🫀 AEGIS Clinical Decision Support P1 · static

This app is part of the AEGIS ("Artificial Intelligence Guided Engineering System") research programme. AEGIS's founding philosophy: AI augments a person's judgment, it does not replace it.

Production model: Random Forest, selected after independently benchmarking 8 models (Logistic Regression, Random Forest, CatBoost, SVM, XGBoost, Explainable Boosting Machine, LightGBM, Stacking Ensemble) — all of which converged on the same top-5 predictors: medication adherence, BNP, HCAS, age, and distance to hospital. The two runner-ups (LightGBM, Logistic Regression) are included for side-by-side comparison.

HCAS (Human Clinical Abnormality Score): a composite feature summarizing cardiovascular severity from routine vitals (BMI, BNP, Creatinine, Sodium, Systolic BP, Heart Rate, ACE Inhibitor use, Beta Blocker use). It's computed automatically from the inputs — enable "Show computed HCAS" to inspect it. Caveat: the original script used to compute HCAS for model training no longer exists among the surviving project files; the version here is a faithful reconstruction from the documented methodology using standard clinical thresholds, not a byte-exact replica.

Accuracy is not the goal of this build. This tool wraps the chosen model in something simple and testable. It has not been validated as a clinical decision-making device and should not be used as one.

Static edition — the model runs entirely in your browser; no server. Predictions are bit-exact reproductions of the original trained scikit-learn / LightGBM pipelines (verified to machine epsilon).