Data-Kitchen-Ethiopia/TB_Adherence_prediction_tool

This is a data driven TB treatment adherance prediction model with user interface and backend # TB_Adherence_prediction_tool This is a data driven TB treatment adherance prediction model with user interface and backend ## Problem Statement Clinical & Public Health ContextTuberculosis (TB) requires a minimum of 6 months of continuous multi-drug treatment (typically Isoniazid, Rifampicin, Pyrazinamide, and Ethambutol). Globally, between 10% and 30% of patients fail to complete their treatment regimen due to treatment fatigue, severe adverse drug reactions, socio-economic barriers, or lack of family support. When a patient stops taking anti-TB medication or misses doses intermittently: ### Drug Resistance Escalation: It transforms drug-susceptible TB into Multi-Drug Resistant TB (MDR-TB) or Extensively Drug-Resistant TB (XDR-TB), which requires up to 2 years of expensive, highly toxic second-line regimens. ### Community Transmission: Non-adherent patients remain infectious longer, spreading TB strains directly within households and dense urban areas ### System Strain: Reactive care for defaulted patients costs health systems up to $20\times$ more than preventive adherence management. ## The current model The Industry Challenge"Current TB control programs rely on one-size-fits-all Directly Observed Therapy (DOT) or reactive identification—detecting non-adherence only AFTER a patient has missed multiple clinic visits or defaulted. Healthcare providers lack an automated, data-driven system to stratify newly diagnosed patients into predictive risk tiers (Low, Medium, High) at baseline and dynamically route them to tailored, cost-effective digital and clinical intervention workflows before default occurs. ## Why Existing Solutions Fail? ### Universal DOT Is Resource-Intensive: Expecting every patient to visit a health facility daily for 6 months creates high transportation costs and lost wages, driving default. ### Unstructured Risk Scoring: Standard clinical checks ignore compound non-adherence predictors (e.g., distance to facility combined with food insecurity …