COVID-19 Hospitalizations Accelerate Again
In the first few days following the creation of my projections on November 8, the number of actual hospitalizations rose to match the worst-case scenario. That was a bit of a shock to me because that scenario is little more than a mathematical oddity; not something I really expected to happen.
On Monday and Tuesday the actual level of hospitalizations fell more in line with the base and best-case scenarios. While performing some work for a client I recognized a way the models could be significantly improved. The models are founded on the observation that US COVID-19 developments are similar to what is happening in Europe lagged by a few weeks. That remains true, but there is a significant difference between them that I didn't account for.
By virtue of being smaller and having less diversity of topology and climate European COVID-19 is proceeding relatively uniformly throughout each country than it is in the US. The US has three different COVI-19 waves at different stages of development. 18, primarily Midwestern states form the Under Pressure segment in which the outbreak is most severe. The three states on the West Coast along with seven in the Northeast form the Horizon segment, in which the pandemic is least severe. I estimate the development of this segment lags the Under Pressure segment by a number of weeks. The remaining 23 states I've placed in the "Cusp" segment. The pandemic is more severe in these states and rapidly advancing. i suspect in the next few weeks, the states in this segment will be the primary driver of increased COVID-19 hospitalizations.
The European model applies much more to each of these segments individually than it does the United States as a whole. The single, US model overestimates the pace of hospitalization increases of states that are in the Cusp and Horizon segments. This is why the actual results are falling more toward the best and base cases than they do the worst case (fortunately). In a future blog post I'll describe the three US segments. I'm not going to disturb the model I've created. I've identified how it could be improved, but I am content to let it play out for at least another week to see how actual results line up against it. As a side note, the Institute for Health Metrics and Evaluation updated the Hospital Resource Use models on November 12. In the face of surging hospitalizations, they have baffled me by lowering their hospitalization estimates. As of this writing its best case estimate for November 30 is a full 18,000 rooms lower than they were as of November 17. Inevitably we will surpass their base case estimate for the week is out.
In an odd development the Institute's November 30 worst case scenario matches my best case. As I mentioned above, my best case scenario is too high due to not having taken account of the different stages of development of the various states. I don't know the reason for the the lowering of IHME's estimates.