Nason's picture
Upload README.md with huggingface_hub
0f46730 verified
|
Raw
History Blame Contribute Delete
3.21 kB
---
license: mit
task_categories:
- tabular-classification
- tabular-regression
tags:
- healthcare
- medicare
- open-payments
- sunshine-act
- pharma
- conflicts-of-interest
pretty_name: CMS Open Payments (Sunshine Act) Database
size_categories:
- 100M<n<1B
---
# Open Payments Database (CMS Sunshine Act)
Every disclosed industry payment to physicians, non-physician practitioners,
and teaching hospitals, program years 2013-2025, as a single queryable
DuckDB database. **172,450,995 rows.**
| Table | Description | Rows | Cols | Coverage |
|-------|-------------|------|------|----------|
| `general_payments` | General (non-research) payments and transfers of value to covered recipients, PY2013-2025 | 148,797,140 | 107 | PY2013 to PY2025 |
| `research_payments` | Research payments (wide form, up to 5 principal investigators per record), PY2013-2025 | 10,995,016 | 268 | PY2013 to PY2025 |
| `research_payment_investigators` | Long form of research-payment principal investigators: one row per (record, PI slot) | 10,745,869 | 14 | PY2013 to PY2025 |
| `covered_recipients` | Covered Recipient Profile Supplement: one row per physician/NPP profile with NPI | 1,697,025 | 32 | all years |
| `deleted_records` | Record IDs deleted between publications (PY2016+; earlier years never had these files) | 162,030 | 5 | PY2016 to PY2024 |
| `ownership_payments` | Physician ownership and investment interests, PY2013-2025 | 53,915 | 31 | PY2013 to PY2025 |
## Query it remotely
```sql
INSTALL httpfs; LOAD httpfs;
ATTACH 'https://huggingface.co/datasets/Nason/openpayments-database/resolve/main/openpayments.duckdb'
AS op (READ_ONLY);
SELECT applicable_manufacturer_or_applicable_gpo_making_payment_name AS maker,
ROUND(SUM(total_amount_of_payment_usdollars) / 1e6, 1) AS millions
FROM op.general_payments
WHERE program_year = 2024
GROUP BY 1 ORDER BY 2 DESC LIMIT 10;
```
Joins to the `cms-medicare` datapond database on NPI — see the GitHub README
for the worked Open Payments x Medicare utilization example and measured
join rates.
Or with the datapond package: `pip install datapond`, then
`datapond.connect('openpayments')`.
Build pipeline, per-era column crosswalks, and full documentation:
https://github.com/ian-nason/openpayments-database
# Changelog
## 2026-07-07 — Initial release
- Program years 2013-2025 from the June 2026 publication plus the frozen
archive ZIPs (2013-2018 at their final republications).
- 172.5M rows: general_payments 148.8M, research_payments 11.0M (+10.7M-row
investigators long table), ownership_payments 54k, covered_recipients
1.7M, deleted_records 162k.
- Typed columns (DECIMAL money, DATE dates, VARCHAR identifiers); per-era
column crosswalk checked into the repo (Physician_* -> Covered_Recipient_*
renames for 2013-2015; pre-2016 product columns kept era-specific).
- Validated against CMS published magnitudes (PY2025: $3.9B general +
$9.5B research + $0.2B ownership) and recognizable top manufacturers.
- Cross-database NPI join to cms-medicare measured and documented (61.6%
of PY2023 recipients; 63-83% for physician specialties; dentists ~0%
as expected — they don't bill Medicare Part B).