Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach

Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach | 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 Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach Ambrose Okibure, Jimmy Patrick Alunyo, Sarah Rachael Akello, Tereza Nyapendi, and 11 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8870898/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Background Tuberculosis (TB) remains a major public health challenge, particularly in settings with high HIV prevalence. In Uganda, TB/HIV co-infection contributes to suboptimal treatment success rates (TSR), still below the World Health Organisation (WHO) target of ≥ 90%. Early identification of patients at risk of poor outcomes is essential to mitigate poor adherence, reduce the risk of multidrug-resistant TB (MDR-TB), and improve health outcomes, including reduced morbidity and mortality as well as transmission. This study applied machine learning (ML) techniques to predict TB treatment outcomes among HIV/TB co-infected patients in Uganda and to identify the most influential predictors of treatment failure using feature importance and partial dependence analyses. Methods Data from a retrospective cohort of 5,062 HIV/TB co-infected patients treated in Uganda between 2020 and 2024were analysed. Machine learning models, including logistic regression, Random Forest, Gradient Boosting Machine (GBM), Support Vector Machine (SVM), AdaBoost, and a stacked ensemble, were developed and evaluated using discrimination, calibration, recall, and accuracy metrics. Feature importance and partial dependence analyses were used to interpret model predictions. Results The stacked ensemble and class-balanced logistic regression models achieved the best overall performance (AUC ≈ 0.67; accuracy ≈ 0.62). The Random Forest model exhibited the highest discrimination (ROC-AUC = 0.675), while GBM achieved the highest accuracy (0.768) but low sensitivity to treatment failures. Key predictors of treatment success included age, ART status, sex, marital status, TB classification, and treatment model. Treatment success declined progressively with increasing age, particularly beyond 40 years. Conclusions The models demonstrated moderate predictive performance and identified key demographic and programmatic predictors of TB treatment outcomes. While not suitable for autonomous clinical decision-making, these models may support risk stratification and targeted patient follow-up. Trial registration number: Not Applicable Tuberculosis HIV co-infection Machine learning Treatment outcomes Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background Globally, tuberculosis (TB) remains among the major causes of morbidity and mortality. Approximately an estimated 10.8 million people fell ill with TB worldwide, including 6.0 million men, 3.6 million women and 1.3 million children in 2023 [ 1 ]. The mortality rate for TB disease was disproportionately high, 1.25 million among HIV-negative people compared to 161,000 in HIV-positive people in 2023 [ 1 ]. The health target of the United Nations Sustainable Development Goals (SDGs) is to end the TB epidemic by 2030 [ 2 ]. The burden of TB is high in the African continent, with over 2.5 million people reported diagnosed with TB in the African region in 2022, accounting for a quarter of new TB cases worldwide. The African continent accounted for over 90% of all global TB cases and 87% of all global co-infection cases. The region contributed more than 400,000 (33%) deaths to the global mortality from the disease in the same year [ 1 ]. The World Health Organization (WHO) recommends that early diagnosis and appropriate treatment of tuberculosis (TB) can effectively reduce transmission, avert deaths, and contribute to its elimination. For drug-susceptible TB, a standard six-month course of first-line anti-TB drugs is recommended, while treatment duration may be longer for multidrug-resistant (MDR) or extensively drug-resistant (XDR) TB. Between 2000 and 2020, early diagnosis and treatment of TB were estimated to have averted 66 million deaths [ 2 ]. Despite widespread access to TB prevention, diagnosis, treatment and care services, the global treatment success rate was only 86% among HIV-negative people in 2019. For HIV-associated TB, it was even very low at 77%. Sub-Saharan Africa accounts for approximately 90% of global tuberculosis (TB) cases and about 87% of TB/HIV co-infections. The treatment success rate in the region stands at 78.9% and is even lower among patients co-infected with HIV and TB [ 3 ]. To address the persistently high TB-related mortality, there is a growing need for interventions aimed at predicting treatment outcomes. Machine learning algorithms offer promising approaches for predicting TB treatment success and improving patient outcomes In Uganda, by 2023, TB disease was still a big public health challenge, with more than 200 people falling ill with TB disease and approximately 30 losing their lives [ 4 ]. Even though 94% of TB/HIV co-infected patients initiated antiretroviral therapy (ART), there was a high mortality rate of 12% [ 5 ]. Early detection of TB disease and prompt initiation of treatment with potent anti-TB drug regimens are required to decrease morbidity and mortality from TB disease among HIV-infected patients [ 6 ]. Despite implementing these recommendations and improving access to TB prevention, diagnosis, treatment, and care, Uganda still has a very low TSR of 70% among TB/HIV co-infections and 72% among HIV-negative people[ 7 ]. Additionally, the country is also ranked among the thirty nations globally grappling with high burdens of TB and TB/HIV co-infection [ 8 ]. Annually, an estimated 91,000 individuals in Uganda contract TB, with 32% of these cases occurring in people living with HIV. Additionally, two percent of TB patients face drug-resistant strains that are not responsive to initial medication, and around 15% of TB cases in Uganda involve children under the age of 14 [ 4 ]. This Treatment Success Rate is way below the desired WHO target of ≥ 90% [ 9 ]. TB treatment is typically successful with the standard six-month course of anti-TB drugs. However, treatment outcomes vary widely due to clinical, socio-demographic, environmental, and health-system-related factors. Studies show that treatment success is influenced by patient-related factors (e.g., medication adherence, co-morbidities like HIV), health system factors (e.g., delays in diagnosis, access to healthcare), and environmental factors. However, most previous studies. Few studies in Uganda have applied ML to routine EMR data to predict TB outcomes among HIV/TB co-infected patients, and majority have relied on conventional statistical models, which have limitations in handling complex, high-dimensional data and non-linear relationships among predictors [ 3 , 9 ] Machine learning (ML) offers a more robust, data-driven approach to identifying early predictors of treatment failure and MDR-TB progression by handling large, high-dimensional datasets that include patient characteristics, clinical parameters, identifying hidden patterns that conventional models might miss and improving prediction accuracy by considering both linear and non-linear relationships among predictors. Despite growing interest in machine learning for TB outcome prediction, few studies in Uganda have applied these methods to nationally representative EMR data, particularly among HIV/TB co-infected patients. This study addresses this gap by evaluating multiple ML models and identifying key predictors of treatment outcomes. Methods Study design The study utilised a retrospective cohort study design of people with HIV/TB co-infection receiving TB treatment across the country between 2020 and 2024. Study setting The study was conducted across 15 Regional Referral Hospitals (RRHs) in Uganda, which serve as key centers for TB diagnosis, treatment, and management, particularly for HIV/TB co-infected patients. These hospitals provide specialized services, including multidrug-resistant TB (MDR-TB) management, Directly Observed Therapy (DOTS), and integration with HIV care through antiretroviral therapy (ART) clinics. They also act as referral centers for lower-level health facilities, such as district hospitals and health center IVs, which initiate TB treatment before referring to complex cases to RRHs. The 15 RRHs include Arua RRH, Fort Portal RRH, Gulu RRH, Hoima RRH, Jinja RRH, Kabale RRH, Lira RRH, Mbale RRH, Mbarara RRH, Moroto RRH, Masaka RRH, Mubende RRH, Naguru RRH (China-Uganda Friendship Hospital), Soroti RRH, Kayunga RRH. TB management at these hospitals follows the Uganda National TB and Leprosy Program (NTLP) guidelines, ensuring standardized patient diagnosis, treatment monitoring, adherence support, and follow-up care. Study population The study targeted TB patients diagnosed with HIV or already on HIV care in Regional Referral Hospitals in Uganda. The accessible population were all PLWH diagnosed with TB and started on anti-TB drug regimens in their respective clinics in the country for the period extending from 2020 to 2024. The sample population comprised HIV/TB co-infected patients whose data were randomly selected from the Uganda EMR system and used in this study. Random selection of patient records was employed to ensure that the data used were representative of the broader national population of HIV/TB co-infected patients. This approach minimized selection bias and enhanced the generalizability of the findings by ensuring that patients from different regions, demographic groups, and treatment centers had an equal chance of inclusion. Data Description and data sources The study utilized secondary, routinely collected, anonymized patient data from the Uganda Electronic Medical Records (EMR) system under the Ministry of Health (MOH). Clinical and demographic data: Retrieved from the Uganda EMR system, containing variables on patient demographics, TB disease classification, ART initiation, and health facility characteristics. Study variables The primary outcome of the study is TB treatment outcome, categorized as successful (including cure and treatment completion) or unsuccessful (comprising failure, default, death, and not evaluated). Predictor variables will be in categories: socio-demographic (age, sex, residence), patient-related (e.g., initiation of ART, TB disease classification), health system-related (e.g., distance to the facility, facility ownership). Data preprocessing and exploratory data analysis (EDA) Data cleaning and handling missing data Missing data were carefully assessed before deciding whether to exclude or impute values. The extent and pattern of missingness were analyzed using descriptive statistics and correlation matrices to determine whether the missing data were completely random (MCAR), missing at random (MAR), or missing not at random (MNAR). Variables with more than 30% missing values were excluded if they had limited predictive relevance, whereas those with lower levels of missingness underwent appropriate imputation based on their type and missing data pattern. Categorical variables with missing values that were missing completely at random (MCAR) were imputed using mode imputation. For each categorical variable, the most frequently occurring category (mode) was identified from the observed data, and all missing entries within that variable were replaced with this modal category. This approach assumed that the missing values had the same distribution as the observed data and helped maintain the original variable structure without introducing bias. For numerical variables, mean or median imputation was applied if the data were normally distributed, whereas K-Nearest Neighbors (KNN) or Multiple Imputation by Chained Equations (MICE) were used for non-normal distributions or MNAR patterns. Sensitivity analyses were performed to compare different imputation methods and ensure minimal bias while maintaining predictive accuracy. Data transformation and normality checks Categorical variables (e.g., treatment success/failure, sex, ART status) were encoded and Continuous variables (e.g., age, BMI) were checked for normality using histograms and the Shapiro-Wilk test. If non-normal, transformations (e.g., log transformation) were applied. Outlier detection and treatment Outliers were detected using box plots, Z-scores, and interquartile range (IQR) methods and Winsorization were considered for extreme outliers that significantly affect model performance. Correlation analysis and feature selection Multicollinearity was assessed using the variance inflation factor (VIF) with a threshold of 5 to identify redundant predictors. A correlation matrix and heatmaps were employed to examine dependencies among variables. Feature selection was performed using recursive feature elimination (RFE) and feature importance scores derived from tree-based models. Sample size determination The sample size was determined based on the proportion of unsuccessful TB treatment cases (30%) and on a power analysis to ensure at least 80% statistical power. The final analysis was conducted using data from 5,062 participants, which provided sufficient power for the planned statistical and machine learning analyses. Machine Learning Models Model selection criteria The study evaluated the performance of seven supervised algorithms, logistic regression, random forest, support vector machine and AdaBoost algorithms. Logistic regression offered a straightforward linear approach, while random forest, support vector machine and AdaBoost provide more flexible solutions for complex data. Through this analysis, we aimed at identifying the most effective algorithm for predicting TB treatment outcomes. See Table 1 . Table 1 Machine Learning Algorithms and the rationale for their selection in the model Algorithm Reason for Selection Logistic Regression Baseline model for comparison, interpretable results Random Forest Handles non-linear relationships, robust to outliers Support Vector Machine (SVM) Effective for complex, high-dimensional datasets AdaBoost (Adaptive Boosting) Enhances weak classifiers, improves prediction accuracy Model building The model building methodology involved thorough data preparation, a 70/30 random split into training and test sets, and balanced sampling to ensure equal representation of treatment outcomes. A classifier algorithm was then trained using the training data to establish an optimal predictive model, which was evaluated using the independent test dataset. Python, a freely available programming language, along with TensorFlow libraries, an open-source machine learning framework were used. Hyperparameter optimization Hyperparameter tuning was performed to optimize model performance. Random Forest modal was fine-tuned by adjusting the number of trees, max depth, and minimum samples per split to balance complexity and accuracy. SVM was optimized by selecting the best kernel type (linear or RBF) and regularization parameter (C). AdaBoost was refined by tuning the learning rate and number of estimators to enhance model stability. Logistic Regression was optimized by adjusting L1 and L2 regularization strength to prevent overfitting. The optimizations were conducted using Grid Search CV and Randomized Search CV for efficient parameter selection. See Table 2 . Table 2 Machine learning algorithms and corresponding hyperparameters tuned during model optimisation Algorithm Hyperparameters to Tune Random Forest Number of trees, max depth, min samples split SVM Kernel type (linear, RBF), regularization parameter (C) AdaBoost Learning rate, number of estimators Logistic Regression Regularization strength (L1, L2) Model evaluation The evaluation of the model's performance was based on accuracy (the proportion of correctly classified observations in the unseen test set), sensitivity (the proportion of known failure outcomes correctly identified by the algorithm), precision (the proportion of predicted failure outcomes that match known failures in the test set), and specificity (the proportion of known successful outcomes correctly identified). Additionally, the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was used to assess the overall predictive classification performance (The AUC represents the probability that the model will correctly distinguish between a randomly chosen positive case and a randomly chosen negative case). A range of 0.5 to 1.0 on the AUC scale indicates the predictive power of the model, where 0.5 suggests no predictive ability and 1.0 indicates perfect predictive performance. See Table 3 . Table 3 Model evaluation metrics, their definitions, and interpretations Metric Definition Interpretation Accuracy (TP + TN) / (TP + TN + FP + FN) Measures overall correctness Precision TP / (TP + FP) How many predicted positive cases were true positives Recall (Sensitivity) TP / (TP + FN) How many actual positive cases were correctly identified Specificity TN / (TN + FP) Correct identification of negative cases F1-Score 2 × (Precision × Recall) / (Precision + Recall) Balance between precision and recall AUC-ROC Score Area under ROC curve Discriminative ability of the model Results Baseline characteristics of study participants Table 4 presents the baseline characteristics of patients included in the analysis. We included 5,062 tuberculosis (TB) patients with a mean age of 37.2 years (SD 14.5). Among these, 3,850 (76.1%) achieved treatment success, while 1,212 (23.9%) experienced treatment failure. Men comprised 54.8% of the study population and showed a higher proportion of treatment failure than women (56.4% vs 43.6%). Approximately one-third (34.2%) of the participants were married. Most patients were classified as having pulmonary bacteriologically confirmed TB (61.1%), followed by pulmonary clinically diagnosed TB (33.7%) and extrapulmonary TB (5.2%). The majority were new TB cases (92.9%), while 5.3% were relapse cases and 1.5% had returned to treatment after loss to follow-up. The non-digital community-based directly observed treatment (DOT) model was the predominant treatment-support approach reported by 68.8%, followed by facility-based DOT (16.9%) and digital community DOT (14.3%). Treatment failure was most frequent among patients managed under the facility-based DOT model (20.1%). Regarding ART status, 49.8% of participants were already on ART at baseline, 22.7% were newly initiated during TB treatment, and 27.5% had unknown ART status. Treatment outcomes did not vary substantially across ART categories. See Table 4 . Table 4 Baseline characteristics of patients Variables Total TB Treatment outcome Treatment success Treatment failure n = 5062(%) n = 3850(%) n = 1212(%) Sex Female 2289 (45.2) 1760 (45.7) 529 (43.6) Male 2773 (54.8) 2090 (54.3) 683 (56.4) Marital status Not married 3332 (65.8) 2491 (64.7) 841 (69.4) Married 1730 (34.2) 1359 (35.3) 371 (30.6) Age in years mean(sd) 37.2 (14.5) 37.2 (14.4) 37.2 (14.9) TB Disease Classification (Baseline) Extra pulmonary (EP) TB 262 (5.2) 160 (4.2) 102 (8.4) Pulmonary Clinically Diagnosed (P-CD) TB 1708 (33.7) 1288 (33.5) 420 (34.7) Pulmonary bacteriologically confirmed (P-BC)TB 3092 (61.1) 2402 (62.4) 690 (56.9) Type of patient New 4704 (92.9) 3581 (93.0) 1123 (92.7) Relapse 266 (5.3) 205 (5.3) 61 (5.0) Return after lost to follow up 78 (1.5) 54 (1.4) 24 (2.0) Treatment after failure 11 (0.2) 7 (0.2) 4 (0.3) Treatment history Unknown 3 (0.1) 3 (0.1) 0 (0.0) TB Treatment Model Digital Community DOT 726 (14.3) 564 (14.6) 162 (13.4) Facility DOT 855 (16.9) 611 (15.9) 244 (20.1) Non-Digital Community DOT 3481 (68.8) 2675 (69.5) 806 (66.5) DS-TB regimen 2RHZE/ 10 RH 378 (7.5) 266 (6.9) 112 (9.2) 2RHZE/ 4RH 4684 (92.5) 3584 (93.1) 1100 (90.8) ART Status Already on ART 2521 (49.8) 1908 (49.6) 613 (50.6) Newly initiated on Art 1149 (22.7) 878 (22.8) 271 (22.4) Unknown 1392 (27.5) 1064 (27.6) 328 (27.1) Model performance Figure 1 presents the comparative performance of machine learning models developed to predict TB treatment outcomes among HIV/TB co-infected patients. Model performance varied across discrimination, recall, and accuracy metrics. The stacked ensemble model demonstrated the strongest overall performance, with a recall for treatment failure of 0.60, ROC-AUC of 0.669, and overall accuracy of 0.618. Logistic regression models with class balancing (L1/Lasso and L2/Ridge) achieved similar performance, detecting 58–59% of failures, with ROC-AUC values around 0.65 and accuracy between 0.61 and 0.62. The tuned Random Forest model yielded the highest ROC-AUC (0.675) among individual algorithms, maintaining balanced recall for treatment failures (0.58) and successes (0.67). The tuned Gradient Boosting Machine (GBM) achieved the highest accuracy (0.768) and near-perfect recall for successes (0.99), but its sensitivity to failures was low (recall = 0.07). Baseline Random Forest and Gradient Boosting models with SMOTE achieved accuracies of 0.709 and 0.715, respectively, but recalled ≤ 0.32 of failures. The Support Vector Machine (SVM) model exhibited the weakest discrimination (ROC-AUC = 0.466). AdaBoost with SMOTE achieved moderate accuracy (0.672) and recall for successes (0.77) but detected only 35% of failures. See Fig. 1 . Model calibration Figure 2 shows the calibration and precision-recall performance of the evaluated models. Across models, ROC-AUC values during cross-validation ranged from 0.637 to 0.655, reflecting fair discrimination. Precision-recall AUC (PR-AUC) values were consistently high (≥ 0.836), indicating that model predictions of treatment failure were generally reliable when made. The tuned Random Forest achieved the highest PR-AUC (0.848), although calibration plots indicated slight overconfidence. Logistic regression models and the stacked ensemble exhibited moderate calibration with minor deviation from the ideal line at higher predicted probabilities. The tuned GBM model achieved the best calibration performance (Brier score = 0.172), although this improvement did not enhance failure detection. See Fig. 2 Feature importance analysis Figure 3 presents the relative importance of predictors of TB treatment outcomes derived from the Random Forest model. Age was the most influential predictor, accounting for nearly half of the model’s predictive strength. Other key demographic and clinical predictors included sex (male), marital status (married), ART status (unknown or newly initiated), TB classification (pulmonary bacteriologically confirmed and clinically diagnosed), and treatment model (community vs facility DOT). Programmatic variables, such as hospital care and treatment regimen (2RHZE/4RH), also contributed moderately to model predictions. Logistic regression analysis further identified facility-specific and clinical predictors associated with treatment success. Facilities including Mbale (OR = 2.82), Naguru (OR = 2.47), Hoima (OR = 2.25), Gulu (OR = 2.13), and Lira (OR = 1.98) Regional Referral Hospitals exhibited higher odds of treatment success. In contrast, Soroti (OR = 0.49), Moroto (OR = 0.49), Masaka (OR = 0.57), and St. Kizito Matany (OR = 0.63) hospitals had lower odds. Patients with pulmonary bacteriologically confirmed TB (OR = 2.15) and clinically diagnosed TB (OR = 1.64) were more likely to achieve treatment success, whereas those classified as “treatment after failure” (OR = 0.51) or managed under facility-based DOT (OR = 0.68) had lower odds. See Fig. 3 . Comparison of feature importance across models Figure 4 compares normalised feature importance across Random Forest, GBM, and Logistic Regression models. Age consistently emerged as the most influential predictor across all models. Random Forest and GBM models additionally emphasised sex (male), ART status (unknown or newly initiated), marital status (married), X-ray result (unknown), and treatment model (non-digital community DOT). The Logistic Regression model identified male sex (OR = 0.33), unknown ART status (OR = 0.40), and newly initiated ART (OR = 0.37) as factors associated with lower odds of treatment success. Patients under the non-digital community DOT model (OR = 0.43) also exhibited poorer outcomes. See Fig. 4 . Partial dependence analysis Figure 5 illustrates the partial dependence of age on the predicted probability of TB treatment success. The relationship was non-linear, with younger patients (< 20 years) demonstrating the highest predicted probabilities of treatment success. The effect plateaued during early adulthood and declined progressively beyond 40 years, reaching the lowest probabilities among patients aged 70 years or older. See Fig. 5 . Discussions The study evaluated the predictive capacity of several machine learning algorithms, including logistic regression, Random Forest, Gradient Boosting Machine (GBM), Support Vector Machine (SVM), AdaBoost, and a stacked ensemble modelling forecasting tuberculosis (TB) treatment outcomes among HIV/TB co-infected patients. Model performance was evaluated across key metrics of discrimination, recall, accuracy, and calibration to determine both their predictive power and clinical applicability. Furthermore, feature importance analysis was used to identify the most influential predictors of treatment outcomes, while partial dependence analysis illustrated the marginal effects of specific predictors, particularly age, on treatment success probabilities. These findings indicate that the stacked ensemble and class-balanced logistic regression models achieved the most favourable balance between discrimination, calibration, and recall for treatment failures, supporting their application in programmatic contexts where accurate identification of patients at risk of poor outcomes is critical. The Random Forest model achieved the highest ROC-AUC, indicating strong discriminative ability, whereas the GBM attained the highest overall accuracy but was less sensitive to treatment failures. Across all models, key predictors of TB treatment outcome included age, ART status, sex, marital status, TB classification, and treatment modality. The partial dependence analysis further revealed a clear decline in treatment success with increasing age, particularly beyond 40 years, highlighting the clinical relevance of demographic and programmatic factors in predicting TB treatment outcomes. Model performance and predictive utility The study revealed that the stacked ensemble and class-balanced logistic regression models provided the most balanced and clinically useful predictions of TB treatment outcomes. These models were able to handle the imbalance between treatment success and failure effectively, achieving moderate discrimination and calibration. Their performance reflects the ability of ensemble approaches to combine complementary strengths of multiple algorithms, while regularized logistic models remain robust in managing collinearity and mixed data types. Comparable studies have reported similar findings, where ensemble and logistic regression models achieved an optimal balance between interpretability and accuracy in TB outcome prediction [ 10 , 11 ]. This consistency supports the application of these models in real-world TB programs, where interpretability and operational feasibility are as crucial as accuracy. Discrimination of individual models The tuned Random Forest exhibited the highest ROC-AUC among single models, while the tuned GBM yielded the highest overall accuracy but failed to adequately detect treatment failures. The Random Forest’s superior discrimination may be attributed to its ability to capture complex, non-linear interactions among heterogeneous predictors. In contrast, the GBM’s overemphasis on optimising overall accuracy led to overfitting to the dominant outcome class, treatment success, resulting in poor sensitivity to failures. This trade-off between accuracy and recall has been consistently documented in previous studies [ 12 , 13 ], highlighting the need for careful tuning of machine learning models when dealing with imbalanced clinical datasets. We found that age was the most influential predictor of TB treatment outcomes. The partial dependence analysis indicated a non-linear relationship between age and the probability of treatment success, with younger patients achieving higher success rates than older adults, particularly those aged 40 years and above, showing declining probabilities. This trend is biologically plausible, as older patients often have weakened immunity, higher prevalence of comorbidities, and greater treatment fatigue, which collectively undermine adherence and clinical response. Studies conducted in Uganda and elsewhere corroborate this finding, demonstrating that older age is associated with increased risk of poor TB outcomes[ 14 , 15 ]. These observations emphasize the need for age-sensitive treatment support and follow-up strategies. We also found that the ART status of these patients significantly influenced treatment outcomes. Patients who were newly initiated on ART or had unknown ART status were more likely to experience treatment failure compared to those already established on ART at baseline. This finding likely reflects late initiation of HIV care, poor linkage between TB and ART services, or incomplete treatment documentation. Unknown ART status also signals potential programmatic weaknesses in data recording or continuity of care. Similar associations have been reported in Ethiopia and Uganda, where delayed or interrupted ART initiation was a major contributor to unfavourable TB outcomes [ 16 ]. Strengthening ART–TB integration and ensuring early initiation of ART remain key programmatic priorities. The study also found that male patients were more likely to experience treatment failure compared to female patients. This difference could be attributed to gendered disparities in health-seeking behaviour, adherence, and social support. Men are often diagnosed later, face occupational mobility challenges, and may have weaker support systems during treatment. Consistent with this, [ 17 ] and [ 14 ] observed that male patients had lower treatment success and higher default rates in Uganda. These findings underscore the importance of incorporating gender-sensitive approaches into TB programs, including flexible clinic schedules and targeted counselling for men. In addition, marital status was also associated with treatment outcomes, with unmarried individuals experiencing higher failure rates. Married participants likely benefit from stronger social and emotional support, enhancing adherence and clinic attendance. Similar associations between marital status and TB treatment success have been reported in sub-Saharan Africa [ 3 ], emphasizing the role of social support structures in sustaining adherence. Patients with pulmonary bacteriologically confirmed or clinically diagnosed TB achieved better outcomes compared with those categorized under retreatment or treatment-after-failure groups. This likely reflects the benefits of early detection and diagnostic certainty, which facilitate appropriate treatment initiation and monitoring. In contrast, retreatment and treatment-after-failure categories typically include patients with prior drug resistance, adherence difficulties, or advanced disease. Comparable findings were reported by [ 3 ] and [ 16 ], who observed that diagnostic category and treatment history were strong predictors of TB treatment success in African settings. These results reinforce the importance of early diagnosis and differentiated care models for high-risk or previously treated patients. The facility-based DOT model was associated with higher rates of treatment failure compared with community-based DOT models. This could be due to limited patient follow-up in busy facility settings and the convenience and social support offered by community-based models. Community DOT allows treatment supervision closer to patients’ homes, reducing transportation barriers and improving adherence. Similar findings were observed in systematic reviews by [ 18 ] and Ugandan programmatic studies by [ 19 ], which showed superior treatment success under community-based approaches. However, [ 20 ]reported reduced success rates among previously treated patients managed under digital DOT, suggesting that the effectiveness of DOT interventions depends on context, implementation fidelity, and patient characteristics. Our results highlight the need to refine digital and facility-based DOT strategies to ensure optimal adherence support and supervision. Model calibration and reliability Model calibration and precision, recall analyses demonstrated fair discrimination and high precision for predicted failures, suggesting that when the models predicted a failure, they did so with high reliability. However, some models, particularly GBM, showed reduced recall, indicating difficulty in identifying all failure cases. This challenge arises from the inherent class imbalance in programmatic datasets, where treatment success overwhelmingly dominates. Recent studies emphasize that in clinical prediction, calibration metrics such as the Brier score and PR-AUC are as important as ROC-AUC for assessing real-world utility [ 11 , 13 ]. The fair calibration observed in this study suggests that the models provide reasonably accurate probabilities that could be integrated into decision-support systems to identify high-risk patients. Strengths and Limitations This study benefited from a large, nationally representative large dataset that enhanced the reliability and generalizability of model findings. The application of multiple machine learning algorithms allowed for robust performance comparisons, while inclusion of calibration and interpretability analyses strengthened clinical relevance. Feature importance and partial dependence analyses further added explanatory value by identifying key drivers of treatment outcomes. Nevertheless, several limitations should be acknowledged. The retrospective design limited control over unmeasured confounders such as nutritional status, adherence behaviours, and socioeconomic conditions. Missing data, especially for ART status, may have introduced bias in model training. The imbalance between treatment success and failure restricted the models’ sensitivity to failures, even after applying balancing techniques. Additionally, lack of external validation constrains the generalizability of findings to other settings. Conclusions This study demonstrates that machine learning models, particularly ensemble and regularised logistic regression approaches, can moderately predict TB treatment outcomes among HIV/TB co-infected patients in Uganda. The models achieved fair discrimination and calibration, highlighting their potential utility in programmatic risk assessment, although their performance remains insufficient for autonomous clinical decision-making. Key predictors of treatment success identified across models included age, ART status, sex, marital status, TB classification, and treatment model. Among these, age and ART status were the most influential variables, reflecting the complex interplay between biological, clinical, and behavioural factors in determining treatment outcomes. The findings also revealed that older adults, males, and patients newly initiated or with unknown ART status were at higher risk of treatment failure. These insights highlighted the importance of integrating data-driven predictive models into routine TB and HIV program management to enable early identification and targeted support for patients at risk of poor outcomes. Overall, this study adds to the existing body of evidence on the application of machine learning in TB treatment monitoring and offers a structured framework for the development of predictive tools designed to strengthen existing clinical and surveillance systems, particularly in resource-limited settings. Recommendations Based on the study findings, several recommendations are proposed to advance both research and practice. For researchers, future investigations should prioritize prospective and external validation of the best-performing models using independent datasets from diverse geographical regions in Uganda. Such validation will enhance the generalizability and robustness of predictive models, allowing them to be effectively adapted to various programmatic contexts. Researchers should also consider including additional clinical, behavioural, and socioeconomic variables, such as HIV viral load, adherence patterns, nutritional indicators, and income levels, in future model development to improve predictive accuracy and interpretability. For the Ministry of Health (MOH) and the National TB and Leprosy Program (NTLP), integration of predictive models into electronic medical record systems and digital health dashboards within national TB control programs should be prioritized. Embedding predictive analytics in these systems would enable real-time identification of high-risk patients, facilitate data-driven decision-making, and strengthen monitoring and evaluation of treatment outcomes. For health workers and program implementers, TB control programs should strengthen community-based and digital directly observed treatment (DOT) models to improve adherence and continuity of care. Tailored follow-up strategies, such as home-based visits and mobile health interventions, should be expanded to support patients who face structural or socioeconomic barriers to treatment completion. Finally, for policymakers and program managers, targeted interventions should focus on vulnerable subgroups, particularly older adults, men, and patients newly initiated or with unknown ART status. Differentiated adherence counselling, personalised treatment support, and closer clinical monitoring for these groups can reduce treatment failure, enhance survival, and contribute to Uganda’s broader TB and HIV control goals. Abbreviations AUC Area Under the Curve ART Antiretroviral Therapy DOT Directly Observed Therapy DS TB–Drug–Susceptible Tuberculosis EDA Exploratory Data Analysis EMR Electronic Medical Records EP Extra–Pulmonary F1 Score–Harmonic Mean of Precision and Recall GBM Gradient Boosting Machine HIV Human Immunodeficiency Virus IQR Interquartile Range KNN K–Nearest Neighbors L1 Lasso Regularization L2 Ridge Regularization MAR Missing At Random MCAR Missing Completely At Random MDR TB–Multidrug–Resistant Tuberculosis MICE Multiple Imputation by Chained Equations ML Machine Learning MNAR Missing Not At Random MOH Ministry of Health NTLP National TB and Leprosy Program PLWH People Living With HIV PR AUC–Precision–Recall Area Under the Curve RBF Radial Basis Function REC Research Ethics Committee RFE Recursive Feature Elimination ROC Receiver Operating Characteristic ROC AUC–Area Under the Receiver Operating Characteristic Curve RRH Regional Referral Hospital SD Standard Deviation SDGs Sustainable Development Goals SMOTE Synthetic Minority Oversampling Technique SVM Support Vector Machine TB Tuberculosis TN True Negative TP True Positive TSR Treatment Success Rate VIF Variance Inflation Factor WHO World Health Organization XDR TB–Extensively Drug–Resistant Tuberculosis Declarations Ethics approval and consent to participate Ethical approval for this study was obtained from the Makerere University School of Public Health Research Ethics Committee, Approval No . 542 . This study involved a secondary analysis of routinely collected, de-identified clinical data, and no direct interaction with human participants occurred. All study procedures were conducted in accordance with applicable ethical guidelines and regulations for research involving human participants. A waiver of informed consent was granted by the Research Ethics Committee due to the use of de-identified data. Data confidentiality and participant privacy were strictly maintained throughout all stages of the research process. Consent for publication Not applicable Availability of data and materials The datasets used and/or analysed during the current study are available from the corresponding author upon reasonable request. Competing interests The authors declare that they have no competing interests. Funding There was no funding for this study Authors' contributions AO conceptualised and designed the study. JPA participated in data analysis, report writing, and drafting of the manuscript. AO and FO conducted data analysis. MN, EOE, EI, NC, and AK contributed to data acquisition, data management, and preliminary analyses. JPA, SRA, TN, RN, FO, and NRAO provided methodological and subject-matter expertise and critically reviewed and revised the manuscript for important intellectual content. RK supervised and reviewed the manuscript. NK provided overall supervision and reviewed the manuscript for scientific soundness. All authors read and approved the final version of the manuscript. Acknowledgements The authors gratefully acknowledge the Uganda Ministry of Health, Division of Health Information, for providing access to the data used in this study. References WHO, Tuberculosis. https://www.who.int/news-room/fact-sheets/detail/tuberculosis (2025). World Health Organization. The End Strategy TB. End TB Strateg. 2015;53:1689–99. Teferi MY, El-Khatib Z, Boltena MT et al. Tuberculosis treatment outcome and predictors in africa: A systematic review and meta-analysis. Int J Environ Res Public Health; 18. Epub ahead of print 2021. 10.3390/ijerph182010678 WHO. World Tuberculosis Day. 2025: Uniting to End TB in Uganda. https://www.afro.who.int/countries/uganda/news/world-tuberculosis-day-2025-uniting-end-tb-uganda (2023). MOH. Uganda National TB and Leprosy Program Republic of Uganda Ministry of Health . 2020. Ali SA, Mavundla TR, Fantu R et al. Outcomes of TB treatment in HIV co-infected TB patients in Ethiopia: a cross-sectional analytic study. BMC Infect Dis 2016; 1–9. MoH. Acceleration of HIV prevention in Uganda: A road map towards zero new infections by 2030. WHO. Global tuberculosis report, 2020 . 2021. Izudi J, Tamwesigire IK, Bajunirwe F. Explaining the successes and failures of tuberculosis treatment programs; a tale of two regions in rural eastern Uganda. 2019; 2: 1–10. Gichuhi HW, Magumba M, Kumar M, et al. A machine learning approach to explore individual risk factors for tuberculosis treatment non-adherence in Mukono district. PLOS Glob Public Heal. 2023;3:1–20. Hosu MC, Faye LM, Apalata T. Optimizing Drug-Resistant Tuberculosis Treatment Outcomes in a High HIV-Burden Setting: A Study of Sputum Conversion and Regimen Efficacy in Rural South Africa. Pathogens ; 14. Epub ahead of print 2025. 10.3390/pathogens14050441 Wang K, Wang Z, Li Z et al. Oriented Object Detection in Optical Remote Sensing Images using Deep Learning: A Survey. Epub ahead of print 2023. 10.1007/s10462-025-11256-0 Zhang L, Yu T, Zheng G, et al. Using machine learning to predict selenium content in crops: Implications for soil health and agricultural land utilization in longevity regions. Sci Total Environ. 2025;964:178520. Kirenga BJ, Ssengooba W, Muwonge C, et al. Tuberculosis risk factors among tuberculosis patients in Kampala, Uganda: Implications for tuberculosis control. BMC Public Health. 2015;15:1–7. Omara G, Bwayo D, Mukunya D, et al. Tuberculosis treatment success rate and its predictors among TB HIV co-infected patients in East and North Eastern Uganda. Sci Rep. 2025;15:5532. Omara G, Bwayo D, Mukunya D, et al. Tuberculosis treatment success rate and its predictors among TB HIV co-infected patients in East and North Eastern Uganda. Sci Rep. 2025;15:5532. Baluku JB, Mukasa D, Bongomin F, et al. Gender differences among patients with drug resistant tuberculosis and HIV co-infection in Uganda: a countrywide retrospective cohort study. BMC Infect Dis. 2021;21:1–11. Wright DM, Reid N, Montgomery WI, et al. Herd-level bovine tuberculosis risk factors: Assessing the role of low-level badger population disturbance. Sci Rep. 2015;5:1–11. Makabayi-Mugabe R, Musaazi J, Zawedde-Muyanja S, et al. Community-based directly observed therapy is effective and results in better treatment outcomes for patients with multi-drug resistant tuberculosis in Uganda. BMC Health Serv Res. 2023;23:1–12. Izudi J, Okello G, Bajunirwe F. Low treatment success rate among previously treated persons with drug-susceptible pulmonary tuberculosis in Kampala, Uganda. J Clin Tuberc Other Mycobact Dis. 2023;32:100375. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 02 May, 2026 Reviewers agreed at journal 12 Mar, 2026 Reviewers invited by journal 06 Mar, 2026 Editor invited by journal 23 Feb, 2026 Editor assigned by journal 22 Feb, 2026 Submission checks completed at journal 22 Feb, 2026 First submitted to journal 13 Feb, 2026 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-8870898","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":603054472,"identity":"21245669-e15b-40a8-9f16-242b721e1af2","order_by":0,"name":"Ambrose Okibure","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIiWNgGAWjYFACHgglwQxm2AAxY+MBUrSkgbQ0EKkFwjgM5uDVwj/t7LEPH/7YyEm28x6Trqg4b7e2/TDQlhqbaFxaJG7nJc+c2ZZmLM3MlyZ55szt5G1nEoFajqXlNuDSczvHmJm34XDiPGYeM8nGttvJZgeAWhgbDuPUIg/SwvPnP1TLv3PJZucf4tdiANbCdiBxNlhLwwE7sxsEbDEE+oVxZluysWQzX7Jlw7HkBLMbQFsS8PhF7nbuYYYPf+zkJM6fPXizocbO3ux8+sMHH2pscHsfHSSCVSYQqxwE7ElRPApGwSgYBSMDAADDm19O7fwtwQAAAABJRU5ErkJggg==","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":true,"prefix":"","firstName":"Ambrose","middleName":"","lastName":"Okibure","suffix":""},{"id":603054473,"identity":"3dbf40c2-3414-4b47-b22f-b399eade6fb5","order_by":1,"name":"Jimmy Patrick Alunyo","email":"","orcid":"","institution":"Busitema University","correspondingAuthor":false,"prefix":"","firstName":"Jimmy","middleName":"Patrick","lastName":"Alunyo","suffix":""},{"id":603054474,"identity":"dc807297-c2e7-4936-9273-fc6f01bbd1c6","order_by":2,"name":"Sarah Rachael Akello","email":"","orcid":"","institution":"Busitema University","correspondingAuthor":false,"prefix":"","firstName":"Sarah","middleName":"Rachael","lastName":"Akello","suffix":""},{"id":603054475,"identity":"fcad3047-9e55-4234-b704-f0dcbe030a86","order_by":3,"name":"Tereza Nyapendi","email":"","orcid":"","institution":"AIDS Information Centre","correspondingAuthor":false,"prefix":"","firstName":"Tereza","middleName":"","lastName":"Nyapendi","suffix":""},{"id":603054476,"identity":"7a3e28f2-06b8-49b9-8655-b649a4b7444f","order_by":4,"name":"Mabel Nakawooya","email":"","orcid":"","institution":"Ministry of Health","correspondingAuthor":false,"prefix":"","firstName":"Mabel","middleName":"","lastName":"Nakawooya","suffix":""},{"id":603054477,"identity":"8e66eb6c-2555-46d7-a341-42bbc911bb4c","order_by":5,"name":"Francis Okello","email":"","orcid":"","institution":"Busitema University","correspondingAuthor":false,"prefix":"","firstName":"Francis","middleName":"","lastName":"Okello","suffix":""},{"id":603054480,"identity":"c9dbc572-28d7-4ebb-99fa-cf8f6622c72f","order_by":6,"name":"Noela Regina Akwi Okalany","email":"","orcid":"","institution":"University of Bergen","correspondingAuthor":false,"prefix":"","firstName":"Noela","middleName":"Regina Akwi","lastName":"Okalany","suffix":""},{"id":603054482,"identity":"d4ff819b-be90-4587-96f8-be80f5d7a0f4","order_by":7,"name":"Rebecca Nekaka","email":"","orcid":"","institution":"Busitema University","correspondingAuthor":false,"prefix":"","firstName":"Rebecca","middleName":"","lastName":"Nekaka","suffix":""},{"id":603054485,"identity":"6bae26e4-a652-42fc-a081-e6c988844337","order_by":8,"name":"Ernest Ochepa Ekiru","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Ernest","middleName":"Ochepa","lastName":"Ekiru","suffix":""},{"id":603054486,"identity":"354e0142-c648-4968-866f-42fa542ea4fa","order_by":9,"name":"Winfred Jenifer Namuyanja","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Winfred","middleName":"Jenifer","lastName":"Namuyanja","suffix":""},{"id":603054487,"identity":"be267439-e3a4-4871-a9c8-954330661ec1","order_by":10,"name":"Edward Ikoona","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Edward","middleName":"","lastName":"Ikoona","suffix":""},{"id":603054488,"identity":"d3daf33a-5ea5-4912-8f42-d9b6c12cc34d","order_by":11,"name":"Nancy chemutai","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Nancy","middleName":"","lastName":"chemutai","suffix":""},{"id":603054489,"identity":"4b7ab229-1891-4440-921c-2889a6cc759b","order_by":12,"name":"Azaria Kanyamibwa","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Azaria","middleName":"","lastName":"Kanyamibwa","suffix":""},{"id":603054490,"identity":"5cc1de32-3d25-4691-949c-64f6fe397bf8","order_by":13,"name":"Ronald Kusolo","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Ronald","middleName":"","lastName":"Kusolo","suffix":""},{"id":603054492,"identity":"746c0072-4d04-407d-be1f-e3bd567cbe3a","order_by":14,"name":"Noah Kiwanuka","email":"","orcid":"","institution":"Makerere School of Public Health","correspondingAuthor":false,"prefix":"","firstName":"Noah","middleName":"","lastName":"Kiwanuka","suffix":""}],"badges":[],"createdAt":"2026-02-13 10:54:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8870898/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8870898/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104547327,"identity":"69605f13-be44-40cf-a509-8b4ac0b188a5","added_by":"auto","created_at":"2026-03-13 07:35:06","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":110371,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of Machine Learning Models for TB Treatment Outcomes\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/54f526c0c2ba0c1d14273536.png"},{"id":104547328,"identity":"6084b68a-1fa1-46e7-95cd-349ae0650140","added_by":"auto","created_at":"2026-03-13 07:35:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":357803,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration and precision-recall analysis of machine learning models.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/b5bfa067b60b9717adaf9d8d.png"},{"id":104547330,"identity":"60351427-8c67-4e34-8899-829e3b4ff84f","added_by":"auto","created_at":"2026-03-13 07:35:06","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":490155,"visible":true,"origin":"","legend":"\u003cp\u003eTop 15 predictors of TB treatment outcomes identified by the Random Forest model\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/732af501a62184a15ea3fdbb.png"},{"id":104547329,"identity":"c9c862e3-349f-4ffe-bf46-19df40cd26a3","added_by":"auto","created_at":"2026-03-13 07:35:06","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":412019,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of feature importance across machine learning models\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/6490d69f6a67d0c728ace2e9.png"},{"id":104547331,"identity":"57ce0cd2-d607-4289-b797-a5e64f4616a5","added_by":"auto","created_at":"2026-03-13 07:35:06","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":87533,"visible":true,"origin":"","legend":"\u003cp\u003ePartial dependence plot of age and predicted probability of TB treatment success\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/1bbbabc57ae035cdcd1bbfb1.png"},{"id":104781345,"identity":"7b3747aa-e76e-41c0-820c-e74a5ea9dfae","added_by":"auto","created_at":"2026-03-17 07:55:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2633868,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8870898/v1/ccd82f33-9a8d-440b-8bf2-17f0dcbd0f1c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach","fulltext":[{"header":"Background","content":"\u003cp\u003eGlobally, tuberculosis (TB) remains among the major causes of morbidity and mortality. Approximately an estimated 10.8\u0026nbsp;million people fell ill with TB worldwide, including 6.0\u0026nbsp;million men, 3.6\u0026nbsp;million women and 1.3\u0026nbsp;million children in 2023 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The mortality rate for TB disease was disproportionately high, 1.25\u0026nbsp;million among HIV-negative people compared to 161,000 in HIV-positive people in 2023 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The health target of the United Nations Sustainable Development Goals (SDGs) is to end the TB epidemic by 2030 [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The burden of TB is high in the African continent, with over 2.5\u0026nbsp;million people reported diagnosed with TB in the African region in 2022, accounting for a quarter of new TB cases worldwide. The African continent accounted for over 90% of all global TB cases and 87% of all global co-infection cases. The region contributed more than 400,000 (33%) deaths to the global mortality from the disease in the same year [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe World Health Organization (WHO) recommends that early diagnosis and appropriate treatment of tuberculosis (TB) can effectively reduce transmission, avert deaths, and contribute to its elimination. For drug-susceptible TB, a standard six-month course of first-line anti-TB drugs is recommended, while treatment duration may be longer for multidrug-resistant (MDR) or extensively drug-resistant (XDR) TB. Between 2000 and 2020, early diagnosis and treatment of TB were estimated to have averted 66\u0026nbsp;million deaths [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Despite widespread access to TB prevention, diagnosis, treatment and care services, the global treatment success rate was only 86% among HIV-negative people in 2019. For HIV-associated TB, it was even very low at 77%. Sub-Saharan Africa accounts for approximately 90% of global tuberculosis (TB) cases and about 87% of TB/HIV co-infections. The treatment success rate in the region stands at 78.9% and is even lower among patients co-infected with HIV and TB [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. To address the persistently high TB-related mortality, there is a growing need for interventions aimed at predicting treatment outcomes. Machine learning algorithms offer promising approaches for predicting TB treatment success and improving patient outcomes\u003c/p\u003e \u003cp\u003eIn Uganda, by 2023, TB disease was still a big public health challenge, with more than 200 people falling ill with TB disease and approximately 30 losing their lives [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Even though 94% of TB/HIV co-infected patients initiated antiretroviral therapy (ART), there was a high mortality rate of 12% [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Early detection of TB disease and prompt initiation of treatment with potent anti-TB drug regimens are required to decrease morbidity and mortality from TB disease among HIV-infected patients [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite implementing these recommendations and improving access to TB prevention, diagnosis, treatment, and care, Uganda still has a very low TSR of 70% among TB/HIV co-infections and 72% among HIV-negative people[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Additionally, the country is also ranked among the thirty nations globally grappling with high burdens of TB and TB/HIV co-infection [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Annually, an estimated 91,000 individuals in Uganda contract TB, with 32% of these cases occurring in people living with HIV. Additionally, two percent of TB patients face drug-resistant strains that are not responsive to initial medication, and around 15% of TB cases in Uganda involve children under the age of 14 [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. This Treatment Success Rate is way below the desired WHO target of \u0026ge;\u0026thinsp;90% [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTB treatment is typically successful with the standard six-month course of anti-TB drugs. However, treatment outcomes vary widely due to clinical, socio-demographic, environmental, and health-system-related factors. Studies show that treatment success is influenced by patient-related factors (e.g., medication adherence, co-morbidities like HIV), health system factors (e.g., delays in diagnosis, access to healthcare), and environmental factors. However, most previous studies. Few studies in Uganda have applied ML to routine EMR data to predict TB outcomes among HIV/TB co-infected patients, and majority have relied on conventional statistical models, which have limitations in handling complex, high-dimensional data and non-linear relationships among predictors [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eMachine learning (ML) offers a more robust, data-driven approach to identifying early predictors of treatment failure and MDR-TB progression by handling large, high-dimensional datasets that include patient characteristics, clinical parameters, identifying hidden patterns that conventional models might miss and improving prediction accuracy by considering both linear and non-linear relationships among predictors. Despite growing interest in machine learning for TB outcome prediction, few studies in Uganda have applied these methods to nationally representative EMR data, particularly among HIV/TB co-infected patients. This study addresses this gap by evaluating multiple ML models and identifying key predictors of treatment outcomes.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy design\u003c/h2\u003e \u003cp\u003eThe study utilised a retrospective cohort study design of people with HIV/TB co-infection receiving TB treatment across the country between 2020 and 2024.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy setting\u003c/h3\u003e\n\u003cp\u003e The study was conducted across 15 Regional Referral Hospitals (RRHs) in Uganda, which serve as key centers for TB diagnosis, treatment, and management, particularly for HIV/TB co-infected patients. These hospitals provide specialized services, including multidrug-resistant TB (MDR-TB) management, Directly Observed Therapy (DOTS), and integration with HIV care through antiretroviral therapy (ART) clinics. They also act as referral centers for lower-level health facilities, such as district hospitals and health center IVs, which initiate TB treatment before referring to complex cases to RRHs. The 15 RRHs include Arua RRH, Fort Portal RRH, Gulu RRH, Hoima RRH, Jinja RRH, Kabale RRH, Lira RRH, Mbale RRH, Mbarara RRH, Moroto RRH, Masaka RRH, Mubende RRH, Naguru RRH (China-Uganda Friendship Hospital), Soroti RRH, Kayunga RRH. TB management at these hospitals follows the Uganda National TB and Leprosy Program (NTLP) guidelines, ensuring standardized patient diagnosis, treatment monitoring, adherence support, and follow-up care.\u003c/p\u003e\n\u003ch3\u003eStudy population\u003c/h3\u003e\n\u003cp\u003e The study targeted TB patients diagnosed with HIV or already on HIV care in Regional Referral Hospitals in Uganda. The accessible population were all PLWH diagnosed with TB and started on anti-TB drug regimens in their respective clinics in the country for the period extending from 2020 to 2024. The sample population comprised HIV/TB co-infected patients whose data were randomly selected from the Uganda EMR system and used in this study.\u003c/p\u003e \u003cp\u003eRandom selection of patient records was employed to ensure that the data used were representative of the broader national population of HIV/TB co-infected patients. This approach minimized selection bias and enhanced the generalizability of the findings by ensuring that patients from different regions, demographic groups, and treatment centers had an equal chance of inclusion.\u003c/p\u003e\n\u003ch3\u003eData Description and data sources\u003c/h3\u003e\n\u003cp\u003eThe study utilized secondary, routinely collected, anonymized patient data from the Uganda Electronic Medical Records (EMR) system under the Ministry of Health (MOH). Clinical and demographic data: Retrieved from the Uganda EMR system, containing variables on patient demographics, TB disease classification, ART initiation, and health facility characteristics.\u003c/p\u003e\n\u003ch3\u003eStudy variables\u003c/h3\u003e\n\u003cp\u003eThe primary outcome of the study is TB treatment outcome, categorized as successful (including cure and treatment completion) or unsuccessful (comprising failure, default, death, and not evaluated). Predictor variables will be in categories: socio-demographic (age, sex, residence), patient-related (e.g., initiation of ART, TB disease classification), health system-related (e.g., distance to the facility, facility ownership).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eData preprocessing and exploratory data analysis (EDA)\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003eData cleaning and handling missing data\u003c/h2\u003e \u003cp\u003eMissing data were carefully assessed before deciding whether to exclude or impute values. The extent and pattern of missingness were analyzed using descriptive statistics and correlation matrices to determine whether the missing data were completely random (MCAR), missing at random (MAR), or missing not at random (MNAR). Variables with more than 30% missing values were excluded if they had limited predictive relevance, whereas those with lower levels of missingness underwent appropriate imputation based on their type and missing data pattern.\u003c/p\u003e \u003cp\u003eCategorical variables with missing values that were missing completely at random (MCAR) were imputed using mode imputation. For each categorical variable, the most frequently occurring category (mode) was identified from the observed data, and all missing entries within that variable were replaced with this modal category. This approach assumed that the missing values had the same distribution as the observed data and helped maintain the original variable structure without introducing bias. For numerical variables, mean or median imputation was applied if the data were normally distributed, whereas K-Nearest Neighbors (KNN) or Multiple Imputation by Chained Equations (MICE) were used for non-normal distributions or MNAR patterns. Sensitivity analyses were performed to compare different imputation methods and ensure minimal bias while maintaining predictive accuracy.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e\n\u003ch3\u003eData transformation and normality checks\u003c/h3\u003e\n\u003cp\u003eCategorical variables (e.g., treatment success/failure, sex, ART status) were encoded and Continuous variables (e.g., age, BMI) were checked for normality using histograms and the Shapiro-Wilk test. If non-normal, transformations (e.g., log transformation) were applied.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eOutlier detection and treatment\u003c/h2\u003e \u003cp\u003eOutliers were detected using box plots, Z-scores, and interquartile range (IQR) methods and Winsorization were considered for extreme outliers that significantly affect model performance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eCorrelation analysis and feature selection\u003c/h2\u003e \u003cp\u003eMulticollinearity was assessed using the variance inflation factor (VIF) with a threshold of 5 to identify redundant predictors. A correlation matrix and heatmaps were employed to examine dependencies among variables. Feature selection was performed using recursive feature elimination (RFE) and feature importance scores derived from tree-based models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eSample size determination\u003c/h2\u003e \u003cp\u003eThe sample size was determined based on the proportion of unsuccessful TB treatment cases (30%) and on a power analysis to ensure at least 80% statistical power. The final analysis was conducted using data from 5,062 participants, which provided sufficient power for the planned statistical and machine learning analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eMachine Learning Models\u003c/h2\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003eModel selection criteria\u003c/h2\u003e \u003cp\u003eThe study evaluated the performance of seven supervised algorithms, logistic regression, random forest, support vector machine and AdaBoost algorithms. Logistic regression offered a straightforward linear approach, while random forest, support vector machine and AdaBoost provide more flexible solutions for complex data. Through this analysis, we aimed at identifying the most effective algorithm for predicting TB treatment outcomes. See Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMachine Learning Algorithms and the rationale for their selection in the model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eReason for Selection\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBaseline model for comparison, interpretable results\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHandles non-linear relationships, robust to outliers\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSupport Vector Machine (SVM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEffective for complex, high-dimensional datasets\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaBoost (Adaptive Boosting)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eEnhances weak classifiers, improves prediction accuracy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eModel building\u003c/h2\u003e \u003cp\u003eThe model building methodology involved thorough data preparation, a 70/30 random split into training and test sets, and balanced sampling to ensure equal representation of treatment outcomes. A classifier algorithm was then trained using the training data to establish an optimal predictive model, which was evaluated using the independent test dataset. Python, a freely available programming language, along with TensorFlow libraries, an open-source machine learning framework were used.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eHyperparameter optimization\u003c/h2\u003e \u003cp\u003eHyperparameter tuning was performed to optimize model performance. Random Forest modal was fine-tuned by adjusting the number of trees, max depth, and minimum samples per split to balance complexity and accuracy. SVM was optimized by selecting the best kernel type (linear or RBF) and regularization parameter (C). AdaBoost was refined by tuning the learning rate and number of estimators to enhance model stability. Logistic Regression was optimized by adjusting L1 and L2 regularization strength to prevent overfitting. The optimizations were conducted using Grid Search CV and Randomized Search CV for efficient parameter selection. See Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMachine learning algorithms and corresponding hyperparameters tuned during model optimisation\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHyperparameters to Tune\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNumber of trees, max depth, min samples split\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eKernel type (linear, RBF), regularization parameter (C)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLearning rate, number of estimators\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRegularization strength (L1, L2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eModel evaluation\u003c/h2\u003e \u003cp\u003eThe evaluation of the model's performance was based on accuracy (the proportion of correctly classified observations in the unseen test set), sensitivity (the proportion of known failure outcomes correctly identified by the algorithm), precision (the proportion of predicted failure outcomes that match known failures in the test set), and specificity (the proportion of known successful outcomes correctly identified). Additionally, the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was used to assess the overall predictive classification performance (The AUC represents the probability that the model will correctly distinguish between a randomly chosen positive case and a randomly chosen negative case). A range of 0.5 to 1.0 on the AUC scale indicates the predictive power of the model, where 0.5 suggests no predictive ability and 1.0 indicates perfect predictive performance. See Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel evaluation metrics, their definitions, and interpretations\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDefinition\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e(TP\u0026thinsp;+\u0026thinsp;TN) / (TP\u0026thinsp;+\u0026thinsp;TN\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMeasures overall correctness\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTP / (TP\u0026thinsp;+\u0026thinsp;FP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow many predicted positive cases were true positives\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRecall (Sensitivity)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTP / (TP\u0026thinsp;+\u0026thinsp;FN)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHow many actual positive cases were correctly identified\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTN / (TN\u0026thinsp;+\u0026thinsp;FP)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCorrect identification of negative cases\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 \u0026times; (Precision \u0026times; Recall) / (Precision\u0026thinsp;+\u0026thinsp;Recall)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBalance between precision and recall\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAUC-ROC Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArea under ROC curve\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDiscriminative ability of the model\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eBaseline characteristics of study participants\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the baseline characteristics of patients included in the analysis. We included 5,062 tuberculosis (TB) patients with a mean age of 37.2 years (SD 14.5). Among these, 3,850 (76.1%) achieved treatment success, while 1,212 (23.9%) experienced treatment failure. Men comprised 54.8% of the study population and showed a higher proportion of treatment failure than women (56.4% vs 43.6%). Approximately one-third (34.2%) of the participants were married.\u003c/p\u003e \u003cp\u003eMost patients were classified as having pulmonary bacteriologically confirmed TB (61.1%), followed by pulmonary clinically diagnosed TB (33.7%) and extrapulmonary TB (5.2%). The majority were new TB cases (92.9%), while 5.3% were relapse cases and 1.5% had returned to treatment after loss to follow-up.\u003c/p\u003e \u003cp\u003eThe non-digital community-based directly observed treatment (DOT) model was the predominant treatment-support approach reported by 68.8%, followed by facility-based DOT (16.9%) and digital community DOT (14.3%). Treatment failure was most frequent among patients managed under the facility-based DOT model (20.1%).\u003c/p\u003e \u003cp\u003eRegarding ART status, 49.8% of participants were already on ART at baseline, 22.7% were newly initiated during TB treatment, and 27.5% had unknown ART status. Treatment outcomes did not vary substantially across ART categories. See Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline characteristics of patients\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eTB Treatment outcome\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTreatment success\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTreatment failure\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;5062(%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;3850(%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003en\u0026thinsp;=\u0026thinsp;1212(%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2289 (45.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1760 (45.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e529 (43.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2773 (54.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2090 (54.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e683 (56.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarital status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot married\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3332 (65.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2491 (64.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e841 (69.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMarried\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1730 (34.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1359 (35.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e371 (30.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge in years mean(sd)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e37.2 (14.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e37.2 (14.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e37.2 (14.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTB Disease Classification (Baseline)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExtra pulmonary (EP) TB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e262 (5.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e160 (4.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e102 (8.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePulmonary Clinically Diagnosed (P-CD) TB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1708 (33.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1288 (33.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e420 (34.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePulmonary bacteriologically confirmed (P-BC)TB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3092 (61.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2402 (62.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e690 (56.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eType of patient\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNew\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4704 (92.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3581 (93.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1123 (92.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRelapse\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e266 (5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e205 (5.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e61 (5.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReturn after lost to follow up\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e78 (1.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e54 (1.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e24 (2.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTreatment after failure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11 (0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7 (0.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4 (0.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTreatment history Unknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3 (0.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3 (0.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTB Treatment Model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDigital Community DOT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e726 (14.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e564 (14.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e162 (13.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFacility DOT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e855 (16.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e611 (15.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e244 (20.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-Digital Community DOT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3481 (68.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2675 (69.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e806 (66.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDS-TB regimen\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2RHZE/ 10 RH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e378 (7.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e266 (6.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e112 (9.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2RHZE/ 4RH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4684 (92.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3584 (93.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1100 (90.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eART Status\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlready on ART\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2521 (49.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1908 (49.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e613 (50.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNewly initiated on Art\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1149 (22.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e878 (22.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e271 (22.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1392 (27.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1064 (27.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e328 (27.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eModel performance\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the comparative performance of machine learning models developed to predict TB treatment outcomes among HIV/TB co-infected patients. Model performance varied across discrimination, recall, and accuracy metrics.\u003c/p\u003e \u003cp\u003eThe stacked ensemble model demonstrated the strongest overall performance, with a recall for treatment failure of 0.60, ROC-AUC of 0.669, and overall accuracy of 0.618. Logistic regression models with class balancing (L1/Lasso and L2/Ridge) achieved similar performance, detecting 58\u0026ndash;59% of failures, with ROC-AUC values around 0.65 and accuracy between 0.61 and 0.62.\u003c/p\u003e \u003cp\u003eThe tuned Random Forest model yielded the highest ROC-AUC (0.675) among individual algorithms, maintaining balanced recall for treatment failures (0.58) and successes (0.67). The tuned Gradient Boosting Machine (GBM) achieved the highest accuracy (0.768) and near-perfect recall for successes (0.99), but its sensitivity to failures was low (recall\u0026thinsp;=\u0026thinsp;0.07). Baseline Random Forest and Gradient Boosting models with SMOTE achieved accuracies of 0.709 and 0.715, respectively, but recalled\u0026thinsp;\u0026le;\u0026thinsp;0.32 of failures.\u003c/p\u003e \u003cp\u003eThe Support Vector Machine (SVM) model exhibited the weakest discrimination (ROC-AUC\u0026thinsp;=\u0026thinsp;0.466). AdaBoost with SMOTE achieved moderate accuracy (0.672) and recall for successes (0.77) but detected only 35% of failures. See Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eModel calibration\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the calibration and precision-recall performance of the evaluated models. Across models, ROC-AUC values during cross-validation ranged from 0.637 to 0.655, reflecting fair discrimination. Precision-recall AUC (PR-AUC) values were consistently high (\u0026ge;\u0026thinsp;0.836), indicating that model predictions of treatment failure were generally reliable when made.\u003c/p\u003e \u003cp\u003eThe tuned Random Forest achieved the highest PR-AUC (0.848), although calibration plots indicated slight overconfidence. Logistic regression models and the stacked ensemble exhibited moderate calibration with minor deviation from the ideal line at higher predicted probabilities. The tuned GBM model achieved the best calibration performance (Brier score\u0026thinsp;=\u0026thinsp;0.172), although this improvement did not enhance failure detection. See Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eFeature importance analysis\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents the relative importance of predictors of TB treatment outcomes derived from the Random Forest model. Age was the most influential predictor, accounting for nearly half of the model\u0026rsquo;s predictive strength. Other key demographic and clinical predictors included sex (male), marital status (married), ART status (unknown or newly initiated), TB classification (pulmonary bacteriologically confirmed and clinically diagnosed), and treatment model (community vs facility DOT). Programmatic variables, such as hospital care and treatment regimen (2RHZE/4RH), also contributed moderately to model predictions. Logistic regression analysis further identified facility-specific and clinical predictors associated with treatment success. Facilities including Mbale (OR\u0026thinsp;=\u0026thinsp;2.82), Naguru (OR\u0026thinsp;=\u0026thinsp;2.47), Hoima (OR\u0026thinsp;=\u0026thinsp;2.25), Gulu (OR\u0026thinsp;=\u0026thinsp;2.13), and Lira (OR\u0026thinsp;=\u0026thinsp;1.98) Regional Referral Hospitals exhibited higher odds of treatment success. In contrast, Soroti (OR\u0026thinsp;=\u0026thinsp;0.49), Moroto (OR\u0026thinsp;=\u0026thinsp;0.49), Masaka (OR\u0026thinsp;=\u0026thinsp;0.57), and St. Kizito Matany (OR\u0026thinsp;=\u0026thinsp;0.63) hospitals had lower odds. Patients with pulmonary bacteriologically confirmed TB (OR\u0026thinsp;=\u0026thinsp;2.15) and clinically diagnosed TB (OR\u0026thinsp;=\u0026thinsp;1.64) were more likely to achieve treatment success, whereas those classified as \u0026ldquo;treatment after failure\u0026rdquo; (OR\u0026thinsp;=\u0026thinsp;0.51) or managed under facility-based DOT (OR\u0026thinsp;=\u0026thinsp;0.68) had lower odds. See Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eComparison of feature importance across models\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e compares normalised feature importance across Random Forest, GBM, and Logistic Regression models. Age consistently emerged as the most influential predictor across all models. Random Forest and GBM models additionally emphasised sex (male), ART status (unknown or newly initiated), marital status (married), X-ray result (unknown), and treatment model (non-digital community DOT).\u003c/p\u003e \u003cp\u003eThe Logistic Regression model identified male sex (OR\u0026thinsp;=\u0026thinsp;0.33), unknown ART status (OR\u0026thinsp;=\u0026thinsp;0.40), and newly initiated ART (OR\u0026thinsp;=\u0026thinsp;0.37) as factors associated with lower odds of treatment success. Patients under the non-digital community DOT model (OR\u0026thinsp;=\u0026thinsp;0.43) also exhibited poorer outcomes. See Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003ePartial dependence analysis\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e illustrates the partial dependence of age on the predicted probability of TB treatment success. The relationship was non-linear, with younger patients (\u0026lt;\u0026thinsp;20 years) demonstrating the highest predicted probabilities of treatment success. The effect plateaued during early adulthood and declined progressively beyond 40 years, reaching the lowest probabilities among patients aged 70 years or older. See Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Discussions","content":"\u003cp\u003eThe study evaluated the predictive capacity of several machine learning algorithms, including logistic regression, Random Forest, Gradient Boosting Machine (GBM), Support Vector Machine (SVM), AdaBoost, and a stacked ensemble modelling forecasting tuberculosis (TB) treatment outcomes among HIV/TB co-infected patients. Model performance was evaluated across key metrics of discrimination, recall, accuracy, and calibration to determine both their predictive power and clinical applicability. Furthermore, feature importance analysis was used to identify the most influential predictors of treatment outcomes, while partial dependence analysis illustrated the marginal effects of specific predictors, particularly age, on treatment success probabilities.\u003c/p\u003e \u003cp\u003eThese findings indicate that the stacked ensemble and class-balanced logistic regression models achieved the most favourable balance between discrimination, calibration, and recall for treatment failures, supporting their application in programmatic contexts where accurate identification of patients at risk of poor outcomes is critical. The Random Forest model achieved the highest ROC-AUC, indicating strong discriminative ability, whereas the GBM attained the highest overall accuracy but was less sensitive to treatment failures. Across all models, key predictors of TB treatment outcome included age, ART status, sex, marital status, TB classification, and treatment modality. The partial dependence analysis further revealed a clear decline in treatment success with increasing age, particularly beyond 40 years, highlighting the clinical relevance of demographic and programmatic factors in predicting TB treatment outcomes.\u003c/p\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003eModel performance and predictive utility\u003c/h2\u003e \u003cp\u003e The study revealed that the stacked ensemble and class-balanced logistic regression models provided the most balanced and clinically useful predictions of TB treatment outcomes. These models were able to handle the imbalance between treatment success and failure effectively, achieving moderate discrimination and calibration. Their performance reflects the ability of ensemble approaches to combine complementary strengths of multiple algorithms, while regularized logistic models remain robust in managing collinearity and mixed data types. Comparable studies have reported similar findings, where ensemble and logistic regression models achieved an optimal balance between interpretability and accuracy in TB outcome prediction [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. This consistency supports the application of these models in real-world TB programs, where interpretability and operational feasibility are as crucial as accuracy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003eDiscrimination of individual models\u003c/h2\u003e \u003cp\u003eThe tuned Random Forest exhibited the highest ROC-AUC among single models, while the tuned GBM yielded the highest overall accuracy but failed to adequately detect treatment failures. The Random Forest\u0026rsquo;s superior discrimination may be attributed to its ability to capture complex, non-linear interactions among heterogeneous predictors. In contrast, the GBM\u0026rsquo;s overemphasis on optimising overall accuracy led to overfitting to the dominant outcome class, treatment success, resulting in poor sensitivity to failures. This trade-off between accuracy and recall has been consistently documented in previous studies [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], highlighting the need for careful tuning of machine learning models when dealing with imbalanced clinical datasets.\u003c/p\u003e \u003cp\u003eWe found that age was the most influential predictor of TB treatment outcomes. The partial dependence analysis indicated a non-linear relationship between age and the probability of treatment success, with younger patients achieving higher success rates than older adults, particularly those aged 40 years and above, showing declining probabilities. This trend is biologically plausible, as older patients often have weakened immunity, higher prevalence of comorbidities, and greater treatment fatigue, which collectively undermine adherence and clinical response. Studies conducted in Uganda and elsewhere corroborate this finding, demonstrating that older age is associated with increased risk of poor TB outcomes[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. These observations emphasize the need for age-sensitive treatment support and follow-up strategies.\u003c/p\u003e \u003cp\u003eWe also found that the ART status of these patients significantly influenced treatment outcomes. Patients who were newly initiated on ART or had unknown ART status were more likely to experience treatment failure compared to those already established on ART at baseline. This finding likely reflects late initiation of HIV care, poor linkage between TB and ART services, or incomplete treatment documentation. Unknown ART status also signals potential programmatic weaknesses in data recording or continuity of care. Similar associations have been reported in Ethiopia and Uganda, where delayed or interrupted ART initiation was a major contributor to unfavourable TB outcomes [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Strengthening ART\u0026ndash;TB integration and ensuring early initiation of ART remain key programmatic priorities.\u003c/p\u003e \u003cp\u003eThe study also found that male patients were more likely to experience treatment failure compared to female patients. This difference could be attributed to gendered disparities in health-seeking behaviour, adherence, and social support. Men are often diagnosed later, face occupational mobility challenges, and may have weaker support systems during treatment. Consistent with this, [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] and [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] observed that male patients had lower treatment success and higher default rates in Uganda. These findings underscore the importance of incorporating gender-sensitive approaches into TB programs, including flexible clinic schedules and targeted counselling for men.\u003c/p\u003e \u003cp\u003eIn addition, marital status was also associated with treatment outcomes, with unmarried individuals experiencing higher failure rates. Married participants likely benefit from stronger social and emotional support, enhancing adherence and clinic attendance. Similar associations between marital status and TB treatment success have been reported in sub-Saharan Africa [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], emphasizing the role of social support structures in sustaining adherence.\u003c/p\u003e \u003cp\u003ePatients with pulmonary bacteriologically confirmed or clinically diagnosed TB achieved better outcomes compared with those categorized under retreatment or treatment-after-failure groups. This likely reflects the benefits of early detection and diagnostic certainty, which facilitate appropriate treatment initiation and monitoring. In contrast, retreatment and treatment-after-failure categories typically include patients with prior drug resistance, adherence difficulties, or advanced disease. Comparable findings were reported by [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] and [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e], who observed that diagnostic category and treatment history were strong predictors of TB treatment success in African settings. These results reinforce the importance of early diagnosis and differentiated care models for high-risk or previously treated patients.\u003c/p\u003e \u003cp\u003eThe facility-based DOT model was associated with higher rates of treatment failure compared with community-based DOT models. This could be due to limited patient follow-up in busy facility settings and the convenience and social support offered by community-based models. Community DOT allows treatment supervision closer to patients\u0026rsquo; homes, reducing transportation barriers and improving adherence. Similar findings were observed in systematic reviews by [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] and Ugandan programmatic studies by [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], which showed superior treatment success under community-based approaches. However, [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]reported reduced success rates among previously treated patients managed under digital DOT, suggesting that the effectiveness of DOT interventions depends on context, implementation fidelity, and patient characteristics. Our results highlight the need to refine digital and facility-based DOT strategies to ensure optimal adherence support and supervision.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec29\" class=\"Section2\"\u003e \u003ch2\u003eModel calibration and reliability\u003c/h2\u003e \u003cp\u003eModel calibration and precision, recall analyses demonstrated fair discrimination and high precision for predicted failures, suggesting that when the models predicted a failure, they did so with high reliability. However, some models, particularly GBM, showed reduced recall, indicating difficulty in identifying all failure cases. This challenge arises from the inherent class imbalance in programmatic datasets, where treatment success overwhelmingly dominates. Recent studies emphasize that in clinical prediction, calibration metrics such as the Brier score and PR-AUC are as important as ROC-AUC for assessing real-world utility [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The fair calibration observed in this study suggests that the models provide reasonably accurate probabilities that could be integrated into decision-support systems to identify high-risk patients.\u003c/p\u003e \u003c/div\u003e"},{"header":"Strengths and Limitations","content":"\u003cp\u003eThis study benefited from a large, nationally representative large dataset that enhanced the reliability and generalizability of model findings. The application of multiple machine learning algorithms allowed for robust performance comparisons, while inclusion of calibration and interpretability analyses strengthened clinical relevance. Feature importance and partial dependence analyses further added explanatory value by identifying key drivers of treatment outcomes.\u003c/p\u003e \u003cp\u003eNevertheless, several limitations should be acknowledged. The retrospective design limited control over unmeasured confounders such as nutritional status, adherence behaviours, and socioeconomic conditions. Missing data, especially for ART status, may have introduced bias in model training. The imbalance between treatment success and failure restricted the models\u0026rsquo; sensitivity to failures, even after applying balancing techniques. Additionally, lack of external validation constrains the generalizability of findings to other settings.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study demonstrates that machine learning models, particularly ensemble and regularised logistic regression approaches, can moderately predict TB treatment outcomes among HIV/TB co-infected patients in Uganda. The models achieved fair discrimination and calibration, highlighting their potential utility in programmatic risk assessment, although their performance remains insufficient for autonomous clinical decision-making.\u003c/p\u003e \u003cp\u003eKey predictors of treatment success identified across models included age, ART status, sex, marital status, TB classification, and treatment model. Among these, age and ART status were the most influential variables, reflecting the complex interplay between biological, clinical, and behavioural factors in determining treatment outcomes. The findings also revealed that older adults, males, and patients newly initiated or with unknown ART status were at higher risk of treatment failure. These insights highlighted the importance of integrating data-driven predictive models into routine TB and HIV program management to enable early identification and targeted support for patients at risk of poor outcomes. Overall, this study adds to the existing body of evidence on the application of machine learning in TB treatment monitoring and offers a structured framework for the development of predictive tools designed to strengthen existing clinical and surveillance systems, particularly in resource-limited settings.\u003c/p\u003e \u003cdiv id=\"Sec32\" class=\"Section2\"\u003e \u003ch2\u003eRecommendations\u003c/h2\u003e \u003cp\u003eBased on the study findings, several recommendations are proposed to advance both research and practice. For researchers, future investigations should prioritize prospective and external validation of the best-performing models using independent datasets from diverse geographical regions in Uganda. Such validation will enhance the generalizability and robustness of predictive models, allowing them to be effectively adapted to various programmatic contexts. Researchers should also consider including additional clinical, behavioural, and socioeconomic variables, such as HIV viral load, adherence patterns, nutritional indicators, and income levels, in future model development to improve predictive accuracy and interpretability.\u003c/p\u003e \u003cp\u003eFor the Ministry of Health (MOH) and the National TB and Leprosy Program (NTLP), integration of predictive models into electronic medical record systems and digital health dashboards within national TB control programs should be prioritized. Embedding predictive analytics in these systems would enable real-time identification of high-risk patients, facilitate data-driven decision-making, and strengthen monitoring and evaluation of treatment outcomes.\u003c/p\u003e \u003cp\u003eFor health workers and program implementers, TB control programs should strengthen community-based and digital directly observed treatment (DOT) models to improve adherence and continuity of care. Tailored follow-up strategies, such as home-based visits and mobile health interventions, should be expanded to support patients who face structural or socioeconomic barriers to treatment completion.\u003c/p\u003e \u003cp\u003eFinally, for policymakers and program managers, targeted interventions should focus on vulnerable subgroups, particularly older adults, men, and patients newly initiated or with unknown ART status. Differentiated adherence counselling, personalised treatment support, and closer clinical monitoring for these groups can reduce treatment failure, enhance survival, and contribute to Uganda\u0026rsquo;s broader TB and HIV control goals.\u003c/p\u003e \u003c/div\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eART\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAntiretroviral Therapy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDOT\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDirectly Observed Therapy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTB\u0026ndash;Drug\u0026ndash;Susceptible Tuberculosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEDA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExploratory Data Analysis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEMR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eElectronic Medical Records\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExtra\u0026ndash;Pulmonary\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eF1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eScore\u0026ndash;Harmonic Mean of Precision and Recall\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eGBM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGradient Boosting Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eHIV\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHuman Immunodeficiency Virus\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eIQR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eInterquartile Range\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKNN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eK\u0026ndash;Nearest Neighbors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eL1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLasso Regularization\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eL2\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRidge Regularization\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMAR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMissing At Random\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMCAR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMissing Completely At Random\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMDR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTB\u0026ndash;Multidrug\u0026ndash;Resistant Tuberculosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMICE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMultiple Imputation by Chained Equations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eML\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMachine Learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMNAR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMissing Not At Random\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMOH\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMinistry of Health\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNTLP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNational TB and Leprosy Program\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePLWH\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePeople Living With HIV\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAUC\u0026ndash;Precision\u0026ndash;Recall Area Under the Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRBF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRadial Basis Function\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eREC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eResearch Ethics Committee\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRFE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRecursive Feature Elimination\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAUC\u0026ndash;Area Under the Receiver Operating Characteristic Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRRH\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRegional Referral Hospital\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eStandard Deviation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSDGs\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSustainable Development Goals\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSMOTE\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSynthetic Minority Oversampling Technique\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSVM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSupport Vector Machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTuberculosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTrue Negative\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTrue Positive\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTSR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTreatment Success Rate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eVIF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eVariance Inflation Factor\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eWHO\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eWorld Health Organization\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eXDR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTB\u0026ndash;Extensively Drug\u0026ndash;Resistant Tuberculosis\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval for this study was obtained from the Makerere University School of Public Health Research Ethics Committee, Approval No\u003cstrong\u003e. 542\u003c/strong\u003e. This study involved a secondary analysis of routinely collected, de-identified clinical data, and no direct interaction with human participants occurred. All study procedures were conducted in accordance with applicable ethical guidelines and regulations for research involving human participants. A waiver of informed consent was granted by the Research Ethics Committee due to the use of de-identified data. Data confidentiality and participant privacy were strictly maintained throughout all stages of the research process.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analysed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere was no funding for this study\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAO conceptualised and designed the study. JPA participated in data analysis, report writing, and drafting of the manuscript. AO and FO conducted data analysis. MN, EOE, EI, NC, and AK contributed to data acquisition, data management, and preliminary analyses. JPA, SRA, TN, RN, FO, and NRAO provided methodological and subject-matter expertise and critically reviewed and revised the manuscript for important intellectual content. RK supervised and reviewed the manuscript. NK provided overall supervision and reviewed the manuscript for scientific soundness. All authors read and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors gratefully acknowledge the Uganda Ministry of Health, Division of Health Information, for providing access to the data used in this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eWHO, Tuberculosis. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/news-room/fact-sheets/detail/tuberculosis\u003c/span\u003e\u003cspan address=\"https://www.who.int/news-room/fact-sheets/detail/tuberculosis\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization. The End Strategy TB. End TB Strateg. 2015;53:1689\u0026ndash;99.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeferi MY, El-Khatib Z, Boltena MT et al. Tuberculosis treatment outcome and predictors in africa: A systematic review and meta-analysis. Int J Environ Res Public Health; 18. Epub ahead of print 2021. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/ijerph182010678\u003c/span\u003e\u003cspan address=\"10.3390/ijerph182010678\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO. World Tuberculosis Day. 2025: Uniting to End TB in Uganda. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.afro.who.int/countries/uganda/news/world-tuberculosis-day-2025-uniting-end-tb-uganda\u003c/span\u003e\u003cspan address=\"https://www.afro.who.int/countries/uganda/news/world-tuberculosis-day-2025-uniting-end-tb-uganda\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMOH. \u003cem\u003eUganda National TB and Leprosy Program Republic of Uganda Ministry of Health\u003c/em\u003e. 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli SA, Mavundla TR, Fantu R et al. Outcomes of TB treatment in HIV co-infected TB patients in Ethiopia: a cross-sectional analytic study. BMC Infect Dis 2016; 1\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoH. Acceleration of HIV prevention in Uganda: A road map towards zero new infections by 2030.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO. \u003cem\u003eGlobal tuberculosis report, 2020\u003c/em\u003e. 2021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIzudi J, Tamwesigire IK, Bajunirwe F. Explaining the successes and failures of tuberculosis treatment programs; a tale of two regions in rural eastern Uganda. 2019; 2: 1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGichuhi HW, Magumba M, Kumar M, et al. A machine learning approach to explore individual risk factors for tuberculosis treatment non-adherence in Mukono district. PLOS Glob Public Heal. 2023;3:1\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosu MC, Faye LM, Apalata T. Optimizing Drug-Resistant Tuberculosis Treatment Outcomes in a High HIV-Burden Setting: A Study of Sputum Conversion and Regimen Efficacy in Rural South Africa. \u003cem\u003ePathogens\u003c/em\u003e; 14. Epub ahead of print 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/pathogens14050441\u003c/span\u003e\u003cspan address=\"10.3390/pathogens14050441\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang K, Wang Z, Li Z et al. Oriented Object Detection in Optical Remote Sensing Images using Deep Learning: A Survey. Epub ahead of print 2023. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10462-025-11256-0\u003c/span\u003e\u003cspan address=\"10.1007/s10462-025-11256-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang L, Yu T, Zheng G, et al. Using machine learning to predict selenium content in crops: Implications for soil health and agricultural land utilization in longevity regions. Sci Total Environ. 2025;964:178520.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKirenga BJ, Ssengooba W, Muwonge C, et al. Tuberculosis risk factors among tuberculosis patients in Kampala, Uganda: Implications for tuberculosis control. BMC Public Health. 2015;15:1\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOmara G, Bwayo D, Mukunya D, et al. Tuberculosis treatment success rate and its predictors among TB HIV co-infected patients in East and North Eastern Uganda. Sci Rep. 2025;15:5532.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOmara G, Bwayo D, Mukunya D, et al. Tuberculosis treatment success rate and its predictors among TB HIV co-infected patients in East and North Eastern Uganda. Sci Rep. 2025;15:5532.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaluku JB, Mukasa D, Bongomin F, et al. Gender differences among patients with drug resistant tuberculosis and HIV co-infection in Uganda: a countrywide retrospective cohort study. BMC Infect Dis. 2021;21:1\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWright DM, Reid N, Montgomery WI, et al. Herd-level bovine tuberculosis risk factors: Assessing the role of low-level badger population disturbance. Sci Rep. 2015;5:1\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMakabayi-Mugabe R, Musaazi J, Zawedde-Muyanja S, et al. Community-based directly observed therapy is effective and results in better treatment outcomes for patients with multi-drug resistant tuberculosis in Uganda. BMC Health Serv Res. 2023;23:1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIzudi J, Okello G, Bajunirwe F. Low treatment success rate among previously treated persons with drug-susceptible pulmonary tuberculosis in Kampala, Uganda. J Clin Tuberc Other Mycobact Dis. 2023;32:100375.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"bmc-infectious-diseases","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"infd","sideBox":"Learn more about [BMC Infectious Diseases](http://bmcinfectdis.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/infd","title":"BMC Infectious Diseases","twitterHandle":"#bmcinfectdis","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Tuberculosis, HIV co-infection, Machine learning, Treatment outcomes","lastPublishedDoi":"10.21203/rs.3.rs-8870898/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8870898/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eTuberculosis (TB) remains a major public health challenge, particularly in settings with high HIV prevalence. In Uganda, TB/HIV co-infection contributes to suboptimal treatment success rates (TSR), still below the World Health Organisation (WHO) target of \u0026ge;\u0026thinsp;90%. Early identification of patients at risk of poor outcomes is essential to mitigate poor adherence, reduce the risk of multidrug-resistant TB (MDR-TB), and improve health outcomes, including reduced morbidity and mortality as well as transmission. This study applied machine learning (ML) techniques to predict TB treatment outcomes among HIV/TB co-infected patients in Uganda and to identify the most influential predictors of treatment failure using feature importance and partial dependence analyses.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eData from a retrospective cohort of 5,062 HIV/TB co-infected patients treated in Uganda between 2020 and 2024were analysed. Machine learning models, including logistic regression, Random Forest, Gradient Boosting Machine (GBM), Support Vector Machine (SVM), AdaBoost, and a stacked ensemble, were developed and evaluated using discrimination, calibration, recall, and accuracy metrics. Feature importance and partial dependence analyses were used to interpret model predictions.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe stacked ensemble and class-balanced logistic regression models achieved the best overall performance (AUC\u0026thinsp;\u0026asymp;\u0026thinsp;0.67; accuracy\u0026thinsp;\u0026asymp;\u0026thinsp;0.62). The Random Forest model exhibited the highest discrimination (ROC-AUC\u0026thinsp;=\u0026thinsp;0.675), while GBM achieved the highest accuracy (0.768) but low sensitivity to treatment failures. Key predictors of treatment success included age, ART status, sex, marital status, TB classification, and treatment model. Treatment success declined progressively with increasing age, particularly beyond 40 years.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe models demonstrated moderate predictive performance and identified key demographic and programmatic predictors of TB treatment outcomes. While not suitable for autonomous clinical decision-making, these models may support risk stratification and targeted patient follow-up.\u003c/p\u003e\u003ch2\u003eTrial registration number:\u003c/h2\u003e \u003cp\u003eNot Applicable\u003c/p\u003e","manuscriptTitle":"Prediction of TB treatment outcomes among HIV/TB coinfected patients in Uganda using routinely collected clinical data: a machine-learning approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-13 07:35:01","doi":"10.21203/rs.3.rs-8870898/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"181909285474034418409244161011669145461","date":"2026-05-03T01:04:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"310780098088123963195653178118587194644","date":"2026-03-12T12:37:33+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-06T15:45:23+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-23T07:44:16+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-23T01:36:32+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-23T01:36:20+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Infectious Diseases","date":"2026-02-13T10:49:55+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-infectious-diseases","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"infd","sideBox":"Learn more about [BMC Infectious Diseases](http://bmcinfectdis.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/infd","title":"BMC Infectious Diseases","twitterHandle":"#bmcinfectdis","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"93648728-5d74-4a88-8901-08dbcf8b6e7d","owner":[],"postedDate":"March 13th, 2026","published":true,"recentEditorialEvents":[{"type":"reviewerAgreed","content":"181909285474034418409244161011669145461","date":"2026-05-03T01:04:45+00:00","index":43,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-13T07:35:01+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-13 07:35:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8870898","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8870898","identity":"rs-8870898","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}