Managing COVID-19 ICU Need: LA Counties Have 1.5 Weeks till ICU Full, as of 3/26/20

ICU capacity is a primary driver of health outcomes as we manage the COVID-19 pandemic in the US. We should monitor projected COVID-19 need by county with ICU availability in that county using daily case count, annual hospital cost reporting, and demographic-based rate of need. If done properly, counties (and states) can take aggressive measures sooner or determine how to provide testing and care between populations more effectively. The state level, by comparison, is too broad, and the Hospital Referral Region too detached from elected official jurisdiction.

Applying this model to Louisiana counties (parishes) as of 3/26 A.M., we find that for counties with more cases such as Orleans, Jefferson, East Baton Rouge, and Ascension, they might have 15, 18, 12, and 13 days until they reach their ICU capacity (the table below illustrates).

Daily case count per county is relatively easy to gather from sources around the web, and from there you can calculate the "line of best fit" growth rate. USAFacts.org has great daily case counts per county. USAFacts also has FIPS codes for each county.

Determining ICU Bed Availability by County from annual hospital cost reporting is more difficult. CMS releases self-reported hospital/provider cost data, Hospital Cost Report (HCRIS), which is very messy from varying report types, reporting periods, quality of information (lots of clerical errors). Adam Sacarny has done an excellent job consolidating and normalizing all of this information and making it publicly available, http://sacarny.com/data/. Some providers in the data lack county name, and once you fill that in, you can then map county name to FIPS codes. From there, aggregate ICU bed count by all providers within a county.

One more step on ICU availability—the raw count of ICU beds does not reflect how many are currently available for incoming patients. To estimate this, we can divide two variables from the HCRIS, Inpatient Bed Days per year and Available Bed Days per year. That quotient provides the average bed availability throughout the year, which we can assume is what's currently available. (And then, once more, sum all providers' availability by county.)

Demographics have a huge impact on what percentage of COVID-19 cases need ICU beds. From Array Architects, "Reports from China’s Hubei Province and Northern Italy suggest radically different rates of hospitalization (20% vs. 50%) correlated to their respective over-65 populations (12% vs. 23%)." Using US Census data, we can apply these findings to each county's population by age group and determine a unique "ICU rate of need." (In the table above, I used a standard 7.56% rate of need based on Array Architect's model that assumes 27% need hospitalization, and of those, 28% of need ICU beds.)

At this point, we can determine the estimated "ICU-Case Capacity," or the number of COVID-19 Cases the county's providers can support given our assumptions on the percentage of cases that will need ICU beds.

Finally, to estimate how many days are left until ICU capacity is reached, we will take the estimated ICU bed availability, the most recent case count, the growth rate, and find the day at which the case count would equal or exceed the ICU-Case Capacity.

Armed with this information, elected officials at the county and state levels can take the right measures, sooner, to either prevent spread of COVID-19 among their population or to prepare their hospital systems to meet anticipated demand.

While tracking these numbers at the state level is useful, I would argue it's drastically less useful. Clusters don't occur in states but within cities and communities, and preventing a cluster from spreading should involve taking action at the level of that city or community by enforcing social distancing, shelter in place, or migration restriction policies, for example. (If enough clusters spread throughout state, then it does become a state problem.) Then, managing ICU need occurs at the local level, because of our highly decentralized healthcare system and the highly varying rate of ICU need, which is based on the demographics of each population.

Another alternative way to estimate and manage ICU need is the "Hospital Referral Region" unit. Healthcare in the US is really provided on a "Hospital Referral Region" basis that the Dartmouth Atlas of Healthcare explains further. In short, these regions pay little heed to county or state lines. My take on how this impacts a county-level analysis is that, for counties with sufficiently small populations, they should probably be lumped in with some portion of their neighboring counties' population and ICU capacity. At the moment, I don't think this is worth doing, since it is ultimately counties and states which will take (the most impactful) action against this pandemic.