This clinical decision support tool utilizes historical Medical Expenditure Panel Survey (MEPS) data to predict the required Medicaid subsidy allocation for cancer patients. The model identifies the intersection between financial burden, depressive symptoms, and treatment non-adherence to provide practitioners with a quantitative risk profile on after diagnosis and ongoing treatment.
Financial toxicity is a documented side effect of cancer treatment that directly correlates with treatment abandonment. This model provides three key clinical interventions:
- Early Navigation: Predicts statutory Medicaid eligibility and allocation before a patient incurs catastrophic debt.
- Adherence Risk Profiling: Uses SDoH (Social Determinants of Health) to flag patients at high risk of quitting treatment.
- Data-Driven Advocacy: Aggregates predicted subsidy requirements to justify hospital grant applications and community resource funding for at risk patients.
- Modeling: Random Forest Regressor / HistGradientBoosting
- Explainability: SHAP (SHapley Additive exPlanations) for local feature impact
- Interface: Streamlit (Enterprise Clinical UI)
- Data Source: AHRQ Medical Expenditure Panel Survey (MEPS)
App.py: The main Streamlit web application.Clinical Insurance Aid Tool.py: The full codemeps_model_data.pkl: The fully trained machine learning "brain" (exported from the training phase).requirements.txt: Environment dependencies..streamlit/config.toml: Custom clinical theme and color palette settings.
The application is deployed via Streamlit Community Cloud and can be accessed at: https://team8iscool.streamlit.app/
Developed by Safwan, Mehraf, Imanuel University College London*