-linear relationships and reduce overfitting through averaging or boosting techniques. Support vector machines, Na\u0026iuml;ve Bayes, and k-nearest neighbors were included to assess performance under different assumptions of class boundaries and data similarity. Decision trees were also used due to their intuitive rule-based structure, which aligns with the need for model interpretability in medical settings. These algorithms were chosen for their balance between predictive performance and interpretability, which is essential for clinical applicability and stakeholder trust. Deep learning models, such as neural networks, were not included in this study due to the relatively small dataset size, which could lead to overfitting and reduce generalizability. Furthermore, deep learning models often require extensive computational resources and hyperparameter tuning, which were beyond the current study\u0026apos;s scope. Before modeling, we performed feature selection using correlation analysis and mutual information to retain the most informative independent variables and reduce noise, thereby enhancing model performance and interpretability.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eModel Training and Optimization\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe computational environment used for model training and optimization consisted of an Intel(R) Core (TM) i7-8650U CPU running at 1.90 GHz, with 16 GB of installed RAM on a 64-bit Windows operating system. On average, model training and hyperparameter tuning took approximately 3 to 8 minutes per algorithm, depending on the model complexity (e.g., logistic regression and decision tree models converged faster, while ensemble methods like XGBoost and Gradient Boosting required longer due to iterative boosting).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe dataset was split into training and testing subsets, with 80% allocated for training and 20% for testing. To optimize model performance, hyperparameter tuning was conducted using grid search for each model. For logistic regression, tuning focused on regularization strength (C), explored over a logarithmic scale from 10\u003csup\u003e-4\u0026nbsp;\u003c/sup\u003eto 10\u003csup\u003e3\u0026nbsp;\u003c/sup\u003eand penalty type (both L1 and L2, which are for Lasso and Ridge regularization, respectively), while for Random Forest, these were several trees (between 50 to 500), maximum depth between 5 to 50, minimum samples per leaf from 1 to 10. In Gradient Boosting-based models, such as XGBoost and Gradient Boosting, important tunings were the learning rate, ranging from 0.001 to 0.3; estimators ranging from 50 to 500; and maximum depth between 3 to 15. For support vector machines (SVM), tuning was performed on kernel type (linear, polynomial, and RBF (Radial Basis Function), regularization parameter (C) ranging from 10\u003csup\u003e-3\u003c/sup\u003e to 10\u003csup\u003e2\u003c/sup\u003e, and gamma values between 10\u003csup\u003e-4\u003c/sup\u003e to 10\u003csup\u003e3\u0026nbsp;\u003c/sup\u003eand decision trees were fine-tuned by varying max depth (3 to 20), minimum samples split (2 to 10), and evaluating both Gini impurity and entropy criteria.\u003c/p\u003e\n\u003cp\u003eKNN (k-Nearest Neighbors) tuned the number of neighbors from 1 to 50 and the distance metric by Euclidean, Manhattan, and Minkowski. Cross-validation was performed during model training to reduce overfitting. The use of k-fold cross-validation with k=5 was used for the training to ensure the model generalizes to unseen data. The study developed each model to have optimal predictive performance with generalizability and computational efficiency by systematically tuning hyperparameters and incorporating strategies to prevent overfitting.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eEvaluation Metrics\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eModel performance was evaluated using a suite of metrics selected for clinical relevance and robustness. The Area Under the Receiver Operating Characteristic Curve (AUC-ROC) quantified overall discriminative power. Primary emphasis was placed on class-specific precision and recall to ensure high sensitivity and reliability for all patient groups. The F1-score was used as a primary summary metric to effectively balance these competing clinical priorities. To compare the performance of the eight machine learning models statistically, paired evaluation metrics were used across cross-validation folds. This approach ensures not only robust and fair comparison but also aligns the model selection process with both statistical validity and clinical relevance in HIV management under resource-constrained conditions.\u003c/p\u003e\n\u003cp\u003eOverall, the process involves collecting EMR data, preprocessing it, selecting and training models, optimizing performance, and evaluating results to identify the best model (Fig. 4).\u003c/p\u003e\n\u003ch4\u003e\u003cstrong\u003eImplementation\u003c/strong\u003e\u003c/h4\u003e\n\u003cp\u003eAll of the models were implemented in the Python programming language using the machine learning libraries. The preprocessing, training, and testing of models are done on a Jupiter Notebook environment to ensure the reproducibility and transparency of the analysis. Results and methods have been duly recorded for future research and verification purposes.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eEthical Considerations\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eEthical approval for this study was obtained from the Institutional Review Board, Debre Markos University. Since all data are anonymized before the analysis, patient confidentiality is strictly assured. Even though these data are secondary, access to the data set was restricted to authorized researchers only to comply with ethical guidelines and data protection policies.\u003c/p\u003e"},{"header":"Results","content":"\u003ch2\u003eDescriptive Statistics\u003c/h2\u003e\n\u003cp\u003eThe dataset consisted of 4,152 patients, with 50.4% having suppressed viral load and 49.6 % having unsuppressed viral load. The mean age of participants was 46.72 years (SD = 11.38), with the majority being female (59.51%) (Table 1).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1: Socio-demographic variables\u003c/strong\u003e\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"579\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFrequency\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e2471\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e59.51\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e1681\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e40.49\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003eMarital Status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eMarried\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e2700\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e65.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eDivorced\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e698\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e16.81\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eNever Married\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e437\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e10.53\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eWidowed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e317\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e7.63\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003eEducation level\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eSecondary Education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e2145\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e51.66\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eNo Education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e805\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e19.39\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003ePrimary Education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e789\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e19.00\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eHigher Education\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e413\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e9.95\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003eResidence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eUrban\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e3177\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e76.52\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eRural\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e975\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e23.48\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\n \u003cp\u003eReligion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eOrthodox\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e3910\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e94.17\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eMuslim\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e206\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e4.96\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eProtestant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e0.63\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eCatholic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e0.22\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 106px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 179px;\"\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 152px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e0.02\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eRegarding clinical characteristics, the median duration of ART was approximately 12.7 years (152.5 months), and most patients were in WHO clinical stage 4. The dataset exhibited a balanced class, with suppressed viral load being the predominant class (50.4%), and unsuppressed (49.6 %) viral load categories (Table 2).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2:Clinical-related variables\u003c/strong\u003e\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"605\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFrequency\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eNormal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e2607\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e62.79\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eOverweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e733\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e17.65\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eUnderweight\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e630\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e15.17\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eObese\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e182\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e4.38\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eTB screening Result\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e4070\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e98.03\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e1.97\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eFunctional Status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eWorking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e3861\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e92.99\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eAmbulatory\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e163\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e3.93\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eBedridden\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e128\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e3.08\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eViral Load Status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eSuppressed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e2092\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e50.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eUnsuppressed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e2060\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e49.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eAdherence\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eGood\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e3611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e86.97\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003ePoor\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e355\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e8.55\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eFair\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e186\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e4.48\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eRegimen Line\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eFirst Line\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e3811\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e91.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eSecond Line\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e319\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e7.68\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eThird Line\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.53\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eTPT Started\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e2958\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e71.24\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e1194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e28.76\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eCPT Use\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e3313\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e79.79\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 184px;\"\u003e\n \u003cp\u003eCPT Use\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 168px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 143px;\"\u003e\n \u003cp\u003e839\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e20.21\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003ch2\u003eModel Training Results\u003c/h2\u003e\n\u003cp\u003eIn this study, the comparative evaluation of machine learning models revealed clear distinctions in performance across the selected algorithms. \u003cstrong\u003eGradient Boosting\u003c/strong\u003e and \u003cstrong\u003eRandom Forest\u003c/strong\u003e emerged as the top-performing models, achieving accuracies of 0.73 and 0.72, respectively, with consistently balanced precision, recall, and F1 scores (all around 0.70). This highlights their robustness and reliability in classifying the data effectively. \u003cstrong\u003eNaive Bayes\u003c/strong\u003e stood out for its exceptionally high recall (0.84), suggesting a strong ability to identify positive cases, albeit at the expense of lower precision (0.59), which may increase false positives. \u003cstrong\u003eDecision Tree\u003c/strong\u003e and \u003cstrong\u003eXGBoost\u003c/strong\u003e offered stable mid-range performance, with accuracies of 0.67 and 0.69 and balanced metrics, making them reasonable alternatives. \u003cstrong\u003eLogistic Regression\u003c/strong\u003e yielded moderate results (accuracy = 0.62), while \u003cstrong\u003eSVM\u003c/strong\u003e and \u003cstrong\u003eKNN\u003c/strong\u003e underperformed, both with accuracies of 0.54 and comparatively lower overall scores (Table 3). Collectively, these findings indicate that ensemble-based methods, particularly Gradient Boosting and Random Forest, are the most effective approaches for this dataset, providing a promising direction for building accurate and reliable predictive models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Model Training Results\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"591\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.60\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eGradient Boosting\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.73\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.70\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.70\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eNaive Bayes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.84\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eDecision Tree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 183px;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 97px;\"\u003e\n \u003cp\u003e0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 98px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 91px;\"\u003e\n \u003cp\u003e0.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 122px;\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe ROC curves shown in Figure 5 illustrate the trade-off between the true positive rate and false positive rate for all models evaluated. The \u003cstrong\u003eGradient Boosting\u003c/strong\u003e model achieved the highest area under the curve (AUC = 0.79), closely followed by \u003cstrong\u003eRandom Forest\u003c/strong\u003e (AUC = 0.78) and \u003cstrong\u003eXGBoost\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e(AUC = 0.76), confirming their strong discriminative ability. \u003cstrong\u003eNaive Bayes\u003c/strong\u003e also performed well, with an AUC of 0.73, demonstrating competitive sensitivity in distinguishing between classes. In contrast, \u003cstrong\u003eLogistic Regression\u003c/strong\u003e (AUC = 0.66) and \u003cstrong\u003eDecision Tree\u003c/strong\u003e (AUC = 0.67) provided moderate results, while \u003cstrong\u003eSVM\u003c/strong\u003e (AUC = 0.57) and \u003cstrong\u003eKNN\u003c/strong\u003e (AUC = 0.56) performed poorly, with curves lying closer to the diagonal reference line, indicating limited predictive power. Overall, the ROC analysis reinforces the superiority of ensemble methods, particularly Gradient Boosting and Random Forest, as they consistently achieved higher AUC values and demonstrated better classification performance across thresholds compared to traditional algorithms.\u003c/p\u003e\n\u003cp\u003eThe Gradient Boosting model, identified as the best-performing algorithm in the initial evaluation, was further optimized through hyperparameter tuning using \u003cstrong\u003eGridSearchCV\u003c/strong\u003e. \u0026nbsp;After tuning, the optimized Gradient Boosting classifier achieved an \u003cstrong\u003eaccuracy of 0.76\u003c/strong\u003e, with a \u003cstrong\u003eprecision of 0.73\u003c/strong\u003e\u003cstrong\u003e,\u0026nbsp;\u003c/strong\u003ea\u003cstrong\u003e\u0026nbsp;\u003cstrong\u003erecall of 0.75\u003c/strong\u003e\u003c/strong\u003e, and an \u003cstrong\u003eF1 score of 0.74\u003c/strong\u003e. The confusion matrix (Fig. 6) shows that the model correctly classified 339 suppressed cases and 211 unsuppressed cases, while misclassifying 108 suppressed and 113 unsuppressed instances. These results demonstrate a well-balanced performance between sensitivity and precision, reflecting the model\u0026rsquo;s reliability in distinguishing between suppressed and unsuppressed viral load statuses. Overall, hyperparameter tuning improved the Gradient Boosting model\u0026rsquo;s predictive capability, making it the most effective and robust approach among the models tested.\u003c/p\u003e\n\u003ch2\u003eFeature Importance\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe feature importance analysis from the tuned Gradient Boosting model (Fig. 7) highlights the variables that most strongly contributed to predicting viral load suppression status. \u003cstrong\u003eBaseline CD4 Category\u003c/strong\u003e emerged as the most influential predictor, contributing more than half of the total importance score, underscoring the critical role of immunological status at treatment initiation. The second most significant predictor was \u003cstrong\u003eDuration on ART in Months\u003c/strong\u003e, indicating that longer treatment duration is strongly associated with viral load suppression outcomes. Other features, such as \u003cstrong\u003esex\u003c/strong\u003e and \u003cstrong\u003eage,\u003c/strong\u003e contributed moderately, while factors like \u003cstrong\u003eregimen line\u003c/strong\u003e\u003cstrong\u003e, \u003cstrong\u003eCPT use\u003c/strong\u003e\u003c/strong\u003e, and \u003cstrong\u003emarital status\u003c/strong\u003e had relatively smaller impacts. Variables including \u003cstrong\u003eWHO stage, adherence, BMI, educational level, TPT initiation, and TB screening result\u003c/strong\u003e showed minimal importance in the model. These findings suggest that immunological markers and treatment duration are the most decisive factors in determining viral load suppression, while sociodemographic and clinical background factors play a comparatively limited role in prediction.\u003c/p\u003e\n\u003ch2\u003eModel Explainability\u003c/h2\u003e\n\u003cp\u003eThe interpretation of the model using SHAP revealed that the most important features driving predictions were the duration on ART, baseline CD4 category, and patient age. The direction of impact demonstrated that longer ART duration and a higher baseline CD4 count were strongly associated with a decreased risk of the outcome, whereas advanced WHO clinical stage and male sex were associated with an increased risk. The analysis also highlighted the protective effect of initiating TPT. This feature importance ranking and directional analysis align with established clinical understanding, thereby enhancing the credibility and interpretability of the model\u0026apos;s decision-making process (Fig. 8).\u003c/p\u003e\n\u003cp\u003eThe model\u0026apos;s prediction for a specific individual was further elucidated using a SHAP waterfall plot. The analysis begins with the baseline model output, E[f(X)] = -0.201, representing the average prediction across the dataset. For this patient, the most influential factors increasing their predicted risk were a \u003cstrong\u003eWHO Stage of 4\u003c/strong\u003e (+1.3) and being \u003cstrong\u003emale\u003c/strong\u003e (Sex=1, +1.27). Their \u003cstrong\u003eage of 53\u003c/strong\u003e also contributed significantly to a higher risk (+0.86). These factors were partially offset by a protective effect from a \u003cstrong\u003elonger duration on ART (162 months)\u003c/strong\u003e\u003cstrong\u003e,\u003c/strong\u003e which decreased the risk (-0.41)(Fig. 9). Contributions from other features, such as \u003cstrong\u003eCPT Use\u003c/strong\u003e and \u003cstrong\u003eBaseline CD4 Category\u003c/strong\u003e, had smaller positive effects. Ultimately, the cumulative effect of all these feature contributions shifted the model\u0026apos;s prediction from the population average to a final, higher value for this individual, clearly illustrating how their specific clinical and demographic profile led to the elevated risk score.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study sought to develop and interpret a machine learning classifier for estimating HIV viral load suppression from regularly collected clinical information in a low-resource environment. The principal finding is that ensemble-based machine learning models, namely Gradient Boosting, can generate stable and clinically sound estimates of viral load status. The best-performing optimized Gradient Boosting model achieved a balanced accuracy of 76%, with superior performance in precision, recall, and F1-score measures. This performance, coupled with a model explainability evaluation via SHAP, is an actionable and interpretable tool consistent with existing clinical evidence, with great potential for the facilitation of HIV care management in resource-limited environments.\u003c/p\u003e \u003cp\u003eHigher performance of tree-based ensemble methods like Gradient Boosting and Random Forest aligns with existing literature on healthcare predictive modeling [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. These models are specially adept at making inferences of complex, non-linear interactions between features, characteristic of clinical information, wherein patient outcomes are a function of a complex interaction between demographic, clinical, and treatment-related variables. Less fundamental models like Logistic Regression and KNN were found to perform poorly, as they may not be able to accurately represent such complex relationships. Naive Bayes' high recall, while intriguing, came with low precision, rendering it less desirable for the clinical setting where a false positive equates to wasteful resource use.\u003c/p\u003e \u003cp\u003eA second foundation of this work's value is its emphasis on model explainability. Feature importance analysis and SHAP plots provide beyond a \"black box\" prediction, clinically interpretable information. The observation that Baseline CD4 Category and Duration on ART (in months) were the strongest predictors is well-aligned with established principles of virology and immunology[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Higher baseline CD4 levels are in agreement with an intact immune system upon initiation of treatment, and higher duration on ART is indicative of cumulative treatment efficacy and adherence, both of which are established predictors of viral suppression[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Moreover, the SHAP summary plot supported the protective direction of these associations: longer ART duration and higher CD4 levels always pushed model prediction away from the unsuppressed class. The findings of male gender and advanced WHO Stage (Stage 4) as risk factors for unsuppressed viral load are also supported by epidemiologic studies that report differential treatment outcomes by stage of disease and gender[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe plot of a SHAP waterfall on one patient's data is a lovely way to demonstrate the utility of the model for personalized clinical decision-making. For the example given, the model quantified how much the patient's risk factors (WHO Stage 4 disease, male, old age) were offset by the beneficial effect of long ART duration. This granularity of explanation might help clinicians understand the \"why\" behind a model's risk score and facilitate appropriate interventions. For instance, a patient who has a high-risk score because of poor adherence (a modifiable factor) would be managed differently from one whose risk is largely owing to poor baseline CD4 count (a non-modifiable factor).\u003c/p\u003e \u003cp\u003eIncorporation of such a model into EMR systems of low-resource clinic settings can be an early warning system. With automated risk score calculation during the encounter of the patient, the model has the potential to notify clinicians to flag patients at increased risk of viral load suppression for priority counseling regarding adherence, escalated follow-up, or enhanced clinical monitoring before treatment failure. This pre-emptive strategy may make the best of finite healthcare resources and also enhance the overall outcomes of therapy. That the model depends upon variables generally found in EMRs for ART clinics makes it most feasible for deployment without new, expensive testing.\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eLimitations and Future Directions\u003c/h2\u003e \u003cp\u003eWhile having supportive results, this study has several limitations. First, the data are derived from a single tertiary hospital in Ethiopia and hence may have limited generalizability to other populations with their own unique demographic and epidemiological characteristics. External verification with datasets from other regions is required. Second, while the dataset was balanced for the outcome variable, it may not have contained all relevant predictors, e.g., certain socio-economic variables, psychosocial stressors, or genetic markers. Future research would be well-advised to incorporate these variables to further enhance predictive power. Third, the model's performance, while good, indicates that there remains some unexplained variance, and viral load suppression must therefore rely, at least in part, on factors yet to be captured in the current feature set.\u003c/p\u003e \u003cp\u003eFuture studies would need to focus on three areas: 1) Multi-center verification to test the strength and transportability of the model to a variety of sub-Saharan African health settings. 2) Prospective studies of implementation to quantify the real-world impact of deploying the model into clinical practice on patient outcomes and utilization of resources. 3) Exploration of temporal modeling techniques to estimate viral load suppression risk dynamically through time, rather than at a snapshot in time, providing yet another more persuasive tool for long-term management of patients.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, this study successfully developed an interpretable machine learning model for predicting viral load suppression in a low-resource setting. The Gradient Boosting model not only demonstrated superior classification performance but also, through SHAP analysis, provided insights that are clinically coherent and actionable. By bridging the gap between predictive accuracy and interpretability, this research offers a practical and trustworthy decision-support tool that can empower healthcare providers to optimize HIV care, ultimately contributing to the global effort to achieve sustained virological suppression for all people living with HIV.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cstrong\u003eAI:\u003c/strong\u003e Artificial Intelligence\u003cbr\u003e\u003cstrong\u003eART:\u003c/strong\u003e Antiretroviral Therapy\u003cbr\u003e\u003cstrong\u003eAUC-ROC:\u003c/strong\u003e Area Under the Receiver Operating Characteristic Curve\u003cbr\u003e\u003cstrong\u003eBMI:\u003c/strong\u003e Body Mass Index\u003cbr\u003e\u003cstrong\u003eCPT:\u003c/strong\u003e Cotrimoxazole Preventive Therapy\u003cbr\u003e\u003cstrong\u003eEMR:\u003c/strong\u003e Electronic Medical Record\u003cbr\u003e\u003cstrong\u003eHIV:\u003c/strong\u003e Human Immunodeficiency Virus\u003cbr\u003e\u003cstrong\u003eKNN:\u003c/strong\u003e k-Nearest Neighbors\u003cbr\u003e\u003cstrong\u003eLMICs:\u003c/strong\u003e Low- and Middle-Income Countries\u003cbr\u003e\u003cstrong\u003eML:\u003c/strong\u003e Machine Learning\u003cbr\u003e\u003cstrong\u003ePLHIV:\u003c/strong\u003e People Living with HIV\u003cbr\u003e\u003cstrong\u003eSHAP:\u003c/strong\u003e SHapley Additive exPlanations\u003cbr\u003e\u003cstrong\u003eSVM:\u003c/strong\u003e Support Vector Machine\u003cbr\u003e\u003cstrong\u003eTB:\u003c/strong\u003e Tuberculosis\u003cbr\u003e\u003cstrong\u003eTPT:\u003c/strong\u003e Tuberculosis Preventive Therapy\u003cbr\u003e\u003cstrong\u003eVL:\u003c/strong\u003e Viral Load\u003cbr\u003e\u003cstrong\u003eVLS:\u003c/strong\u003e Viral Load Suppression\u003cbr\u003e\u003cstrong\u003eWHO:\u003c/strong\u003e World Health Organization\u003cbr\u003e\u003cstrong\u003eXGBoost:\u003c/strong\u003e Extreme Gradient Boosting\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthical approval and consent to participate\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe study was approved by the Institutional Review Board (IRB) of Debre Markos University College of Medicine and Health Sciences. The need for individual consent was waived by the IRB, as the research utilized anonymized and de-identified data from existing records, making it unfeasible to obtain consent from individuals. The study was conducted in accordance with the principles of the Helsinki Declaration and adhered to all relevant national regulations and ethical guidelines. All data were kept confidential, and no personal identifiers were included in the analysis or reporting. The research processes were performed and secured under the appropriate ethical and regulatory standards by national regulations and ethical guidelines.\u003c/p\u003e\n\u003ch2\u003eClinical Trial Number\u003c/h2\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003ch2\u003eConsent for publication\u003c/h2\u003e\n\u003cp\u003e\u0026nbsp;Not applicable\u003c/p\u003e\n\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e\n\u003cp\u003eData will be available upon request from the corresponding author. The code and experiments used for this research are available on the following public GitHub repository. \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(https://github.com/abrahamekeffale/Viral_suppression_project.git)\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eFunding not applicable\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u0026nbsp;\u003c/h2\u003e\n\u003cp\u003eThe authors declare that they have no competing interests\u003c/p\u003e\n\u003ch2\u003eAuthors\u0026rsquo; contributions\u003c/h2\u003e\n\u003cp\u003eA.K.M. conceptualized the study, and A.E.G. was involved in design, analysis, interpretation, report, and manuscript writing. A.W.S. and M.B.M. edited the manuscript for clarity and correctness. All the authors read and approved the final manuscript\u003cstrong\u003e.\u003c/strong\u003e\u003c/p\u003e\n\u003ch2\u003eAcknowledgments\u003c/h2\u003e\n\u003cp\u003eWe would like to express our heartfelt thanks to the University of Gondar Hospital for cooperating and permitting the use of the data. Moreover, we would like to give our special appreciation to Debre Markos University for its contribution. Finally, thanks to all who made significant contributions to the success of this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003e\u0026ldquo;Guterres A. (2023). Viral load: We need a new look at an old problem?. Journal of medical virology, 95(8), e29061. https://doi.org/10.1002/jmv.29061\u0026rdquo;.\u003c/li\u003e\n\u003cli\u003eT. Marin, \u0026ldquo;The Importance of Viral Load in Human Body Fluids and its Role in Infectious Diseases,\u0026rdquo; 2023, doi: 10.35248/2161-0517.23.12.276.\u003c/li\u003e\n\u003cli\u003eK. Lakshmanan and B. M. Liu, \u0026ldquo;Impact of Point-of-Care Testing on Diagnosis, Treatment, and Surveillance of Vaccine-Preventable Viral Infections,\u0026rdquo; Jan. 01, 2025, \u003cem\u003eMultidisciplinary Digital Publishing Institute (MDPI)\u003c/em\u003e. doi: 10.3390/diagnostics15020123.\u003c/li\u003e\n\u003cli\u003eF. Rouet and C. Rouzioux, \u0026ldquo;The measurement of HIV-1 viral load in resource-limited settings: How and where?,\u0026rdquo; \u003cem\u003eClin Lab\u003c/em\u003e, vol. 53, pp. 135\u0026ndash;148, Feb. 2007.\u003c/li\u003e\n\u003cli\u003eF. Rouet and C. Rouzioux, \u0026ldquo;The measurement of HIV-1 viral load in resource-limited settings: How and where?,\u0026rdquo; \u003cem\u003eClin Lab\u003c/em\u003e, vol. 53, pp. 135\u0026ndash;148, Feb. 2007.\u003c/li\u003e\n\u003cli\u003eE. A. Ochodo, E. E. Olwanda, J. J. Deeks, and S. Mallett, \u0026ldquo;Point-of-care viral load tests to detect high HIV viral load in people living with HIV/AIDS attending health facilities,\u0026rdquo; Mar. 10, 2022, \u003cem\u003eJohn Wiley and Sons Ltd\u003c/em\u003e. doi: 10.1002/14651858.CD013208.pub2.\u003c/li\u003e\n\u003cli\u003eJ. Greig \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Viral load testing in a resource-limited setting: Quality control is critical,\u0026rdquo; \u003cem\u003eJ Int AIDS Soc\u003c/em\u003e, vol. 14, no. 1, 2011, doi: 10.1186/1758-2652-14-23.\u003c/li\u003e\n\u003cli\u003eD. N. Mamo \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Machine learning to predict virological failure among HIV patients on antiretroviral therapy in the University of Gondar Comprehensive and Specialized Hospital, in Amhara Region, Ethiopia, 2022,\u0026rdquo; \u003cem\u003eBMC Med Inform Decis Mak\u003c/em\u003e, vol. 23, no. 1, Dec. 2023, doi: 10.1186/s12911-023-02167-7.\u003c/li\u003e\n\u003cli\u003eD. Dixon \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Unveiling the Influence of AI Predictive Analytics on Patient Outcomes: A Comprehensive Narrative Review,\u0026rdquo; \u003cem\u003eCureus\u003c/em\u003e, May 2024, doi: 10.7759/cureus.59954.\u003c/li\u003e\n\u003cli\u003eM. Maskew \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Applying machine learning and predictive modeling to retention and viral suppression in South African HIV treatment cohorts,\u0026rdquo; \u003cem\u003eSci Rep\u003c/em\u003e, vol. 12, no. 1, Dec. 2022, doi: 10.1038/s41598-022-16062-0.\u003c/li\u003e\n\u003cli\u003eS. Payagala and A. Pozniak, \u0026ldquo;The global burden of HIV,\u0026rdquo; \u003cem\u003eClin Dermatol\u003c/em\u003e, vol. 42, no. 2, pp. 119\u0026ndash;127, 2024, doi: https://doi.org/10.1016/j.clindermatol.2024.02.001.\u003c/li\u003e\n\u003cli\u003eA. Kippen, L. Nzimande, D. Gareta, and C. Iwuji, \u0026ldquo;The viral load monitoring cascade in HIV treatment programmes in sub-Saharan Africa: a systematic review,\u0026rdquo; \u003cem\u003eBMC Public Health\u003c/em\u003e, vol. 24, no. 1, p. 2603, Dec. 2024, doi: 10.1186/s12889-024-20013-x.\u003c/li\u003e\n\u003cli\u003eT. A. Kitaw and R. N. Haile, \u0026ldquo;Virological outcomes of antiretroviral therapy and its determinants among HIV patients in Ethiopia: Implications for achieving the 95\u0026ndash;95\u0026ndash;95 target,\u0026rdquo; \u003cem\u003ePLoS One\u003c/em\u003e, vol. 20, no. 1, Jan. 2025, doi: 10.1371/journal.pone.0313481.\u003c/li\u003e\n\u003cli\u003eE. Moyo, P. Moyo, G. Murewanhema, M. Mhango, I. Chitungo, and T. Dzinamarira, \u0026ldquo;Key populations and Sub-Saharan Africa\u0026rsquo;s HIV response,\u0026rdquo; 2023, \u003cem\u003eFrontiers Media S.A.\u003c/em\u003e doi: 10.3389/fpubh.2023.1079990.\u003c/li\u003e\n\u003cli\u003eJ. L. Marcus, W. C. Sewell, L. B. Balzer, and D. S. Krakower, \u0026ldquo;Artificial Intelligence and Machine Learning for HIV Prevention: Emerging Approaches to Ending the Epidemic,\u0026rdquo; Jun. 01, 2020, \u003cem\u003eSpringer\u003c/em\u003e. doi: 10.1007/s11904-020-00490-6.\u003c/li\u003e\n\u003cli\u003eY. Xiang, J. Du, K. Fujimoto, F. Li, J. Schneider, and C. Tao, \u0026ldquo;Application of artificial intelligence and machine learning for HIV prevention interventions,\u0026rdquo; \u003cem\u003eLancet HIV\u003c/em\u003e, vol. 9, Nov. 2021, doi: 10.1016/S2352-3018(21)00247-2.\u003c/li\u003e\n\u003cli\u003eS. Pal, \u0026ldquo;A Comparative Analysis of Machine Learning Algorithms for Predictive Analytics in Healthcare,\u0026rdquo; vol. 72, pp. 10\u0026ndash;25, Mar. 2024.\u003c/li\u003e\n\u003cli\u003eM. Javaid, A. Haleem, R. Pratap Singh, R. Suman, and S. Rab, \u0026ldquo;Significance of machine learning in healthcare: Features, pillars and applications,\u0026rdquo; \u003cem\u003eInternational Journal of Intelligent Networks\u003c/em\u003e, vol. 3, pp. 58\u0026ndash;73, 2022, doi: https://doi.org/10.1016/j.ijin.2022.05.002.\u003c/li\u003e\n\u003cli\u003eS. A. Alowais \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Revolutionizing healthcare: the role of artificial intelligence in clinical practice,\u0026rdquo; Dec. 01, 2023, \u003cem\u003eBioMed Central Ltd\u003c/em\u003e. doi: 10.1186/s12909-023-04698-z.\u003c/li\u003e\n\u003cli\u003eA. Chekol, A. Ketemaw, A. Endale, A. Aschale, B. Endalew, and M. A. Asemahagn, \u0026ldquo;Data quality and associated factors of routine health information system among health centers of West Gojjam Zone, northwest Ethiopia, 2021,\u0026rdquo; \u003cem\u003eFrontiers in Health Services\u003c/em\u003e, vol. 3, 2023, doi: 10.3389/frhs.2023.1059611.\u003c/li\u003e\n\u003cli\u003eL. G. Gebretsadik \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Health information system in primary health care units of the Central Zone, Tigray, Northern Ethiopia,\u0026rdquo; \u003cem\u003eBMC Med Inform Decis Mak\u003c/em\u003e, vol. 25, no. 1, Dec. 2025, doi: 10.1186/s12911-025-03078-5.\u003c/li\u003e\n\u003cli\u003eH. Abebe \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Health Management Information System Data Quality and its associated factors in Addis Ababa Public Hospitals, Ethiopia, 2022.A cross-sectional study,\u0026rdquo; Jan. 21, 2025. doi: 10.1101/2025.01.19.25320816.\u003c/li\u003e\n\u003cli\u003eE. Tucker \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Using a Health Information Exchange to Characterize Changes in HIV Viral Load Suppression and Disparities During the COVID-19 Pandemic in New York City,\u0026rdquo; \u003cem\u003eOpen Forum Infect Dis\u003c/em\u003e, vol. 10, no. 12, p. ofad584, Dec. 2023, doi: 10.1093/ofid/ofad584.\u003c/li\u003e\n\u003cli\u003eZ. Xie \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Prevention of adverse HIV treatment outcomes: machine learning to enable proactive support of people at risk of HIV care disengagement in Tanzania,\u0026rdquo; \u003cem\u003eBMJ Open\u003c/em\u003e, vol. 14, no. 9, p. e088782, Sep. 2024, doi: 10.1136/bmjopen-2024-088782.\u003c/li\u003e\n\u003cli\u003eV. Zuhair \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Exploring the Impact of Artificial Intelligence on Global Health and Enhancing Healthcare in Developing Nations,\u0026rdquo; Jan. 01, 2024, \u003cem\u003eSAGE Publications Inc.\u003c/em\u003e doi: 10.1177/21501319241245847.\u003c/li\u003e\n\u003cli\u003eA. Z. Al Meslamani, I. Sobrino, and J. de la Fuente, \u0026ldquo;Machine learning in infectious diseases: potential applications and limitations,\u0026rdquo; 2024, \u003cem\u003eTaylor and Francis Ltd.\u003c/em\u003e doi: 10.1080/07853890.2024.2362869.\u003c/li\u003e\n\u003cli\u003eH. Wang, M. Zhang, L. Mai, X. Li, A. Bellou, and L. Wu, \u0026ldquo;An effective multi-step feature selection framework for clinical outcome prediction using electronic medical records,\u0026rdquo; \u003cem\u003eBMC Med Inform Decis Mak\u003c/em\u003e, vol. 25, no. 1, Dec. 2025, doi: 10.1186/s12911-025-02922-y.\u003c/li\u003e\n\u003cli\u003eI. Mienye and Y. Sun, \u0026ldquo;A Survey of Ensemble Learning: Concepts, Algorithms, Applications, and Prospects,\u0026rdquo; \u003cem\u003eIEEE Access\u003c/em\u003e, vol. PP, p. 1, Sep. 2022, doi: 10.1109/ACCESS.2022.3207287.\u003c/li\u003e\n\u003cli\u003eG. G. Gebrerufael and Z. G. Asfaw, \u0026ldquo;Predictors of change in CD4 cell count over time for HIV/AIDS patients on ART follow-up in northern Ethiopia: a retrospective longitudinal study,\u0026rdquo; \u003cem\u003eBMC Immunol\u003c/em\u003e, vol. 25, no. 1, Dec. 2024, doi: 10.1186/s12865-024-00659-3.\u003c/li\u003e\n\u003cli\u003eT. Y. Birhan, L. D. Gezie, D. F. Teshome, and M. M. Sisay, \u0026ldquo;Predictors of CD4 count changes over time among children who initiated highly active antiretroviral therapy in Ethiopia,\u0026rdquo; \u003cem\u003eTrop Med Health\u003c/em\u003e, vol. 48, no. 1, May 2020, doi: 10.1186/s41182-020-00224-9.\u003c/li\u003e\n\u003cli\u003eA. Phillips \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;HIV Viral Load Response to Antiretroviral Therapy According to the Baseline CD4 Cell Count and Viral Load,\u0026rdquo; \u003cem\u003eJAMA\u003c/em\u003e, vol. 286, pp. 2560\u0026ndash;2567, Nov. 2001, doi: 10.1001/jama.286.20.2560.\u003c/li\u003e\n\u003cli\u003eA. S. Tegegne, \u0026ldquo;Robustness of viral load over CD4 cell count in measuring the quality of life of people with HIV at second line regimen in Amhara region,\u0026rdquo; \u003cem\u003eSci Rep\u003c/em\u003e, vol. 15, no. 1, Dec. 2025, doi: 10.1038/s41598-025-93608-y.\u003c/li\u003e\n\u003cli\u003eN. S. Muhie and A. S. Tegegne, \u0026ldquo;CD4 cell count and viral load count association and its joint risk factors among adult TB/HIV co-infected patients: a retrospective follow-up study,\u0026rdquo; \u003cem\u003eBMC Res Notes\u003c/em\u003e, vol. 18, no. 1, Dec. 2025, doi: 10.1186/s13104-025-07428-4.\u003c/li\u003e\n\u003cli\u003eS. Mukwenha \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Predictors of Unsuppressed HIV Viral Load and Low CD4 Count Among ZIMPHIA 2020 Survey Participants,\u0026rdquo; \u003cem\u003eTexila International Journal of Public Health\u003c/em\u003e, vol. 12, no. 4, Dec. 2024, doi: 10.21522/TIJPH.2013.12.04.Art062.\u003c/li\u003e\n\u003cli\u003eD. Fernandez \u003cem\u003eet al.\u003c/em\u003e, \u0026ldquo;Assessing sex differences in viral load suppression and reported deaths using routinely collected program data from PEPFAR-supported countries in sub-Saharan Africa,\u0026rdquo; \u003cem\u003eBMC Public Health\u003c/em\u003e, vol. 23, no. 1, Dec. 2023, doi: 10.1186/s12889-023-16453-6.\u003c/li\u003e\n\u003cli\u003eR. Wisaksana, Y. Hartantri, and E. Hutajulu, \u0026ldquo;Risk Factors Associated with Unsuppressed Viral Load in People Living with HIV Receiving Antiretroviral Treatment in Jawa Barat, Indonesia,\u0026rdquo; \u003cem\u003eHIV/AIDS - Research and Palliative Care\u003c/em\u003e, vol. 16, pp. 1\u0026ndash;7, 2024, doi: 10.2147/HIV.S407681.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"HIV, Viral Load Suppression, Machine Learning, Predictive Modeling, SHAP, Low-Resource Settings, Clinical Decision Support","lastPublishedDoi":"10.21203/rs.3.rs-8799312/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8799312/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eEffective viral load (VL) monitoring is crucial in the management of HIV care, but is difficult in resource-constrained settings due to limited access to laboratory examinations. Machine learning (ML) has a promising approach to viral load suppression (VLS) prediction using normal clinical information. This study aimed to develop and interpret an ML model for VLS classification among an Ethiopian cohort.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA retrospective analysis was undertaken with electronic medical records of 4,152 patients on antiretroviral therapy (ART) in the University of Gondar Comprehensive Specialized Hospital. Eight ML algorithms, namely Logistic Regression, Random Forest, and Gradient Boosting, were trained and optimized to classify a binary VLS outcome. Model performance was assessed based on accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). The best-performing model was interpreted with SHapley Additive exPlanations (SHAP) to identify the significant predictors and their sign of impact.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe best-performing Gradient Boosting model performed the best with 76% accuracy, 0.74 F1-score, and 0.79 AUC-ROC. Baseline CD4 Category and Duration on ART in Months were identified as the most impactful predictors through feature importance evaluation. SHAP analysis supported that longer ART duration and larger baseline CD4 count were associated with increased odds of VLS, and that higher WHO clinical stage and male sex were associated with unsuppressed VL. The model's decision-making was further depicted for individual patients by waterfall plots, which enhanced clinical interpretability.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis work demonstrates that one can have an interpretable Gradient Boosting model to properly predict viral load suppression in a low-resource setting. The predictions of the model are made from clinically reasonable factors, linking algorithmic performance to corresponding clinical insight. The tool can potentially assist healthcare workers in identifying patients at risk of treatment failure, enabling the implementation of early interventions and optimizing HIV care management in settings where routine VL testing is not feasible.\u003c/p\u003e","manuscriptTitle":"Explainable Machine Learning for Classification of HIV Viral Load Suppression in Resource- Limited Settings","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-27 16:18:07","doi":"10.21203/rs.3.rs-8799312/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1257c326-fd37-428d-88b2-965de22849f5","owner":[],"postedDate":"February 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-22T08:43:12+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-27 16:18:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8799312","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8799312","identity":"rs-8799312","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}