bashirssuna/Impaired-Fasting-Glucose-Risk-Model-

This is the model for a tool that estimates the probability of Impaired Fasting Glucose (IFG) for Uganda using selected demographic and clinical variables. You can use either the full model (with cholesterol) or a simplified model. # IFG Risk Stratification Model (Uganda) A machine-learning risk-stratification model for **Impaired Fasting Glucose (IFG)** in Ugandan adults, built on the nationally representative 2023 WHO STEPS survey. The final model is a LASSO logistic regression with isotonic probability calibration, evaluated with **out-of-fold (OOF) predictions** from nested cross-validation, and paired with a threshold-based low / moderate / high risk stratification. This repository reproduces the analysis behind the manuscript *"Calibrated Risk Stratification Model for Impaired Fasting Glucose in the Ugandan Population: A Nested Cross-Validation and Threshold Optimization Approach."* ## Key results (out-of-fold) | Metric | Value (95% CI) | |--------|----------------| | ROC AUC | 0.68 (0.64–0.71) | | PR AUC | 0.40 (0.34–0.46) | | Brier score | 0.155 (0.143–0.168) | | Sensitivity / Specificity | 0.60 / 0.69 | | PPV / NPV | 0.34 / 0.86 | At the balanced (G-mean) operating point, threshold ≈ 0.21. Risk bands: low ( **On evaluation:** all reported performance is from OOF predictions, not from evaluating > the model on its own training data. See docs/EVALUATION_NOTES.md. ## Repository structure ``` ifg-risk-model/ ├── preprocessing/ │ ├── steps2.0.1.R # R: cleaning, MICE imputation, IFG derivation │ └── README.md ├── src/ # model modules (data processing, model zoo, evaluation, plots) ├── pipeline/ # original entry scripts (train / evaluate / stratify / app) ├── analysis/ # reproducible OOF analysis (primary) │ ├── run_analysis.py # discovery + nested-CV OOF + bootstrap CIs + thresholds │ ├── make_figures.py # all analysis figures │ └── recompute_tables.py # descriptive Tables 3 & 4 ├── results/ # committed outputs (reports, metrics, OOF predictions, figures) ├── data/ │ └── raw/ # place case_cohort_IFG.xlsx here (not distributed; see data/README.md) ├── docs/EVALUAT …