The following report uses only data from the Covid Tracking Project API feeds.
The main reference for Covid 19 metrics is the The Atlantic initiated COVID Tracking Project (CTP), in particular the alarming increase in hospitalizations and resulting ICU usage.
However, recently the CTP data defies anecdotal reference or even common sense. It seems the hospitalization and ICU use data is based on an algo or is formula, likely supposition and is likely not actuality. The main data input for this algo seems to be new cases.
Hospitalization and resulting ICU data is notoriously hard to find in the US. Data is lagged or spotty of sometimes withheld or presented in ways that seem to support specific agendas.
Data scientists and scientists that have been commissioned by key states to define hospital bed usage and capacity given Covid 19 cannot find robust data.
CTP has noted this problem on their website yet they still provide hopsitalization metrics. COVID Tracking Project National Data
Currently hospitalized/Now hospitalized Individuals who are currently hospitalized with COVID-19. Definitions vary by state / territory. Where possible, we report hospitalizations with confirmed or probable COVID-19 cases per the expanded CSTE case definition of April 5th, 2020 approved by the CDC.
From the above hyperlink of the CTP “faq” on the “problem”:
‘Why have you stopped reporting national cumulative hospitalizations, ICU, and ventilation numbers on your website?
Only about two-thirds of states and territories report data for Cumulative hospitalized/Ever hospitalized, and even fewer states report data for Cumulative in ICU/Ever in ICU and Cumulative on ventilator/Ever hospitalized.
Therefore, adding these state and territory figures together to get a national count (as we do for other COVID-19 metrics with complete reporting such as cases and tests) drastically undercounts the true cumulative national number of COVID-19 patients who have ever been hospitalized, admitted to the the ICU, or placed on a ventilator.
This spotty reporting of US hopsitalization is misleading. For example, since more states report the number of people currently in the ICU or on a ventilator than report them cumulatively, the national numbers for individuals currently in the ICU or on a ventilator sometimes exceed the cumulative values.
To avoid confusion, we have therefore decided to stop displaying these national sums of cumulative hospitalization, ICU, and ventilated values on our website, although the fields remain available in our API. We will continue to ask states to report cumulative hospitalization figures and hope to restore the national sums to our website when a critical mass of states report them.’
Despite this, CTP hospitalization data continues to be provided. The data is alarming even to the point of prompting panic, indicating levels that exceed those experienced in NY, CT and NJ in the March to May.
The hospitalization data is obviously smoothed, suggesting it is derived from an algo using confirmed case. New cases are shown then cumulative summed for a rolling 8 days, the typical infectious time or “gamma”.
Then the hospitalized is compared to this with the rolling cumulative cases “boot strapped” to a fit. The fit has high correlation such that it suggests the hospitalization is derived from new cases data and not actually observed.
Covid 19 has well understood severity, the progression from infection to death or recovery - or as epidemiologists call it, “resolved”. The ratio of those infected that become hospitalized is not known as there is no statistically robust test data given the large bias in the application of tests. But there large meta data sets on once being hospitalized the odds entering the ICU and then the odds of death. That is, in a rough off the cuff, 20% odds that if hospitalized ICU admission will result, and that case fatality rate (CFR), will be 2%, depending on the demographics of the admitted.
Using these understood metrics means that deaths, assuming accurate reporting over time, should end with the above ratios, and also show that the understood time to death from infection and time to death once admitted to an ICU is apparent.
This is not the case with CTP hospitalization data, yet there is a very close fit of CTP hopsitalization to reported daily deaths, given which specific period is considerd. The factors that make for this fit is not at all those of the consensus on severity that has developed since February.
That unusual correlation, based upon which time period/phase, is a sign the data for hospitalization not actual but is derived from an algo. Alarming considering how central the CTP hospitalization data has become with the current discussion and has become the basis for the political discussion on public health policy - including lock-downs or mitigation orders.
Below CTP hospitalization data is compared to daily deaths. First the daily deaths CTP provides is smoothed via a 14 day moving average - this removes the “weekend effect”.
Then the CTP data for hopsitalization is boot strapped into fitting the daily deaths. High correlation can be found for three phases: a) the initial surge in March to May b) the surge as Florida, Arizona, Texas commenced and the current surge as Upper Midwest and Mississippi/St Louis river basins increased infections. But to reach these high correlations each phase requires a different time of being infectious (gamma) and a different Case Fatality Rate (CFR) which clearly provide a fit for that specific period. Since this is also the case for hopsitalizations to new cases, perhaps the cases data is also derived from new deaths. The factors required for each of these three phases is provided in the following plots:
“Assuming 14 Days to Death Upon Infection and a Case Fatality Rate of 2.2%” for the current phase.
“Assuming 18 Days to Death Upon Infection and a Case Fatality Rate of 1.8%” for the August phase.
“Assuming 4 Days to Death Upon Infection and a Case Fatality Rate of 3.4%” for the initial March to May phase.
That the three phases fit is so good yet requiring a nonsensical factor change in infectious time (gamma) and CFR implies the CTP hopsitalization data is derived and not actual.