Validate hospital bed stats against other datasets

[lossyrob]

The goal of this issue is to cross validate the statistics in the dataset generated by this project for the number of hospital beds per county against external datasources. Any mismatches of data should be documented and accounted for; numbers won't match exactly but making sure external sources are indicating that the number's were generating are at least close will provide a great test to make sure we're generating valid data.

https://www.kff.org/other/state-indicator/beds-by-ownership/?currentTimeframe=0&sortModel=%7B%22colId%22:%22Location%22,%22sort%22:%22asc%22%7D

• 2.4 staffed beds per 1,000 pop across USA

https://www.sccm.org/Communications/Critical-Care-Statistics

AHA data: According to the AHA 2015 annual survey, the United States had

• 4862 acute care registered hospitals;
  • 2814 of these had at least 10 acute care beds and at least 1 ICU bed.
• These hospitals had a total of 540,668 staffed beds and 94,837 ICU beds (14.3% ICU beds/total beds) in 5229 ICUs.
  • There were 46,490 medical-surgical beds in 2644 units,
  • 14,731 cardiac beds in 976 units,
  • 6588 other beds in 379 units,
  • 4698 pediatric beds in 307 units, and
  • 22,330 neonatal beds in 920 units.
• The median number of beds in medical-surgical, cardiac, and other units was 12, with 10 beds in pediatrics and 18 in neonatal. Fifty-two percent of hospitals had 1 unit, 24% had 2 units, and 24% had 3 or more units.

HCRIS data:

• In 2010 there were 2977 acute care hospitals with ICU beds.
• In these, there were 641,395 total acute care beds with 103,900 ICU beds (16.2% ICU beds/total beds).
• From 2000 to 2010, the number of critical care beds in the United States increased by 17.8%, from 88,235 to 103,900. However, the majority of the growth in critical care bed supply is occurring in a small number of U.S. regions that tend to have large populations, fewer baseline ICUs per 100,000 capita, higher baseline ICU occupancy, and increased market competition. Additionally, between 2000 and 2010, the greatest percentage increases were in neonatal beds (29%), followed by adult beds (26%); there were minimal changes in pediatric beds (2.7%).
• Of the 103,900 ICU beds in 2010,
  • 83,417 (80.3%) were adult,
  • 1917 (1.8%) were pediatric, and
  • 18,567 (17.9%) were neonatal.
• In total, there were 33.6 beds per 100,000 population,
  • 35.5 beds per 100,000 adult beds (age > 18 years),
  • 2.7 beds/100,000 pediatric beds (age 1-17 years),
  • and 470 beds/100,000 neonatal beds (age < 1 year).

ICU days: HCRIS analysis showed that there were 150.9 million hospital days, including 25 million ICU days in 2010 (16.5% ICU days/total days). Medicare accounted for 7.9 million ICU days (31.4%) and Medicaid 4.3 million ICU days (17.2%).

Occupancy: Occupancy rates were calculated from HCRIS (days/possible days) data. In 2010, hospital and ICU occupancy rates were 64.6% and 68%, respectively. Occupancy rates vary by hospital size, with higher occupancy rates associated with larger hospitals.

[lossyrob]

https://github.com/daveluo/covid19-healthsystemcapacity/issues/7#issuecomment-599308581

[daveluo]

For another source of beds data to crosscheck and sanity check our methodology, here is the American Hospital Directory data which gives by state and nationally the total staffed beds count using the same HCRIS data source as the approach in #21.

https://www.ahd.com/definitions/statistics.html:

Staffed Beds Numbers of staffed beds are taken from a hospital's most recent Medicare cost report (W/S S-3, Part I, col.1). Cost report instructions define staffed beds as, "the number of beds available for use by patients at the end of the cost reporting period. A bed means an adult bed, pediatric bed, birthing room, or newborn bed maintained in a patient care area for lodging patients in acute, long term, or domiciliary areas of the hospital. Beds in labor room, birthing room, postanesthesia, postoperative recovery rooms, outpatient areas, emergency rooms, ancillary departments, nurses' and other staff residences, and other such areas which are regularly maintained and utilized for only a portion of the stay of patients (primarily for special procedures or not for inpatient lodging) are not termed a bed for these purposes." The total number of general med/surg beds plus special care beds are reported:

• General Medical/Surgical Beds are the beds used for routine care.
• Special Care Beds include Intensive Care Units, Coronary Care Units, etc.

Source	Total Staffed Beds count (all states & territories)
AHD.com	745,957
Our notebook	742,562

Still a slight discrepancy to figure out (maybe AHD.com uses a different FY of HCRIS reports vs us using FY2018) but more in the right neighborhood than the ~200K difference in beds that AHA's numbers show per https://github.com/daveluo/covid19-healthsystemcapacity/issues/7#issuecomment-599308581

[lossyrob]

Difference between KFF hospital bed counts and the v2 HCRIS data (based on us census 2019 estimates), ordered by difference.

[lasingallday]

I've started looking into comparisons of these datasets, and I agree that SCCM data will vary from that of KFF and AHD. From what I've read on the links provided, SCCM comes from a combination of 2015 AHA Annual Survey data and 2010 HCRIS data. So SCCM results will differ from KFF results (which use 2018 AHA Annual Survey data) and AHD results (which use a combination of the following: most recently reported HCRIS dataset (assuming 2018), MedPAR, OPPS, and propietary sources). Additionally, I feel like AHD should be the most trusted, as it is merging multiple datasets.

I also agree that there is something to look into in regards to the differences in total staffed beds between KFF and v2 HCRIS, particularly for the states . @lossyrob Is there a particular notebook to use for looking into these state-level total staffed bad differences? Or, alternatively, are these datasets already loaded and I can see the raw data in python? Lastly, is it fine for me to roll up the raw data to state-level, if such needs to be done?

[Yhinkar]

I found this issue on ovio.org and would love to contribute!