You are an administrator at a large urban hospital. Many of the hospital’s patients are elderly, and use Medicare to pay a portion of what the hospital charges. You wonder:

• How your hospital’s charges for various medical conditions compare to other hospitals in the state and around the country?
• In terms of the percentage of the costs that are covered by Medicare Plan A payments, how does your hospital compare to other hospitals?

You ask one of the hospital’s data scientists to provide some insight your questions.

The U.S. Government provides by-hospital public data:

Centers for Medicare and Medicaid Services

• Diagnostic-Related Condition (DRG)
• Discharges: number of patients treated for a given DRG, by a given hospital, in a given year)
• Medicare Payments: how much Medicare Plan A paid for the the discharges associated with the DRG during the year, for that hospital.

The data scientist might produce an interactive app like the following:

https://homer.shinyapps.io/medicare

• Downloading from the cms.gov site is not difficult, but …
• the data format varied from year to year

So the data scientist had to do some programming in order to combine the individual-year data sets.

• CMS data was quite repetitive, resulting in files too large for an app to handle quickly.
• We had to separate the data into smaller “relational” tables that can be joined quickly in response to particular queries.

• CMS data tells us the zip code for each provider.
• We had to use external data to determine the county for each zip code.

• Spatial data is a large, important topic in contemporary data science.
• You have to think about
  • projection systems
  • standards for encoding spatial data
  • how to access the right online mapping service,
  • etc.

• What factors affect total costs for a given DRG?
• What factors affect the percentage covered by Medicare?

We probably need access to per-patient data, not just per-hospital data.