Background: Drug-resistant tuberculosis remains a major challenge in resource-limited settings, particularly in rural regions of the Eastern Cape Province, where limited laboratory infrastructure, constrained access to advanced molecular diagnostics, shortages of specialized healthcare personnel, and prolonged diagnostic turnaround times can delay appropriate treatment initiation. This study examined whether routinely detectable genomic resistance markers could be integrated with parsimonious machine learning approaches to support early risk stratification for isoniazid (INH) and/or rifampicin (RIF) resistance and multidrug-resistant tuberculosis (MDR-TB). Methods: We conducted a retrospective analysis of clinical, demographic, and genomic data from 207 Mycobacterium tuberculosis isolates representing 207 unique patients. Resistance was classified as INH and/or RIF resistance or MDR-TB (concurrent resistance to both drugs). Predictors included age, sex, and canonical resistance-associated mutations (katG S315T, inhA −15C>T, and rpoB codon substitutions). Logistic regression was used to estimate adjusted odds ratios (aORs), while Random Forest models were applied to assess non-linear feature importance. Internal validation was performed using 10-fold cross-validation. A systems network analysis mapped the integration of model-derived risk bands into Clinical Governance structures and Community-Engaged Education pathways, including interventions delivered by Community Health Workers (CHWs). Results: INH and/or RIF resistance was identified in 58.9% of isolates, with 21.7% classified as MDR-TB. The most frequently detected mutations were katG S315T (29.0%) and rpoB S450L (26.6%). Logistic regression identified rpoB S450L (aOR 4.20; 95% CI: 2.10–8.45) and katG S315T (aOR 2.85; 95% CI: 1.40–5.80) as the strongest independent predictors, while age and sex were not statistically significant. Models demonstrated strong internal discrimination (AUCs of 0.96 for INH and/or RIF resistance and 0.99 for MDR-TB). Risk stratification categorized 18% of patients as high risk. Scenario-based modelling suggested that prioritizing high-risk patients for reflex Line Probe Assay testing could reduce the median time to appropriate treatment from 14 to 3 days and may reduce progression from isoniazid-resistant TB to MDR-TB under specified operational assumptions. Conclusions: Mutation-informed predictive modelling demonstrates strong internally validated discrimination and provides a structured framework for risk-stratified intervention. Integrating probability-based risk thresholds within Clinical Governance systems and community-level support structures, including CHW-led adherence and education strategies, may support earlier treatment optimization in high-burden rural settings. External validation and prospective implementation studies are required before broader programmatic adoption.

Background: Drug-resistant tuberculosis (DR-TB) remains a critical challenge in high-burden rural settings. This study aims to bridge genomic resistance markers with machine learning to create a predictive model for timely diagnosis and the prevention of multidrug-resistant tuberculosis (MDR-TB). Methods: We conducted a retrospective analysis of clinical, demographic, and genomic data from 207 Mycobacterium tuberculosis isolates (representing 207 unique patients). Resistance was categorized as any resistance or MDR-TB.” Predictors included age, sex, and key mutations (S315T, -15C>T, and substitutions). Logistic regression (LR) was used to calculate adjusted odds ratios (aORs), while Random Forest (RF) assessed non-linear feature importance. Model validity was confirmed via 10-fold cross-validation. A Systems Network Analysis mapped the integration of model outputs into Clinical Governance (SOPs/KPIs) and Community-Engaged Education. Results: Resistance to at least one drug was found in 58.9% of isolates, with 21.7% classified as MDR-TB. Predominant mutations included S315T (29.0%) and S450L (26.6%). LR identified S450L (aOR 4.20, 95% CI: 2.10–8.45) and S315T (aOR 2.85, 95% CI: 1.40–5.80) as the strongest predictors; demographic variables (age, sex) were not statistically significant. Models achieved high discriminative power (AUC: 0.96 for any resistance; 0.99 for MDR-TB). Risk stratification categorized 18% of patients as high risk. Simulations showed that prioritizing high-risk patients for reflex LPA testing could reduce the median time to appropriate treatment from 14 to 3 days, potentially preventing 12–15% of Isoniazid-resistant TB (Hr-TB) cases from progressing to MDR-TB. Conclusion: Integrating molecular markers with machine learning enables highly accurate, risk-stratified clinical interventions. By embedding these predictive models into governance frameworks and utilizing Community Health Workers (CHWs) for targeted education, rural health systems can significantly reduce diagnostic delays and interrupt the progression of MDR-TB.

Background: Early identification of individuals at heightened risk for drug-resistant tuberculosis (DR-TB) is essential for prioritising rapid diagnostic testing and targeted preventive care, particularly in resource-constrained rural settings. Although routine tuberculosis (TB) surveillance and household contact investigation records provide an accessible data source, their predictive capacity and utility for machine learning-based risk stratification remain under-evaluated. This study evaluated the predictive utility of routine demographic and household contact investigation data for programme-defined risk classification in rural Eastern Cape, South Africa, and identified critical data gaps relevant to the future development of clinically actionable DR-TB prediction models. Methods: A retrospective secondary analysis was conducted using anonymised, routinely collected household contact investigation data from 437 contacts of smear-positive pulmonary TB index cases in King Sabata Dalindyebo Municipality, Eastern Cape. The primary outcome was programme-assigned TB vulnerability classification (high versus low risk), rather than microbiologically confirmed TB or DR-TB. Bivariate analyses and multivariable logistic regression were used to assess factors associated with high-risk classification. Three supervised machine learning algorithms—logistic regression, random forest, and gradient boosting—were evaluated using an 80:20 stratified training–test split and five-fold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), area under the precision–recall curve (AUPRC), calibration measures, and decision curve analysis. A structured variable-gap assessment identified additional clinical, microbiological, and molecular predictors required for future DR-TB prediction. Results: Of 437 household contacts, 290 (66.4%) were classified as high risk and 147 (33.6%) as low risk. Age was significantly associated with programme-defined risk classification (χ² = 16.61, p = 0.005), whereas sex was not. Children aged 0–4 years had the highest proportion classified as high risk (35/41; 85.4%). In multivariable logistic regression, younger age (adjusted OR [aOR] = 2.41; 95% CI: 1.48–3.94) and larger household size (aOR = 1.12 per additional household member; 95% CI: 1.01–1.25) were independently associated with high-risk classification. Gradient boosting demonstrated the strongest cross-validated performance (AUROC = 0.657, SD = 0.052; AUPRC = 0.785, SD = 0.042), although overall discrimination remained modest. Household-level analyses demonstrated clustering of programme-defined vulnerability. The variable-gap assessment identified important unavailable predictors, including previous TB treatment, HIV status, diabetes, microbiological measures, molecular diagnostic results, and drug-susceptibility information. Conclusions: Routine household contact investigation data provide measurable but limited predictive information for programme-defined TB vulnerability, with age and household size emerging as important baseline factors. However, these data should not be interpreted as directly predicting microbiologically confirmed DR-TB. Development of robust and clinically meaningful DR-TB prediction models in rural African settings will require integration of routine programme data with clinical characteristics, molecular diagnostic results, and drug-susceptibility testing, followed by appropriate internal and external validation.

Tuberculosis (TB) treatment outcomes may differ among individuals with similar routinely recorded demographic and clinical characteristics, suggesting limitations in the explanatory scope of programme surveillance data. This study evaluated the capacity of routinely collected demographic, clinical, socioeconomic, and behavioural variables to explain TB treatment outcomes and used identified evidence gaps to define priorities for complementary molecular research. We conducted a retrospective secondary analysis of anonymised routine TB programme data from 422 adults treated in public healthcare facilities in rural OR Tambo District Municipality, Eastern Cape, South Africa, between January 2018 and December 2022. Treatment success was defined as cure or treatment completion, whereas unsuccessful outcomes included death, treatment failure, or loss to follow-up. The explanatory and predictive performance of logistic regression, Poisson regression, and Random Forest models was evaluated, alongside a structured evidence-gap analysis comparing variables available in routine surveillance with molecular determinants of treatment response and drug resistance identified in the literature. Previous TB treatment was associated with lower odds of treatment success in univariable analysis (OR 0.48, 95% CI 0.29–0.78; p=0.003), but the association attenuated after adjustment (OR 0.64, 95% CI 0.38–1.09; p=0.10). HIV status was not independently associated with treatment success in logistic (OR 0.80, 95% CI 0.49–1.29; p=0.36) or Poisson regression (aRR 0.95, 95% CI 0.66–1.37; p=0.79). Random Forest achieved a higher average precision than logistic regression (0.907 versus 0.810), demonstrating improved prediction from routinely available variables; however, molecular mechanisms could not be evaluated because genomic, transcriptomic, immunological, and pharmacological measures were absent from the dataset. These findings define an important boundary between programme-level prediction and mechanistic explanation. We propose a program-to-molecular translational framework in which routine surveillance identifies clinically relevant patterns and knowledge gaps that can subsequently be investigated using genomic, transcriptomic, and functional molecular approaches. This integration may strengthen mechanistic understanding of heterogeneous TB treatment responses and inform future precision-oriented TB research.