---
configs:
- config_name: cohort
  data_files:
  - split: train
    path: cohort_subset.csv
- config_name: covariates
  data_files:
  - split: train
    path: covariate_subset.csv
- config_name: data
  data_files:
  - split: train
    path: data_subset.csv
---

# OMOP POU Data Card

----------------------------------

## Publication

Naderalvojoud, Behzad, et al.  
"Towards global model generalizability: independent cross-site feature evaluation for patient-level risk prediction models using the OHDSI network."  
*Journal of the American Medical Informatics Association* 31.5 (2024): 1051–1061.

Please refer to the publication for full details on cohort construction and model development.

---

## Dataset Specification

- **Source of Data**: STARR OMOP  
- **Prediction outcome**: Prolonged opioid use  
  - 1: Prolonged  
  - 0: Non-prolonged  
- **Covariate count**: 10,636  
- **Total cohort size (original)**: 41,929 patients  
- **Final modeling cohort**: 41,507 patients  
- **Training set size**: 33,206  
- **Test set size**: 8,301  

---

## Note on Data Availability

Due to PHI restrictions, the full dataset cannot be publicly released. The files provided here are **small subset examples with synthetic patients**, intended only to demonstrate the **structure and format** of the dataset.

---

## File Descriptions

### `cohort_subset.csv`

Contains 10 example (synthetic) patients with:
- `rowId`: patient identifier  
- `outcome`: prolonged opioid use label  

---

### `data_subset.csv`

Represents the model input matrix:
- Each row corresponds to a patient  
- Each column corresponds to a covariate (by covariateId)  
- `rowId` links to `cohort_subset.csv`  

---

### `covariate_subset.csv`

Covariate dictionary containing:
- `covariateId`  
- `covariateName`