XGBoost · Explainable Risk Factors · AI-Powered Clinical Interpretation · Clinical Decision Support Tool

Early-Warning Stroke Risk
Prediction for Primary Health Care

An explainable AI clinical decision support tool for frontline PHC workers. Stratifies stroke risk among patients under management for hypertension, using blood pressure at diagnosis, demographic factors and comorbidity burden — to help prioritise scarce follow-up capacity before symptoms appear.

    Developed and internally validated on Nigerian clinical data  ·  Designed for LMIC PHC settings
  

⚠ This tool is designed for non-diagnostic screening only. All results must be reviewed by a qualified clinician before any clinical decision is made.

📋

Patient Assessment Form

Clinical & Demographic Inputs

Demographics

Age (years) *

Sex *

BMI (kg/m²) *

Years since HTN diagnosis *

Blood Pressure

New patient — first presentation today No prior BP on record. Today's reading will be used as baseline. SBP_Delta will be set to 0 automatically.

SBP at HTN diagnosis (mmHg) * First recorded BP when HTN was diagnosed

DBP at HTN diagnosis (mmHg) *

SBP at last/today's visit (mmHg) * Most recent BP reading taken today

DBP at last/today's visit (mmHg) *

Lifestyle

Current Smoker *

Clinical Safety Flags

Prior stroke or TIA

Patient has a documented history of stroke or transient ischaemic attack. Triggers the secondary-prevention pathway regardless of model score.

FAST symptoms present today

Face drooping · Arm weakness · Speech difficulty · Time to act. If any are present, refer immediately – do not wait for the risk score. Triggers emergency referral and bypasses risk scoring entirely.

Comorbidities

Diabetes *

Chronic Kidney Disease (CKD) *

Heart Failure *

Depression *

Sleep Apnea *

🌐 Interpretation Language GPT-4o generates interpretation in selected language. Medical terms remain in English.

📊

Risk Assessment Results

XGBoost · Calibrated Probability · Risk Factor Attribution

🏥

Ready to Assess

Complete the patient form on the left and click
Assess Stroke Risk to generate results.

Running XGBoost model...

—

STROKE RISK SCORE

            0.01%1% (LOG SCALE)100%
          

—

Recommended Action

—

SBP Change

—

Pulse Pressure

—

Comorbidity Score

Top Contributing Factors For This Patient

AI Clinical Interpretation

Generating guideline-grounded interpretation...

AHA/ASA · WHO PEN · Nigeria PHC Protocol

0.776

AUC-ROC (out-of-fold)

97.9×

Imbalance ratio

85%

Sensitivity

N=10K

Training size

Model score–based risk band. Not a clinical diagnosis.
Threshold: —% (sex-specific) | Sensitivity: 85.3% | Specificity: 44.5%

🔒

Model Monitoring Dashboard

Administrator access required. Enter the admin password to view data drift metrics, prediction logs, and model performance monitoring.

Incorrect password. Access denied.

📊 Model Monitoring Dashboard

          StrokeRisk AI · PHC Clinical Decision Support Tool · Admin View
        

—

Total Predictions

This session

—

High Risk Flags

—

New Patient Assessments

SBP Delta = 0

—

Feature Drift Status

—

      Feature Distribution Drift — Session vs Training Reference
    

Feature	Training Mean	Session Mean	Δ Deviation	Drift Status	Inputs (N)

      Recent Prediction Log (Session)
    

#	Time	Age	SBP Last	SBP Delta	Patient Type	Risk Score	Risk Band

⚠ This dashboard monitors session-level data only. No patient identifiers are stored. All inputs are anonymous aggregates for model performance monitoring purposes. Drift tests: continuous — Kolmogorov–Smirnov; binary — rate comparison with Wilson interval; categorical — chi-squared goodness-of-fit. WARNING at p<0.05 or 5–10% rate difference; ALERT at p<0.01 or >10%. An ALERT triggers threshold re-derivation against locally observed outcomes.

Early-Warning Stroke RiskPrediction for Primary Health Care

Model Monitoring Dashboard

📊 Model Monitoring Dashboard

Early-Warning Stroke Risk
Prediction for Primary Health Care