Hospital Deterioration Dataset — Data Dictionary
This dataset represents a high-fidelity simulated hospital cohort designed for early-warning and deterioration modeling.
The records are generated via a rules-based and probabilistic simulation calibrated to typical hospital patterns. They do not correspond to real patients and contain no identifiable information.
- Time is expressed as hours from admission.
- Each patient stay is capped at 72 hours.
- All identifiers (
patient_id) are artificial and non-identifiable. - All features are fully observed (no missing values).
1. patients.csv — Patient-level static data
Granularity: one row per patient (10,000 patients).
| column_name | type | description | allowed_values / range | missing_values |
|---|---|---|---|---|
patient_id |
int | Unique patient identifier | 1–10,000 | None |
age |
int | Age at admission (years) | 18–90 | None |
gender |
category | Biological sex | "M", "F" |
None |
comorbidity_index |
int | Aggregate comorbidity burden (higher = more chronic disease) | 0–8 | None |
admission_type |
category | Admission route to the hospital | "ED", "Elective", "Transfer" |
None |
baseline_risk_score |
float | Latent baseline deterioration risk at admission on a 0–1 scale (simulation parameter, not a clinical score) | ~0.03–0.98 | None |
los_hours |
int | Length of stay in hours (capped at 72 hours) | 12–72 | None |
deterioration_event |
int (0/1) | Indicator for any clinical deterioration event during the stay | 0 = no event, 1 = at least one event | None |
deterioration_within_12h_from_admission |
int (0/1) | Deterioration event occurs within the first 12 hours from admission | 0, 1 | None |
deterioration_hour |
int | Hour-from-admission of the first deterioration event; -1 means no event during the stay |
-1 (no event) or 0–(los_hours - 1) (up to 71 in this dataset) | None |
2. vitals_timeseries.csv — Hourly vital signs
Granularity: one row per (patient_id, hour_from_admission) for vital signs.
| column_name | type | description | allowed_values / range | missing_values |
|---|---|---|---|---|
patient_id |
int | Patient identifier (matches patients.csv) |
1–10,000 | None |
hour_from_admission |
int | Hour index from admission (0 = admission hour) | 0–(los_hours - 1), up to 71 in this dataset | None |
heart_rate |
float | Heart rate at that hour (beats per minute) | ~40–180 | None |
respiratory_rate |
float | Respiratory rate at that hour (breaths per minute) | ~8–45 | None |
spo2_pct |
float | Peripheral oxygen saturation percentage | ~70–100 | None |
temperature_c |
float | Body temperature in degrees Celsius | ~35.2–40.5 | None |
systolic_bp |
float | Systolic blood pressure (mmHg) | ~70–185 | None |
diastolic_bp |
float | Diastolic blood pressure (mmHg) | ~40–110 | None |
oxygen_device |
category | Type of oxygen delivery device | "none", "nasal", "mask", "hfnc", "niv" |
None |
oxygen_flow |
float | Oxygen flow rate (liters per minute). Exactly 0.0 whenever oxygen_device == "none", and strictly positive only when an oxygen device is in use. |
0–~60 (only >0 for "nasal", "mask", "hfnc", "niv") |
None |
mobility_score |
int | Ordinal mobility score (higher = more independent) | 0–4 (ordinal scale) | None |
nurse_alert |
int (0/1) | Whether a nurse alert was triggered in that hour | 0, 1 | None |
3. labs_timeseries.csv — Hourly lab results
Granularity: one row per (patient_id, hour_from_admission) for lab values.
| column_name | type | description | allowed_values / range | missing_values |
|---|---|---|---|---|
patient_id |
int | Patient identifier (matches patients.csv) |
1–10,000 | None |
hour_from_admission |
int | Hour index from admission (aligned with vital signs) | 0–(los_hours - 1), up to 71 in this dataset | None |
wbc_count |
float | White blood cell count (arbitrary clinical units, simulation-based) | ~2–30 | None |
lactate |
float | Serum lactate level (approx. mmol/L) | ~0.5–8.0 | None |
creatinine |
float | Serum creatinine (approx. mg/dL) | ~0.4–4.5 | None |
crp_level |
float | C-reactive protein level (approx. mg/L) | ~0–250 | None |
hemoglobin |
float | Hemoglobin concentration (approx. g/dL) | ~7–17 | None |
sepsis_risk_score |
float | Latent sepsis risk score at that hour on a 0–1 scale (simulation parameter) | ~0.02–1.00 | None |
4. hospital_deterioration_hourly_panel.csv — Full joined hourly panel
Granularity: one row per (patient_id, hour_from_admission) combining:
- Hourly vital signs
- Hourly lab values
- Static patient features
- All deterioration labels
This file is convenient for EDA and modeling where you want everything in one table.
| column_name | type | description | allowed_values / range | missing_values |
|---|---|---|---|---|
patient_id |
int | Patient identifier | 1–10,000 | None |
hour_from_admission |
int | Hour index from admission | 0–(los_hours - 1), up to 71 in this dataset | None |
heart_rate |
float | Heart rate (beats per minute) at this hour | ~40–180 | None |
respiratory_rate |
float | Respiratory rate (breaths per minute) | ~8–45 | None |
spo2_pct |
float | Peripheral oxygen saturation (%) | ~70–100 | None |
temperature_c |
float | Body temperature in °C | ~35.2–40.5 | None |
systolic_bp |
float | Systolic blood pressure (mmHg) | ~70–185 | None |
diastolic_bp |
float | Diastolic blood pressure (mmHg) | ~40–110 | None |
oxygen_device |
category | Oxygen delivery device | "none", "nasal", "mask", "hfnc", "niv" |
None |
oxygen_flow |
float | Oxygen flow rate (L/min). Exactly 0.0 whenever oxygen_device == "none", and strictly positive only when an oxygen device is in use. |
0–~60 (only >0 for "nasal", "mask", "hfnc", "niv") |
None |
mobility_score |
int | Ordinal mobility score | 0–4 | None |
nurse_alert |
int (0/1) | Nurse alert triggered during this hour | 0, 1 | None |
wbc_count |
float | White blood cell count | ~2–30 | None |
lactate |
float | Serum lactate | ~0.5–8.0 | None |
creatinine |
float | Serum creatinine | ~0.4–4.5 | None |
crp_level |
float | C-reactive protein | ~0–250 | None |
hemoglobin |
float | Hemoglobin | ~7–17 | None |
sepsis_risk_score |
float | Latent sepsis risk score (0–1) | ~0.02–1.00 | None |
age |
int | Age at admission | 18–90 | None |
gender |
category | Biological sex | "M", "F" |
None |
comorbidity_index |
int | Aggregate comorbidity burden | 0–8 | None |
admission_type |
category | Admission route | "ED", "Elective", "Transfer" |
None |
baseline_risk_score |
float | Latent baseline deterioration risk at admission on a 0–1 scale (simulation parameter) | ~0.03–0.98 | None |
los_hours |
int | Length of stay in hours | 12–72 | None |
deterioration_event |
int (0/1) | Any deterioration event during the stay | 0, 1 | None |
deterioration_within_12h_from_admission |
int (0/1) | Deterioration occurs within the first 12 hours | 0, 1 | None |
deterioration_hour |
int | Hour of first deterioration event; -1 = no event |
-1 (no event) or 0–(los_hours - 1) | None |
deterioration_next_12h |
int (0/1) | Label: deterioration occurs after this hour and within the next 12 hours (see definition below) | 0, 1 | None |
Definition of deterioration_next_12h:
For a given row with (patient_id, hour_from_admission = t) and corresponding deterioration_hour = h:
deterioration_next_12h = 1ift < h ≤ t + 12deterioration_next_12h = 0otherwise (including no event in the stay).
5. hospital_deterioration_ml_ready.csv — ML-ready classification panel
Granularity: one row per hourly observation (per patient and hour_from_admission).
Intended use: feed directly into ML models for next-12h deterioration prediction.
This file keeps only features + target, and omits identifiers and auxiliary targets to reduce leakage risk.
| column_name | type | description | allowed_values / range | missing_values |
|---|---|---|---|---|
hour_from_admission |
int | Hour index from admission | 0–71 (per-stay capped length in this dataset) | None |
heart_rate |
float | Heart rate (beats per minute) | ~40–180 | None |
respiratory_rate |
float | Respiratory rate (breaths per minute) | ~8–45 | None |
spo2_pct |
float | Peripheral oxygen saturation (%) | ~70–100 | None |
temperature_c |
float | Body temperature (°C) | ~35.2–40.5 | None |
systolic_bp |
float | Systolic blood pressure (mmHg) | ~70–185 | None |
diastolic_bp |
float | Diastolic blood pressure (mmHg) | ~40–110 | None |
oxygen_device |
category | Oxygen delivery device | "none", "nasal", "mask", "hfnc", "niv" |
None |
oxygen_flow |
float | Oxygen flow rate (L/min). Exactly 0.0 whenever oxygen_device == "none", and strictly positive only when an oxygen device is in use. |
0–~60 (only >0 for "nasal", "mask", "hfnc", "niv") |
None |
mobility_score |
int | Ordinal mobility score | 0–4 | None |
nurse_alert |
int (0/1) | Nurse alert active in this hour | 0, 1 | None |
wbc_count |
float | White blood cell count | ~2–30 | None |
lactate |
float | Serum lactate | ~0.5–8.0 | None |
creatinine |
float | Serum creatinine | ~0.4–4.5 | None |
crp_level |
float | C-reactive protein | ~0–250 | None |
hemoglobin |
float | Hemoglobin | ~7–17 | None |
sepsis_risk_score |
float | Latent sepsis risk score (0–1) | ~0.02–1.00 | None |
age |
int | Age at admission | 18–90 | None |
gender |
category | Biological sex | "M", "F" |
None |
comorbidity_index |
int | Aggregate comorbidity burden | 0–8 | None |
admission_type |
category | Admission route | "ED", "Elective", "Transfer" |
None |
deterioration_next_12h |
int (0/1) | Target label: deterioration occurs after this hour and within the next 12 hours | 0, 1 | None |
The definition of deterioration_next_12h is identical to the one in hospital_deterioration_hourly_panel.csv.