LIVE DEMO
Hospital Workforce Intelligence

Predict staffing crises 3 weeks before they happen

MindMedix AI analyses shift rosters and absence patterns to give hospital operations managers an early warning — with enough time to act on it, not react to it.

Try the Model → Book a Call →
Current Week — All Departments
0%
Overall Accuracy
LOWO cross-validation · 10 folds
0%
Crisis Recall
28 / 29 crisis events caught
0
Real Hospital Wards
Italian NHS data · Ceschia 2023
3 Weeks
Advance Warning
Before shortage becomes critical
Hospital Intelligence Board

All Departments — Current Week

Staffing risk across all 10 wards. Color-coded by severity. Sparklines show 6-week trend.

2 CRITICAL ALERTS — Cardiologia and Terapia Intensiva require immediate action. Predicted 3 weeks in advance. Estimated combined saving: €5,900 by acting now vs agency staff.
Department / Reparto Risk Level Probability 6-Week Trend & Key Signal Time to Act Action
2 Departments
Require immediate action
4 Departments
Require monitoring
4 Departments
Operating normally
Interactive Model

Try the Prediction Engine

Adjust the staffing signals below. The model predicts risk in real time — exactly as it runs in production.

← drag sliders to predict →
% of nurse-days absent. AT_RISK >18% · CRISIS >25%
% of shifts beyond 7.2h/day. Triggers AT_RISK only when rising.
Positive = overtime is rising. Required with high overtime for AT_RISK.
Shiftoff requests per nurse assignment. AT_RISK >0.15 · CRISIS >0.25
Positive = absence rate is worsening week-over-week.
—
—
Move the sliders to predict
Probability Distribution
CRISIS
0%
AT RISK
0%
STABLE
0%
Top Driving Signals
Staffing Action Plan
① BEST
Internal Reallocation
Nurses from stable wards · no extra budget needed
1.0×
② IF NEEDED
Planned Overtime
Remaining hours · advance-scheduled
1.25×
③ LAST RESORT
Agency Staff (gettonisti)
Only if no action taken in time
1.9×
Saving vs. waiting for agency
3-week advance warning makes options ① + ② possible
Model Intelligence

What the Model Learned

Feature importance by gain — which signals most powerfully separate risk levels. 22 features total, top 10 shown.

■ Absence signals dominate (54% combined importance) ■ overtime_rising — trend-gated feature · rank #3 · fixes baseline bias
Model Evidence

Validation Results

Leave-One-Ward-Out CV — model never trains on the ward it predicts. Simulates deployment to a new hospital.

Per-Class Performance (LOWO CV · 170 samples)
STABLE
100%
Recall · 91 samples · 0 false negatives
AT RISK
92%
Recall · 50 samples
CRISIS
97%
Recall · 29 samples · 28/29 caught
97.1%
Overall Accuracy
0.97
Macro F1 Score
Methodology & Compliance
Data source34 real Italian hospital wards (Ceschia et al., 2023)
ModelXGBoost multiclass classifier (3 risk levels)
ValidationLeave-One-Ward-Out CV — 10 folds, 170 samples
Temporal CVTrain weeks 1–4, predict week 5 → 97.1%
PrivacyZero personal data — anonymous IDs only
ComplianceGDPR Art.89 · EU AI Act ready
Not a deviceOperational AI only — NOT a medical device
MindMedix AI never processes patient data or personal nurse identifiers.
Platform Pipeline

How It Works

Four automated steps from raw scheduling data to actionable staffing decision.

01
Data Ingestion
Anonymised shift rosters and absence records from your HR system. No patient data ever touched.
Zucchetti · TeamSystem · HL7
02
Signal Extraction
22 weekly features per ward: absence rates, overtime trends, fatigue indices, shift refusals — rolling 4-week windows.
22 Features · Rolling 4-week
03
Risk Prediction
XGBoost classifier outputs STABLE / AT_RISK / CRISIS with confidence scores and top driving signals per ward.
97.1% Accuracy · 97% Crisis Recall
04
Action Report
Weekly dashboard: which wards to act on, available staff IDs, and the cost differential of acting now vs waiting.
3-week advance warning
Cost if you act now
1.25×
Planned overtime — €18.5/h × 1.25
Cost if you wait
1.90×
Agency (gettonisti) — €35/h gross
Avg saving per crisis prevented
€2,400
Per ward event · scales with hospital size
Ready to run a pilot?

48-hour setup. Zero risk. Immediate insight.

Connect your existing HR scheduling data and receive your first staffing risk report within 48 hours. No infrastructure changes. No personal data required.

Schedule a Pilot → Download Sample Report
✓ GDPR compliant ✓ EU AI Act ready ✓ Not a medical device ✓ No patient data