Supplement 2018 HCRIS data with 2017 HCRIS data

[lossyrob]

There are some issues with HCRIS data, e.g. where ICU inpatient days is > ICU available bed days, that could be mitigated by utilizing 2017 HCRIS reporting data where 2018 has missing or bad information.

The goal of this issue is to load HCRIS 2017 data into the Processing_HCRIS_Data notebook and utilize it to correct errors in HCRIS data. Part of this issue is collecting all the errors that could utilize this method; please discuss in the Gitter.

[daveluo]

We should pull in the excellent work by Jacob Fenton and team who already worked through and combined the 2017-2019 HCRIS data on facility bed and occupancy reports: https://github.com/jsfenfen/covid_hospitals_demographics

Briefly looking through their hospital_data.csv (converted to gdoc version here), there's

• 4710 facilities (compared to our combined DH+HCRIS set of 6346 facilities)
• 791K All staffed beds, column all_beds_2700 (compared to our 809K)
• 77K ICU staffed beds, column all_icu_beds (compared to our 68K)
  ^ sorry if I've misinterpreted any of the columns.

I've spoken directly with Jacob on the CodeforPhily chime-analysis slack and he's happy for us to pull this in. Also looping him into convo here (@jsfenfen) if any comments/questions with our approach.