[Updated] Can Hawaii's Hospitals Meet Peak COVID-19 Demand?
[Updated April 7, 2020]
NOTE: The model referenced below is not maintained by the Hawaii Data Collaborative. We are reporting on the results of the model as of the date this article is posted, and encourage you to refer directly to the source for updates going forward.
The University of Washington Institute for Health Metrics and Evaluation (IHME) model for Hawaii was updated on April 6, 2020, with the following changes:
Hawaii’s peak resource usage date moved from May 3, 2020 to April 12, 2020
General bed shortage in Hawaii now not expected at the new peak utilization of April 12, 2020, the projection is corrected to 631 beds at peak.
The ICU bed shortage is still projected and has been kept stable at 66 ICU bed shortage from previous updates.
Source: University of Washington, Institute for Health Metrics and Evaluation (IHME), COVID-19 Projections
Why Hawaii’s Model Changed
Hawaii’s peak resource usage date moved up from May 3, 2020 to April 12, 2020 because of additional data coming in from Spain and Italy that points to a shorter amount of time from the implementation of social distancing policies to the peak of daily deaths.
General bed projection corrected down to 631 general beds needed out of 956 general beds available at peak utilization on April 12, 2020. This correction is because of updated state-level data on the ratio of hospital admissions to COVID-19 deaths, which informs model parameters that are used to project need for hospital beds, ICU beds, and ventilators. Although Hawaii does not yet have enough confirmed COVID-19 deaths to calculate a Hawaii-specific ratio, the UW model now assigns Hawaii a pooled ratio (derived from data from other states) that is lower than previous CDC estimates, contributing to a lower peak need for general beds.
ICU bed shortage still exists at 111 ICU beds needed out of 45 ICU beds available at peak utilization on April 12, 2020. Although the projected number of needed beds total went down for Hawaii, projected need for ICU beds has not. The model re-estimated length of stay in hospital for COVID-19 patients. ICU patients have recently been shown to have longer hospital stays than previously predicted, while general patients have been shown to have shorter hospital stays. Long hospital stays for ICU patients mean that the reduction in ICU bed need is lower than for general bed need.
Lastly, IHME reports they have adopted a new way of estimating uncertainty that is purportedly more "rigorous.” However, the uncertainty bands (shaded area around the curve) are now much wider, even at the national level. We will be able to understand this better as the model’s technical documentation is updated.
[Original post, April 3, 2020]
We have been working to evaluate multiple modeling tools for projecting peak hospital demand and capacity sufficiency in Hawaii. The University of Washington (UW) Institute for Health Metrics and Evaluation (IHME) model was identified as best suited for projecting Hawaii hospital capacity in the coming weeks. Assuming continued social distancing through May, this model currently predicts:
Hawaii’s peak hospitalization will occur on May 3, 2020, with 1,134 hospitalized for COVID-19.
At that time the state will be short 178 general beds and 125 ICU beds.
Source: University of Washington, Institute for Health Metrics and Evaluation (IHME), COVID-19 Projections
It is important to note that the UW model suggests capacity shortages could be as high as 977 general and 261 ICU beds, at peak hospitalization. However, this is at the upper bound of the model’s prediction – it is possible, but not probable.
Lastly, we also employed a second model created Penn Medicine (University of Pennsylvania), which predicts slightly higher but comparable numbers. We do not report the numbers from this second model, because it was designed primarily for assessing individual hospital capacity, not generalized capacity across an entire regional system (the strengths and limitations of these two models are detailed below).
Of the models we have analyzed, we find that the University of Washington model is best suited for state-level health care and policy planning, while the University of Pennsylvania Medicine model is most useful for hospital-level modeling. More specifically, the UW COVID-19 model can inform state-level efforts to prepare healthcare providers for overall resource shortages, while the Penn Medicine model can help individual hospitals to run their own projections and prepare for anticipated resource needs.
University of Washington Model: State-level Decisions
The UW model was built for nationwide, and then state-by-state, projections.
Strengths: Can provide a bird’s eye view of projected hospital resource use for Hawaii; built on available state-level data, the projections from this model can serve as a baseline point of reference for the state
Limitations: Licensed bed and ICU capacity, as well as utilization rates, were obtained from the American Hospital Association, and may not reflect current conditions.
Projections assume full social distancing through May 2020.
Penn Medicine Model: Hospital-level Decisions
The Penn Medicine model allows for user input for variables such as local infection rate, doubling time, and social distancing efficacy. Because this model was built to project resource consumption for individual hospitals, local healthcare officials can input severity parameters (such as the proportion of COVID-19 patients who are hospitalized in their catchment area) as the situation develops.
Strengths: Provides more granular projections for specific hospitals; any local hospital can tailor the Penn Medicine model to generate hospital-level resource consumption for its specific catchment area.
Limitations: Requires input of accurate, reliable parameter data to yield robust hospital impact model results; ideally, hospital-level catchment area populations (rather than state or regional population data) would be used for this purpose.
Further Considerations
As with any model, the UW and Penn Medicine COVID-19 models are only as accurate in anticipating hospital resource usage as the underlying data on which their projections are based. Ensuring that up-to-date, accurate data regarding available hospital resources and COVID-19 infection and mortality rates are made publicly available is critical to the utility of these models.
