At DoltHub we never like to go a day without a bounty. Having just wrapped up our US Housing Prices Bounty we announce without further ado...

$15,0000 US Hospitals Bounty: "third time's the charm"

If you haven't been following us, our past hospitals bounties were our way of pooling 5000+ hospital chargemasters into a single, queryable, and updatable Dolt database. Though hospitals are legally required to make their chargemasters public and machine readable, an (un?)surprisingly high number of hospitals make finding these price lists as hard as legally possible -- if they publish them at all. And because there's no open, centralized compilation of them, it's hard to compare prices between hospitals.

Why try a third time?

Our past efforts were partial successes and learning experiences. We wanted to try out a new payment model and do a better job of producing a clean, researchable table.

We want a simpler, cleaner table

On our first try we allowed users to build the table of hospitals, but the schema, we felt, was messy. On our second try we added a detailed schema with a separate payers table, plus caveats, and modes. Both times, hospitals were identified by their npi_number, and procedures were identified by their codes which, in theory, were shared between hospitals.

But in practice, our assumptions about the schema were only partially true: there are multiple NPI numbers associated with a given hospital, and codes (like CPT and HCPCS) were often reused for multiple procedures, leading to a disorganized table of hospitals and ambiguous codes like C1387, C1387,1, C1387.21, etc.

This time we're starting with the hospitals table already filled out, and we've simplified the schema to two tables: hospitals and prices. A "price" is a combination of hospital cms_certification_num, a code, a payer, and a code_disambiguator. That last column is a catch-all column for the case where there might not be enough data to totally disambiguate the codes for a hospital -- we hope to not have to use it.

As a side benefit of the simplified table, we'll see how each hospital lists its descriptions for each generic code.

A new incentive model

Our previous payout schemes incentivized our participants to collect hospitals with lots of payers and prices. Our participants shooting for the biggest hospitals led to a table with a lot of prices in it but with only 20% hospital coverage. Plus, it made it harder for the latecomers to make money. We hope to change that this time by paying out a fixed $10 per hospital, no matter how many prices are in the chargemaster. And with the hospital table already filled out, our participants will always have clear guidance on where to look.

Schema, plus a twist on the old scorebard

This time we have just two tables: hospitals and prices. The tables are mostly self-explanatory except for we have a new way of recording ownership of a row.

Getting on the scoreboard with the last_edited_by_username column

Whenever you add or edit a hospital's prices you put your username in this column. This claims your ownership of that hospital. You have to input your username to get credit.

For more information on the schema see the README on the bounty page (click "About").

We're paying $10 per hospital, throughout the bounty, up to $15k. Once (or if) we reach 1500 hospitals, we'll cap the max payout and lower the per-hospital rate.

Questions, comments?

Click the invite link to visit us on Discord. We're always here to answer questions.