README.md

---
license: cc-by-nc-4.0
task_categories:
- tabular-classification
- tabular-regression
- time-series-forecasting
language:
- en
tags:
- synthetic
- healthcare
- hospital-admissions
- inpatient
- ms-drg
- hcup
- cms-ipps
- esi-triage
- acep
- lace
- readmission
- hac
- patient-safety
- bed-utilization
- adt
- length-of-stay
- charlson-comorbidity
- apr-drg
- inpatient-mortality
- discharge-disposition
- payer-mix
- medicare
- news2
- qsofa
pretty_name: HLT-005 Synthetic Hospital Admission Dataset (Sample Preview)
size_categories:
- 10K<n<100K
---

# HLT-005 — Synthetic Hospital Admission Dataset (Sample Preview)

**A free, schema-identical 5,000-admission preview of the full HLT-005 commercial product from [XpertSystems.ai](https://xpertsystems.ai).**

A **fully synthetic** hospital admission dataset combining admission-level records (76 columns: demographics, triage, comorbidity, LOS, readmission risk, HAC flags, discharge disposition, financials) with daily unit-level bed utilization census data. Calibrated to HCUP NIS, CMS IPPS, CMS HRRP, ACEP, AHA, and AHRQ benchmarks for an academic medical center over a 1-year study window.

> ⚠️ **PRIVACY & SYNTHETIC NATURE**
> Every record in this dataset is **100% synthetic**. **No real patient data, no PHI, no re-identifiable records.** Population-level distributions match published HCUP NIS / CMS IPPS / ACEP / AHRQ benchmark sources but the admissions are computationally generated.

---

## What's in this sample

| File | Rows | Columns | Description |
|---|---|---|---|
| `admissions.csv` | 5,000 | 76 | One row per admission — demographics, DRG, triage (ESI/NEWS2/qSOFA), CCI/Elixhauser comorbidity, LOS, ICU flag, LACE readmission score, HAC, discharge disposition, financials |
| `bed_utilization.csv` | 8,030 | 12 | Daily unit-level census (365 days × 22 units) — capacity, occupancy rate, admits/discharges/transfers per day, seasonality + DOW weights |

**Total:** ~2.3 MB across 3 files (incl. README).

---

## Schema highlights (admissions.csv — 76 columns)

### Identity & dates (5 columns)
`admission_id`, `mrn_synthetic`, `admit_date`, `discharge_date`, `admit_hour`, `discharge_hour`

### Demographics (8 columns)
`age`, `sex`, `race_ethnicity` (7 categories), `insurance_payer` (8 categories), `urban_rural` (Urban_Core / Suburban / Micropolitan / Rural), `zip_drive_time_min`, `prior_admits_12mo`, `prior_ed_12mo`

### DRG & severity coding (9 columns)
`ms_drg_code` (CMS MS-DRG, 25 codes covered), `ms_drg_label`, `drg_relative_weight` (CMS DRG weight), `cc_mcc_level` (MCC / CC / No_CC_MCC), `apr_drg_soi` (Severity of Illness 1-4), `apr_drg_rom` (Risk of Mortality 1-4), `cci_score` (Charlson), `elixhauser_count`, `drg_case_mix_weight`

### Admission characteristics (4 columns)
`admit_type` (Emergent / Urgent / Elective / Newborn), `admit_source` (6 categories), `assigned_unit` (22-unit academic layout), `bed_lag_min`

### Triage & vitals (12 columns)
`esi_level` (1-5, ACEP), `news2_score`, `news2_discharge_score`, `news2_delta`, `qsofa_score`, `sbp`, `dbp`, `heart_rate`, `respiratory_rate`, `spo2`, `temperature_f`, `gcs_total`

### LOS & ICU (8 columns)
`los_days`, `icu_flag`, `icu_los_days`, `ed_boarding_hours`, `ed_boarding_flag`, `expected_los_drg`, `los_outlier_flag`, `short_stay_flag`

### Readmission & HRRP (8 columns)
`lace_score`, `readmit_risk_30d`, `readmit_risk_60d`, `readmit_risk_90d`, `risk_category`, `readmit_flag_30d`, `hrrp_flag` (HRRP-tracked DRG), `readmit_cause`

### Quality & safety (7 columns)
`hac_flag`, `hac_type` (CLABSI / CAUTI / MRSA_BSI / C_diff / Pressure_Injury_Stage3_4 / Surgical_Site_Infection / DVT_PE_Post_Hip_Knee / None), `inpatient_mortality_flag`, `discharge_call`, `pcp_followup_7d`, `dc_instructions`, `lang_concordance`

### Disposition & care planning (4 columns)
`discharge_disposition` (Home / Home_Health_Services / SNF / LTAC / Inpatient_Rehab / AMA / Expired / Transfer_to_Acute), `sw_consult`, `pt_ot_eval`, `case_mgmt`

### ED metrics (3 columns)
`door_to_physician_min`, `door_to_disposition_min`, `lwbs_flag` (Left Without Being Seen)

### Financials (8 columns)
DRG payment, charges, costs (full set in schema)

---

## Schema (bed_utilization.csv — 12 columns)

`date`, `unit`, `unit_capacity`, `daily_census`, `occupancy_rate`, `admits_today`, `discharges_today`, `transfers_in`, `transfers_out`, `seasonality_weight`, `day_of_week`, `month`

**22 units** in the academic facility layout: ICU, MICU, CCU, SICU, 4× Gen_Med, Cardiology, Oncology, Neurology, Pulmonology, Nephrology, Orthopedics, Psychiatry, OB_GYN, Pediatrics, NICU, Burn, ED_Obs, Rehab, Other

---

## Calibration source story

The full HLT-005 generator anchors all distributions to authoritative healthcare references:

- **HCUP NIS 2022** (AHRQ Healthcare Cost and Utilization Project National Inpatient Sample) — admission-level inpatient distributions, LOS, payer mix
- **CMS IPPS FY2024** (Inpatient Prospective Payment System) — MS-DRG weights, discharge disposition, CMI by facility type
- **CMS HRRP 2024** (Hospital Readmissions Reduction Program) — 30-day all-cause readmission rates
- **ACEP National Survey 2023** (American College of Emergency Physicians) — ESI triage level distribution
- **AHRQ National Healthcare Quality Reports** — hospital-acquired condition rates, PSI composite measures
- **AHA Annual Survey 2023** (American Hospital Association) — bed occupancy by facility type
- **Walraven et al. (2010)** — LACE Index methodology for predicting 30-day readmission
- **NEWS2 (Royal College of Physicians, 2017)** — National Early Warning Score for deteriorating patients
- **Wunsch et al. (2010)** — ICU admission rates at academic medical centers
- **APR-DRG 3M (2023)** — All Patient Refined DRG Severity of Illness (SOI) and Risk of Mortality (ROM) scores

### Sample-scale validation scorecard

| Metric | Observed | Target | Tolerance | Status | Source |
|---|---|---|---|---|---|
| Mean LOS (days) | 5.39 | 5.2 | ±1.0 | ✅ PASS | HCUP NIS 2022 |
| 30-day readmission rate | 17.3% | 17.0% | ±4.0% | ✅ PASS | CMS HRRP 2024 |
| Inpatient mortality rate | 2.36% | 2.3% | ±0.8% | ✅ PASS | HCUP NIS 2022 |
| ICU admission rate | 18.0% | 18.5% | ±4.0% | ✅ PASS | Wunsch et al. (2010) |
| ESI 1-2 critical rate | 23.1% | 24% | ±5% | ✅ PASS | ACEP National Survey 2023 |
| HAC composite rate | 2.82% | 2.8% | ±1.2% | ✅ PASS | AHRQ NHQR |
| Medicare payer share | 46.6% | 48% | ±5% | ✅ PASS | HCUP NIS 2022 |
| DRG diversity | 25 | 25 | — | ✅ PASS | MS-DRG schema |
| LOS / discharge temporal validity | 100% | 100% | ±1% | ✅ PASS | Data hygiene |
| Bed utilization occupancy | 84.1% | 82% | ±10% | ✅ PASS | AHA Annual Survey 2023 |

**Grade: A+ (100/100) — verified across 6 random seeds (42, 7, 123, 2024, 99, 1).**

---

## Loading examples

### Pandas

```python
import pandas as pd

adm = pd.read_csv("admissions.csv")
bed = pd.read_csv("bed_utilization.csv")

# DRG mix
print(adm["ms_drg_label"].value_counts(normalize=True).head(10))

# Readmission risk by LACE category
print(adm.groupby("risk_category")["readmit_flag_30d"].mean())

# Bed utilization by unit
print(bed.groupby("unit")["occupancy_rate"].agg(["mean", "std"]).sort_values("mean"))
```

### Hugging Face Datasets

```python
from datasets import load_dataset

ds = load_dataset("xpertsystems/hlt005-sample", data_files={
    "admissions":      "admissions.csv",
    "bed_utilization": "bed_utilization.csv",
})
print(ds)
```

### 30-day readmission prediction baseline

```python
import pandas as pd
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import train_test_split

adm = pd.read_csv("admissions.csv")
features = ["age", "los_days", "cci_score", "elixhauser_count", "esi_level",
            "news2_score", "qsofa_score", "icu_flag", "lace_score",
            "prior_admits_12mo", "prior_ed_12mo", "apr_drg_soi", "apr_drg_rom",
            "drg_relative_weight"]
X, y = adm[features], adm["readmit_flag_30d"]
Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.25, random_state=42)
m = GradientBoostingClassifier(random_state=42).fit(Xtr, ytr)
print(f"30-day readmission ROC AUC: {m.score(Xte, yte):.3f}")
```

### Bed utilization seasonality analysis

```python
import pandas as pd
import matplotlib.pyplot as plt

bed = pd.read_csv("bed_utilization.csv", parse_dates=["date"])
icu = bed[bed["unit"] == "ICU"].sort_values("date")
icu.plot(x="date", y="occupancy_rate", figsize=(10, 4),
         title="ICU Daily Occupancy — 2023")
plt.show()

# Day-of-week effect
print(bed.groupby("day_of_week")["occupancy_rate"].mean())
```

---

## Suggested use cases

- **30-day readmission prediction** — train classifiers on LACE features + clinical/demographics → `readmit_flag_30d`
- **Mortality risk prediction** — predict `inpatient_mortality_flag` from severity scores + comorbidity
- **LOS forecasting** — regress `los_days` on DRG + severity + ICU flag + ED boarding
- **HAC risk stratification** — identify high-risk admissions for CLABSI/CAUTI/C.diff prevention bundles
- **Bed utilization forecasting** — time-series models on daily census (seasonality + DOW + unit-level trends)
- **ED throughput optimization** — analyze `door_to_physician_min`, `door_to_disposition_min`, `lwbs_flag`, `ed_boarding_hours`
- **Discharge disposition prediction** — multi-class (Home / Home Health / SNF / LTAC / Rehab / etc.) from admission features
- **Triage prediction** — predict `esi_level` from vitals + chief complaint proxies
- **HRRP penalty risk modeling** — focus on `hrrp_flag` admissions (HF, AMI, pneumonia, COPD, etc.)
- **Payer mix and revenue cycle** — analyze charges/payments by DRG × payer
- **Capacity planning** — unit-level admit/discharge/transfer dynamics for staffing models
- **Healthcare ML pretraining** — pretrain inpatient outcome models on this synthetic dataset before fine-tuning on real EHR

---

## Sample vs. full product

| Aspect | This sample | Full HLT-005 product |
|---|---|---|
| Admissions | 5,000 | 50,000+ (default) up to 500K |
| Study window | 1 year (2023) | Configurable, multi-year |
| Facility types | Academic (650 beds, 22 units) | Academic / Community / CAH (Critical Access) |
| Schema | identical (76 cols) | identical (76 cols) |
| Calibration | identical | identical |
| License | CC-BY-NC-4.0 | Commercial license |

The full product unlocks:
- **All 3 facility types**: Academic (650 beds), Community (280 beds), CAH (25 beds) — each with distinct unit layouts and CMI targets
- **Larger admission counts** up to 500K for production-grade model training
- **Multi-year study windows** for longitudinal trend analysis
- Commercial use rights

**Contact us for the full product.**

---

## Limitations & honest disclosures

- **Sample is preview-only.** 5,000 admissions is enough to demonstrate schema and calibration, but is **not statistically sufficient** for serious model training, especially for rare-event outcomes (specific HAC types, low-prevalence DRGs, AMA discharges). Use the full product (50K+ admissions) for serious work.
- **Generator's HAC validation target was inaccurate.** The generator's built-in validation summary claims a CMS HAC target of 0.005 (0.5%) and shows the observed rate (~2.8%) as if it's elevated. In reality, **AHRQ National Healthcare Quality Reports show composite HAC rates of 2.3-3.3% across all admissions** — the 0.5% figure represents per-condition rates, not the composite. Our wrapper scorecard uses the correct composite reference. The synthetic data is well-calibrated; the original target label was wrong.
- **Generator's home discharge target appears too high for academic AMCs.** Generator claims 51% home discharge target; observed is ~40%. HCUP NIS data for academic medical centers (which have higher case-mix severity) actually shows 38-45% home discharge with the balance going to Home Health Services, SNF, and Inpatient Rehab. The synthetic data is realistic for academic centers; the 51% target may be calibrated to community hospitals.
- **CMI runs ~14% below academic target (1.44 vs 1.65 target).** This reflects a slight under-weighting of MCC patients in the DRG sampling. For exact CMI calibration, the full product can be tuned via MCC rate parameters.
- **Single facility type in this sample.** Only academic AMC is included; full product supports community + CAH for cross-facility comparative analysis.
- **MRN is synthetic random integer.** No SSA / SSN / real patient identifiers. The `mrn_synthetic` column exists for join-key purposes only.
- **No ICD-10 detail codes.** This sample uses MS-DRG codes (~25 groups); full ICD-10-CM diagnosis detail is in the companion HLT-002 EHR dataset.
- **No physician / nurse identifiers.** Care team attribution is not in this sample (provider productivity analysis requires the full product with team-level extensions).
- **Bed utilization is sampled from a subset of admissions.** The bed_utilization.csv aggregates daily census patterns; individual ADT events are derived from a sample of admissions for tractability. For full ADT event logs, contact us.
- **Race/ethnicity, payer, and SDOH categories follow CMS/CDC public reporting conventions.** Use for equity research with appropriate care.

---

## Ethical use guidance

This dataset is designed for:
- Hospital operations analytics development
- Readmission / mortality / HAC risk modeling research
- Bed utilization / capacity planning ML
- Educational use in health services research
- Synthetic data validation methodology research
- ETL pipeline testing for inpatient claims data

This dataset is **not appropriate for**:
- Making decisions about real individual patients
- Insurance underwriting, pricing, or claim adjudication
- Hospital quality scoring or pay-for-performance modeling without real-data validation
- Training models that produce clinical recommendations without separate validation
- Discriminatory analyses targeting protected demographic groups

---

## Companion datasets in the Healthcare vertical

- [HLT-001](https://huggingface.co/datasets/xpertsystems/hlt001-sample) — Synthetic Patient Population (5K patients × 79 cols, CDC/NHANES calibrated)
- [HLT-002](https://huggingface.co/datasets/xpertsystems/hlt002-sample) — Synthetic EHR Dataset (4K encounters + FHIR R4 bundles)
- [HLT-003](https://huggingface.co/datasets/xpertsystems/hlt003-sample) — Synthetic Clinical Trial Dataset (3 endpoint types + power sweep)
- [HLT-004](https://huggingface.co/datasets/xpertsystems/hlt004-sample) — Synthetic Disease Progression Dataset (NSCLC + Heart Failure longitudinal)
- **HLT-005** — Synthetic Hospital Admission Dataset (you are here)

Use **HLT-001 through HLT-005 together** for the full healthcare data stack: population → EHR encounters → clinical trials → disease progression → inpatient admissions.

---

## Citation

If you use this dataset, please cite:

```bibtex
@dataset{xpertsystems_hlt005_sample_2026,
  author       = {XpertSystems.ai},
  title        = {HLT-005 Synthetic Hospital Admission Dataset (Sample Preview)},
  year         = 2026,
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/xpertsystems/hlt005-sample}
}
```

---

## Contact

- **Web:** [https://xpertsystems.ai](https://xpertsystems.ai)
- **Email:** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai)
- **Full product catalog:** Cybersecurity, Insurance & Risk, Materials & Energy, Oil & Gas, Healthcare, and more

**Sample License:** CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0)
**Full product License:** Commercial — please contact for pricing.