|
|
--- |
|
|
license: other |
|
|
license_name: commercial |
|
|
license_link: LICENSE |
|
|
task_categories: |
|
|
- text-generation |
|
|
- feature-extraction |
|
|
- summarization |
|
|
- tabular-to-text |
|
|
- table-to-text |
|
|
- text-retrieval |
|
|
tags: |
|
|
- medical |
|
|
- meld |
|
|
- nlp |
|
|
- manuscript |
|
|
- emrs |
|
|
- ehrs |
|
|
- rwd |
|
|
- rwe |
|
|
- harvard |
|
|
- ibm |
|
|
- mgb |
|
|
- mgh |
|
|
- liver |
|
|
- hepatology |
|
|
- predict |
|
|
- unos |
|
|
--- |
|
|
|
|
|
# Synthetic MELD-Plus (1M Patients) |
|
|
|
|
|
This dataset contains **1,000,000 synthetic patients** inspired by the published [MELD-Plus study](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0186301) (a collboration between Massachusetts General Hospital and IBM Research). Each row corresponds to a single admission, with demographics, labs, comorbidities, medications, derived scores (MELD, MELD-Na, MELD-Plus), and the binary outcome **Death_Within_90_Days**. |
|
|
|
|
|
All data are **artificially generated** and contain **no identifiable patient records**. |
|
|
|
|
|
--- |
|
|
|
|
|
## Source and Augmentation |
|
|
|
|
|
- **Original study:** The MELD-Plus study described ~5,000 admissions across its main manuscript and four supplementary documents. These reported **summary statistics only** (means, SDs, prevalences, ranges, quartiles, and units). |
|
|
- **Augmentation process to 1M patients:** |
|
|
1. **Extracted variables** (covariates, outcomes, descriptive stats) from main + supplementary files. |
|
|
2. **Simulated distributions** for continuous labs (Normal with reported mean/SD, with physiologic plausibility bounds). |
|
|
3. **Applied prevalence rates** for comorbidities (zero-inflated Poisson) and for missingness in labs. |
|
|
4. **Modeled medications** with Poisson counts. |
|
|
5. **Computed derived scores:** MELD, MELD-Na, MELD-Plus. |
|
|
6. **Generated outcomes:** Death_Within_90_Days simulated via MELD-Plus logistic model, calibrated to match ~16.3% mortality. |
|
|
7. **Scaled up** to 1,000,000 patients, each with one admission, preserving distributions and correlations. |
|
|
|
|
|
--- |
|
|
|
|
|
## Schema (Highlights) |
|
|
|
|
|
- **Demographics:** Age, Gender, Ethnicity, MaritalStatus, BMI, Insurance (Medicaid/Medicare/Other), Admissions_Prior12mo |
|
|
- **Labs:** TotalBilirubin, Creatinine, INR, Sodium, Albumin, WBC |
|
|
- **Comorbidities:** 20+ variables (e.g., Ascites, HepaticEncephalopathy, Diabetes, Hypertension, COPD) |
|
|
- **Medications:** Anticoagulants, Antiplatelets, Antiarrhythmics_Diuretics, Aspirin, Cardiovascular, DiabetesMeds, etc. |
|
|
- **Derived:** MELD, MELD_Na, MELD_Plus, OnDialysis, Death_Within_90_Days |
|
|
|
|
|
--- |
|
|
|
|
|
## Example Usage |
|
|
|
|
|
```python |
|
|
import pandas as pd |
|
|
|
|
|
df = pd.read_csv("meldplus_synthetic_1m.csv") |
|
|
print(df.shape) # (1000000, ~50 columns) |
|
|
print(df.head()) |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
- **Educational & personal learning** |
|
|
- **Benchmarking methods** for EMR preprocessing, feature extraction, and survival analysis |
|
|
- **Synthetic data methodology testing** |
|
|
|
|
|
Not for clinical decision-making. |
