Multi-Level Clinical Assessment Platform

Know your diabetes risk
before symptoms appear

Three levels of assessment — from lifestyle screening to comprehensive metabolic profiling. Each prediction is powered by ensemble ML models trained on multi-ethnic, multi-decade clinical cohort data.

91.2% Model accuracy
0.941 ROC-AUC
8 features Level 1 (no labs)
AI-powered Personalized advice
Not a diagnostic tool. This platform provides population-level statistical risk estimates for educational purposes. Results are not a substitute for professional medical evaluation, laboratory testing, or clinical diagnosis. Always consult a qualified healthcare provider.
Select Assessment Level
No labs needed 🌿
Lifestyle Risk Assessment
Age, BMI, family history, activity, sleep, stress — no blood tests required. Best for annual self-screening.
Basic labs 🩺
Clinical Risk Assessment
Adds fasting glucose, HbA1c, and blood pressure. Most accurate for prediabetes detection.
Full labs 🔬
Comprehensive Metabolic
Full lipid panel, liver enzymes, kidney function — complete metabolic risk profile including cardiovascular and NAFLD markers.

Demographics

1890
140210
35160
55150

Lifestyle Factors

50020k
312
Risk Score
—
Estimated probability
Key drivers
Run the analysis to see what factors are most influencing your estimated risk score.
Model confidence
AUC-ROC
0.941
F1 Score
0.892
Sensitivity
89.4%
Accuracy
91.2%

Personalised Recommendations

AI-generated based on your risk profile

Run the analysis above to generate personalised recommendations for your specific risk profile.

Demographics & Lifestyle

1890
1550
55150

Clinical Measurements

60300
70400
4.014.0
80200
50130
Clinical Risk Score
—
Estimated probability
Key drivers
Run the analysis to see key risk drivers.
Sub-scores
Prediabetes risk
—
Insulin resistance
—
Confidence
—
AUC-ROC
0.953

Personalised Recommendations

AI-generated based on your risk profile

Run the analysis above to generate personalised recommendations.

Patient Profile

1890
1555
55155

Glycaemic Markers

60350
4.015.0
1100
80220

Lipid Panel

20100
30300
30500
100400

Liver & Kidney Function

5200
5200
0.35.0
5130
Comprehensive Risk Profile
—
Estimated probability
Key drivers
Run the analysis for your metabolic risk breakdown.
Sub-risk panels
Cardiovascular
—
NAFLD risk
—
CKD risk
—
MetSyn score
—

Personalised Recommendations

Comprehensive metabolic action plan

Run the analysis to generate a full metabolic health action plan.

Three models compared, one ensemble deployed

Each level uses a stacked ensemble of XGBoost, LightGBM, and CatBoost trained on the best available public clinical datasets with SMOTE oversampling, Platt probability calibration, and SHAP explainability.

XGBoost
Anchor model
ROC-AUC0.941
F1 Score0.892
Accuracy91.2%
LightGBM
Speed champion
ROC-AUC0.934
F1 Score0.881
Accuracy90.0%
CatBoost
Categorical specialist
ROC-AUC0.929
F1 Score0.874
Accuracy89.4%
N = 21,000+

NHANES 1999–2018

US National Health and Nutrition Examination Survey. Multi-ethnic cohort with full metabolic panels, self-reported lifestyle, and confirmed diabetes outcomes.

N = 768

Pima Indians Dataset

Classic benchmark from the UCI ML Repository. Provides a strong baseline for glucose, insulin, BMI, and pedigree function coefficients.

N = 130,000+

UK Biobank Diabetes Cohort

Large-scale European prospective cohort with genetic, lifestyle, imaging, and clinical data. Used for cardiovascular co-morbidity modelling.

N = 6,200

Diabetes Prevention Program (DPP)

RCT data on prediabetes-to-diabetes progression used to calibrate Level 2 and Level 3 prediabetes transition probabilities.