Vaccronyms: FOIA, RKI, DIVI, German ICU data
Matthew 13:30
In Germany, DIVI has been collecting data on Covid patients on ICU (“with” Covid, not necessarily “because of” Covid). Some data are available for download (also on GitHub), but some have been guarded jealously by the Robert-Koch-Institut (RKI). In particular, information on vaccination status of ICU patients was only published once per month, hidden deeply within the RKI reports. See Section 3.5 in this report from 04.05.2023, after which the RKI stopped the reporting altogether.
This intransparent behaviour annoyed me from the very beginning (cf. this post from November 2022), so I wrote an FOIA-type request via FragDenStaat.de, a very useful German platform that completes your request by listing the required paragraphs, and guides you through the whole process. After two and a half months, and two further enquiries, I finally got an answer.
I had asked for daily data on vaccination status of all new admissions to ICU, stratified by age. Unfortunately, age of new admissions is not available (to DIVI, to RKI, and to me). What I got instead are daily data for all individual hospitals in Germany (more than 1,200) but with hospital identifiers carefully replaced by hash values. The reporting started on 13.12.2021 (almost a year after the first vaccination), and my snapshot ends at 14.11.2023. The two csv files (why two? I don’t know) take up 160 megabytes of disk space.
Maybe some useful information can be squeezed out of the hospital-level data, but for the time being I aggregated over all hospitals. I then computed 7-day sums to iron the in-week oscillation (fewer admissions during weekends, for example) out of the picture. This produces time series with a length of 696 days, starting on 19.12.2021 (sum over seven days from 13.12.2021 until 19.12.2021).
Let me show you the resulting diagram first, and then explain the details:
In June 2022, the interface changed. Before that, hospitals had to classify patients into the five categories listed in the top row of the legend, and after that, into the five categories listed in the second row. I omitted one week of data because of my summing over seven days. I mapped colours to indicate categories that should be comparable. The “recovered & unvaccinated” category (rose) was discontinued after June 2022. Something happened to the higher dose echelons at that point. Maybe hospitals had interpreted “fully vaccinated” differently, and with the new interface it became clearer that they were to count doses.
The data from before the interface gap are useless anyway. To demonstrate this, I counted total cases in the data set (black line, right axis) and compared these to official numbers of new admissions (dashed grey line, right axis as well). Only after the gap do the numbers (almost) coincide.
Finally, “prop. with known vacc. status” (dashed black line, left axis) shows the proportion of patients with known vaccination status. Compliance with the request seems to have dropped from the start. Nowadays, vaccination status is reported for only around a quarter of patients (and there is therefore considerable uncertainty in the proportions of the different dose categories).
The overall picture (after the gap) is quite stable. Short-term oscillations mostly occur during periods with very low numbers of patients (during the summer of 2023). The following table compares total post-gap rates to official vaccination rates.
The vaccination rates of ICU Covid patients are higher than the rates in the population but lower than those in the 60+ population. And this is the crux of the matter: it is hard to say anything about vaccine effectiveness because the ICU population is so different from the general population, and because age and vaccination status can not be linked in the data. The best we can do is display the age distribution of the ICU Covid population (not new admissions!):
In order to illustrate the differences, I amended the population structure (as of 31.12.2022) on the right hand side. The 60+ age group makes up around 80% of Covid ICU patients but only 30% of the population. Therefore, a placebo vaccine would result in ICU vaccination rates closer to the 60+ vaccination rates than to the population vaccination rates. Without further information, it is hard to judge the Covid vaccines (but let’s be clear: anything better than very low 2-digit effectiveness is impossible, and whatever effectiveness remains is partly a selection effect).
However, what is really striking is the proportion of patients with exactly one dose: this proportion is much larger than the official rates suggest. But the official rates have become pretty useless anyway. Even one year ago, I demonstrated that the true rates might be up to 4% higher or lower than the official ones, and now that another year has passed, by the same reasoning the interval is more like +/- 8%.
A government agency (RKI) that starts to think about collecting data one year after the first vaccination. An interface that has to be replaced after six months. Hospitals that fill in the daily forms technically but do not care about the data. A project that surely wasted millions of Euros and hundreds of thousands of person-days. Data that are of limited value because patients’ age was never included in the interface. But even if it had been, population level data on vaccination only distinguish between very broad age categories (18-59, 60+). That’s Germany.
On the other hand, we may have profited from our stupidity and sluggishness. Our lockdowns were never as severe as in, for example, the Southern European countries. But then, population-wide vaccine mandates were on the table until the Bundestag voted against them in April of 2022. Mandates for the medical profession were in place until 31.12.2022, and are still in place for Bundeswehr soldiers (“Duldungspflicht”). We just flattened the totalitarian curve.
If you are interested in the data, or would like to study high-resolution versions of the diagrams, please feel free to download the above Excel file.
Nice work, CM!
Another angle is the revealed preference of the German public in your table
Vax status : unvax ->1+ doses ->2+ doses -> 3+ doses -> 4+ doses
Rates on ICU: 13.5% -> 78.0% -> 76.4% -> 62.7% -> 16.2%
Vax rate 60+ : 8.9% -> 91.1% -> 90.1% -> 85.5% -> 41.4%
Strangely, despite vax-passport restrictions (so-called 2G) in winter 21/22, uptake drops significantly between 2 doses and 3 (why, if so effective and safe?) BUT effectiveness apparently increases - huh?
Then, after vax-passport restrictions are lifted in spring '22, uptake slumps between 3 doses and 4 (why, if so effective and safe?) AND effectiveness increases even further!
For example, in 60+ population this screams (to me):
91.1% -> 90.1% = "serious adverse events"
90.1% -> 85.5% = "buyers remorse" + "serious adverse events"
85.5% -> 41.4% = "buyers remorse" + "removal of government enforcement" + "healthy user bias"
Arguably the discrepancy between 60+ unvaxxed rate and ICU rate points to vax efficacy but is hopelessly confounded by the unknown make up of unvaxxed population (healthy?/unhealthy?).
"Therefore, a placebo vaccine would result in ICU vaccination rates closer to the 60+ population rates than to the general population rates."
Not sure about this, the profile of Covid ICU patients is predominantly older regardless of any vax status, so wouldn't placebo vax result in ICU rates closer to pre-vax rates?