Using detailed data from Florida, we try to explain why the recent acceleration in cases has seemingly not generated any notable spike in hospitalizations in the state. We find that changing age distribution of cases, towards younger cohorts explain about half of the declining hospitalization ratio. But the propensity to need hospitalization is also falling within age groups. It seems like several states, such as Florida and California is avoiding dramatic hospital pressure due to these effects, perhaps similar to to what we have seen in Sweden (cases remain elevated but hospital pressure is moderating). Looking at all US states, and the hospital data that is available, we find that the large majority of states are seeing substantial declines in hospitalization rates, while Texas is the main outlier (case growth has translated almost 1:1 into growth in hospitalizations). We are fairly confident that case growth will remain elevated in coming weeks. But the implications for markets vary dramatically based on whether this is translating into increased hospital pressure or not. The evidence is mixed, but for the majority of states the pressure is moderate, which is different from the first wave.
We have been focussed on accelerating case growth in a number of US states over the past 2-3 weeks. But the second wave is not like the first one. While some states have seen increasing hospital pressure, most have seen dramatic divergence between an accelerating number of cases and fairly benign trends in hospitalizations.
In the appendix, we show hospitalization rates (hospitalizations relative to cases) for all US states, based on two different metrics (subject to data limitations). It is clear that hospitalization rates are dropping sharply in the large majority of states.
Below, we go into more detail with data from Florida. Specifically, we dissect a detailed database of all 90K cases in Florida so far. We find that the observed drop in hospitalization rates can be partly explained by age effects, with younger people making up a much larger share of new infections. The remaining portion (about half of the aggregate drop in hospitalization) is explained by within-group changes in hospitalization rates. We attribute these widespread decreases in hospitalization to changes in behavior and testing. The key conclusion is that as case growth surges in the context of a second-wave, hospital pressure will be slower to react.
The Case of Florida: Dynamics of hospitalization rates
Florida has seen dramatic case growth over the past 3-4 weeks. From a daily average of 760 to a daily average of 3103 using 7-day moving average.
At the same time, the daily hospitalizations have been mostly stable.
In other words, the hospitalization rate among recorded positives has dropped substantially.
We can use the detailed information about cases in Florida to explain why the hospitalization rate has dropped from 25% in March to 8% in June so far (we use data for H1 June to avoid undercounting hospitalizations).
There are fundamentally two effects in play
A) Composition effect: The age distribution of cases in Florida is getting younger and younger, and this is impacting the hospitalization ratios, as the younger cohort is much less likely to need hospitalization.
The chart below shows the those <34 years accounted for less than 25% of recorded cases in March, but near 50% currently.
As illustrated in the chart, the change in the age distribution accounts for about half (47% to be specific) of the drop in the hospitalization rate in June.
For comparison, the composition effect accounts for only 30% in May (using hospitalization ratios from April, we can explain 30% of the drop in the hospitalization ratio with the changing age distribution of recorded cases).
B) Within-group effects account for the part of the decline that cannot be attributed to composition
As the chart below shows, hospitalization rates have dropped notably for all age groups, with the hospitalizations rate cut in half for the oldest cohort, for example.
Some possible explanations for this widespread decline include:
Testing – in March, many local authorities were still in ‘catch up’ mode, where testing was largely happening at hospitals. In that context, the hospitalization rate will certainly be inflated and registered cases will be overwhelmingly skewed towards the most severe patients.
Behavior – earlier infections were results of completely unmodified (relaxed) behavior. People could have been in extremely close contact with sick people for extended periods of time (getting high virus loads). This means that exposure levels could have been far more elevated than a newly infected person today. Moreover, the most vulnerable (ie the oldest) are likely to be the most careful now, creating a different ‘selection bias’.
Appendix: Detailed data on hospitalization ratios for all US states
The charts below show a metric of the ‘hospitalization ratios’ for the states that report cumulative hospitalizations. We focus on the change in this metric (to capture daily hospitalizations) and we look two-week trend in this metric, relative to a two-week trend in new cases, 10-days lagged (the average lag from a test positive to hospitalizations).
The charts below shows an alternative metric of the ‘hospitalization ratio’ based on active hospitalizations relative to active cases. This metric will be influenced by how long patients stay in hospital (as opposed to just how many enter the hospital). The trend is clearly down in almost all states, with key exceptions: Texas, South Dakota, North Dakota. Among the larger states, Texas is the clear outlier.