Datasets:

---

tarekmasryo

/

hospital-deterioration-dataset

Like 1

🏥 A fully simulated clinical time-series benchmark for early warning modeling.
Each admission includes hourly vitals + labs + patient context, with a main label for “deterioration in the next 12 hours”.

📦 Dataset snapshot

• Simulated / privacy-safe (no real patients)
• Hourly granularity per admission (hour_from_admission)
• Main target: deterioration_next_12h (0/1)
• Clean schema with no missing values (simulation-designed consistency)

🗂️ Files included

• patients.csv (static + admission context + summary outcomes)
• vitals_timeseries.csv (hourly vitals + oxygen support + nurse signals)
• labs_timeseries.csv (hourly labs + sepsis_risk_score)
• hospital_deterioration_hourly_panel.csv (joined full panel with labels)
• hospital_deterioration_ml_ready.csv (features + single target; recommended entry point)

🔎 Feature groups (high-level)

• Patient context: age, gender, comorbidity_index, admission_type, baseline_risk_score, los_hours
• Hourly vitals: heart_rate, respiratory_rate, spo2_pct, temperature_c, systolic/diastolic_bp
• Oxygen support: oxygen_device, oxygen_flow
• Care signals: mobility_score, nurse_alert
• Hourly labs: wbc_count, lactate, creatinine, crp_level, hemoglobin, sepsis_risk_score

🎯 Target definition (why it’s useful)

• deterioration_next_12h = 1 if an event happens after the current hour and within the next 12 hours
• This mirrors real early-warning systems: alert before deterioration, not at the same moment

✅ Good for
📊 EDA + monitoring-style dashboards (risk trajectories over time)
🤖 Baseline ML + sequence models (hourly tabular / time-series)
🧪 Calibration + threshold/policy design (trade-off: recall vs false alarms)
🧠 Teaching end-to-end clinical ML pipelines without real EHR data

⚠️ Modeling notes (to avoid leakage)

• Split by patient/admission (patient_id groups), not by rows
• Use time-aware validation when evaluating “next 12h” forecasting
• Treat this as a simulation benchmark — validate on real EHR + clinical oversight before any real deployment

🧾 Educational/research use only — not for clinical decisions.