Fact check: does anti-SARS-CoV-2 vaccination work?

 COVID-19 vaccines do not work against Delta
Our hospitals are full with vaccinated patients
I'd rather wait for an inactivated vaccine
Two third are vaccinated and there's a lockdown again...
I've got my jab and got CoV afterwards, what a junk (des isch a Klumpert)!
I wear a mask, keep my distance, go testing, why I should let me vaccinate with this poison?!
Heard on the street, playgrounds in Tyrol and seen in social media in Austria these days.  
Daily incidences breach every record, breakthrough infections rate climbs toward 50%, the national health care is approaching a collapse, the surprised government seems to have no idea... Hey, but two-third Austrians are fully vaccinated, what went wrong? The politicians claimed: the pandemic is over, CoV is gonna be an individual medical problem, it is this a pandemic of unvaccinated.
This motivates me to write this post. Does the anti-SARS-CoV-2 vaccination works against the Delta virus paralyzing our hospitals? Is it 65% vaccinated population enough to keep the fourth COVID-19 outbreak at bay? Does it protect me from COVID-19? I'm gonna answer those irritating questions in light of the available data.

Technically, my approach is simple. I've downloaded the complete SARS-CoV-2 data set from Our World in Data containing the daily incidences, fatalities, cumulative case numbers and vaccination rate in the entire population for all European countries. Next, I modeled the daily incidence and mortality in 2020 - this is my training data set - with an AI Random Forest algorithm (frankly speaking I've tinkered with SVM and shallow neural networks as well). Finally, I predicted the incidences and mortality in 2021 - this was my test data set. The R pipeline used for the analysis is available from my GitHub.
If it sounds Klingonese to you: I let my machine learn about the course of pandemic in 2020 and apply this knowledge to 2021. The set of parameters driving such learning process is very limited: just incidences during the past 14 days (it's an infectious disease...), total number of cases since beginning of the pandemic (natural immunity) and day of the year (seasonal effect). Importantly, my algorithm is blind to the vaccination rate! The main limitation was the simplicity of my approach. To get a even better prognosis you should include direct effects of lockdowns, school closings, mask wearing, test rate (the more you test the more you see..) and so on. Hence my estimations are a bit overoptimistic, but as you'll see for Austria, by roughly 35%.
The purpose: I wanted to see what went better and what failed in the second pandemic year. Notably, in 2020 we had no mass vaccination and way over 50% of the Europeans fully vaccinated. Hence, by comparing the expected case and fatality numbers and the true figures for 2021 I hoped to see if the vaccine may play a role there. I was really curious about the results!

Finding 1: 2021 pandemic is not so dramatic as it is expected to be

My first impression for almost all European countries: the algorithm seems to do the job and predicts correctly the cases and moralities in the 'training' 2020 year (Figure 1 and 2, left-side plots). By the way: such mode of backwards-checking of the model is called 'internal validation' and is an essential step when predicting weather, stock price and so on. A convenient, and one of many,  measures for this process is so-called 'mean absolute error' (MAE). It gives you an information, how much your model diverts from the reality, in my case per day. For Austria (Figure 1), I'm missing the true cases per million by MAE = 10 per day and deaths per million by MAE = 0.29 per day.

Figure 1. Predictions of daily SARS-CoV-2 cases (A) and fatalities (B) in Austria in the training 2020 pandemic set for the test 2021 data. Bars: real outcome, lines: mean 7 day incidences, red: true outcome, blue: prediction by the algorithm. Mean absolute error (MAE) and cumulative error (CumE) presented in the plot headings.

What is really exciting: predictions by the algorithm for 2021. As you may see for Austria, the algorithm predicts the 2021 cases with high precision till autumn, i.e. the dramatic outbreak we see right now: the number of new cases those days is almost twice as high as predicted. In sum there are 2800 cases per million more in the reality than expected for 2021 till now (see: cumulative error, CumE), which is a clear effect of the Delta pathogen. Given the vaccination rate in Austria, one can conclude that somehow 65% immunized population could not brake the spread of the virus (Figure 1A). By contrast, there's a dramatic discrepancy between the reality and the prediction for mortality in Austria in 2021 (Figure 1B): it's generally far lower than a year ago - there are estimated 3.6 COVID-19 death per million and day (MAE) less than expected or 1100 fatalities (CumE) less per million till today! For 9 million inhabitants this makes an impressive number of 9900 human lives saved in total. This is a bit more than calculated by Austria's Health Authority AGES (6259 fatalities) for the lives saved by the vaccination alone, but my model is extremely simple and blinded to lockdowns, incidences in risk groups and so on. Hence, it seems plausible that approx. 2/3 of the deaths rescued in 2021 are due to vaccination in Austria. In other words: the anti-SARS-CoV2 vaccination saves thousands of human lives in the alpine republic but does not necessarily protects from infection, at least in the whole-population data analyzed here.

Figure 2. Predictions of daily SARS-CoV-2 cases (A) and fatalities (B) in Poland in the training 2020 pandemic set for the test 2021 data. Bars: real outcome, lines: mean 7 day incidences, red: true outcome, blue: prediction by the algorithm. Mean absolute error (MAE) and cumulative error (CumE) presented in the plot headings.

As a data freak, I was curious if the same pattern - again: a measurable amount of human lives saved despite much more infections - is identifiable in other European countries. In fact, by far not for all - see: Poland as an example (Figure 2), where lower numbers of infection than predicted have been registered but, unfortunately, there was somehow 250 CoV deaths more per million inhabitants than expected for 2021 (or 9600 more in total). In this particular case, the 'excess' mortality was concentrated in the first 4 months of the year, when the impact of the mass vaccination campaign was quite unlikely. Alternatively, the spring 2021 mortality in Poland may have resulted from a failure to reach vulnerable groups with the immunization (Figure 2B), which obviously worked much better in Austria (Figure 1B).

A pan-European statistic is shown in Figure 3: you may appreciate that the majority of the 45 analyzed countries succeeded to cut down the mortality by quite a significant numbers and in many cases also the incidence rates.

Figure 3. Numbers of SARS-CoV-2 infections (A) and deaths (B) rescued in 2021 as compared with the machine learning prediction.
The case becomes a bit more intricate when we restrict out modeling to the second half-year 2021, i.e. the outbreaks caused by the highly infectious Delta variant of SARS-CoV-2 (Figure 4).

Figure 4.
Numbers of SARS-CoV-2 infections (A) and deaths (B) rescued in the second half-year 2021 as compared with the machine learning prediction.

Conspicuously, many more countries have to fight with higher incidences than predicted (Figure 4A), however only few - you know them from recent press releases like Russia or Romania - have higher mortality as it could be expected for the incidence, reproduction number and time of the year (Figure 4B).
You may say: hey, there's something wrong with your model. Is Austria really an example of the best-practice at pandemic management, on par with UK, Italy and Spain. Right now? On the eve of a hard lockdown and with the collapsing hospital health care? Obviously yes: look at the blue lines in Figure 1B. The catastrophe we're experiencing now had been expected two moths before at the onset of autumn. It was obviously hold off by a mysterious factor - vaccination?

Finding 2: the number of fatalities rescued during the Delta outbreak correlates with the vaccination rate

The vaccination rates in Europe as of 19th November follow more or less a west-east axis (Figure 5A). They are the lowest in the former soviet republics, Balkan countries and peak in Spain, Portugal, Iceland, Italy, Ireland, Benelux region, Denmark and Finland. Obviously, such high vaccination rate is not a guarantee of low daily incidences in the second half-year 2021 (Figure 5B): there are lots of SARS-CoV-2 cases prevented in Italy or Portugal but more than expected in the similarly well immunized Icelandic, Irish and Finnish populations.

Figure 5. Vaccination rate (A), numbers of SARS-CoV-2 cases (B) and fatalities (C) rescued in the second half-year 2021.
In turn, the association of vaccination rates and COVID-19 deaths is much clearer (Figure 5B). East-European countries with low vaccination rates have also more deaths than expected or are comparable with the expectations. Of course there are 'outliers' like Slovenia (SVN) or Croatia (HRV) with low percent of fully vaccinated inhabitants and lots of lives saved (Figure 6). In those particular cases the model predictions may result from training on extremely high mortality rates in 2020.

Figure 6. Relationship of the rate of fully vaccinated with the number of SARS-CoV-2 cases (A) and COVID-19 deaths (B) rescued. Blue lines: LOESS trend.

Interestingly, the vaccination - deaths saved relationship follows a S-like curve (see: the blue trend line in Figure 6B). It means that the positive effects of the SARS-CoV-2 vaccination on the overall CoV mortality during the Delta outbreaks are seen quite instantly, at approximately 50% of the population fully vaccinated. Interestingly, no such relationship exists for the infection cases rescued (Figure 6A). 
To me, this suggests that high vaccination rates could not break circulation of the Delta variant in the population. But what they can, especially in the vulnerably, elderly and multi-morbid subsets, to drastically cut down the deaths.

Finding 3: the vaccine protects robustly from CoV pneumonia but not from infection with Delta SARS-coV-2

Well, this is maybe a valuable information for the health authority - in fact they get better simulations from their simulation scientists and I'm not gonna get a call from the Complexity Science Hub Vienna with a job offer after this post. But what this means to the European societies struggling against the Delta outbreak or awaiting the same situation as in Austria in the next weeks? What does it mean to single citizens? Does the vaccination makes a sense? A boost?

My modeling results suggest that we're not going to push down the daily incidences by simply immunizing the whole population with two doses of the vaccine. At the population level, the immunization by vaccination or by infection has a minimal effect on the virus spread. Even if 'fully vaccinated', you're likely to get a infected this winter and spring. The good part of news is, assuming you got at least two jabs: you infection is likely to be a common upper respiratory tract stuff like many at this time of the year. Your chances of getting a COVID-19 pneumonia drop down dramatically - by eyeballing of my data by 10 - 20 fold as compared with an unvaccinated or incompletely vaccinated individual. Or even much, much more if you belong to a 'risk group' suffering from cardiovascular disease, lung or kidney conditions, diabetes. Of course, it would be way too far-fetched to present hard numbers based on my simple mode here - I'm thinking on crunching the Austrian data in a post one day to give you more details.

Extrapolated to the whole population: a high rate of full vaccination is hence also the best way to protect our national health system - one of the most egalitarian and robust in the world and highly valued by the citizens. Personally, I think the colloquial discussion on how much vaccinated populate the country's ICUs or COVID-19 stations is simply silly. You can expect 100% breakthrough infections and 100% vaccinated COVID-19 fatalities if you vaccinate every inhabitant of the country. Especially irritating, if people without any knowledge on mathematics, statistics or biology play experts there. The fact is: with the high vaccination rate from kids to elderly, we are unlikely to eradicate the pathogen, but we're going to relieve the yearly misery of the health care system, guarantee integrity of medical support to cancer, trauma and cardio patients. The life-saving potential of the anti-SARS-CoV-2 vaccine reaches far, far beyond COVID-19. For this reason, I support the newly announced compulsory vaccination against SARS-CoV-2 in Austria. Sorry, there is not other way to make the pathogen a (mostly) tame bug and let us enjoy our freedom.

Figure 7. SARS-CoV-2 cases (A) and COVID-19 deaths rescued in Austria per day.

The booster issue is the most tricky one. Apart from just few special regions like Israel there's no enough data to evaluate the booster's efficacy at the population level. In Austria, the majority of the vaccinated got the second jab June till mid July (see: OWID). 4 - 5 months thereafter, we are experiencing a dramatic drop of the infection-rescuing effect (Figure 7A): a strong hint, this is the right time point to boost if you hope to combat the virus spread that way. However, the general life-saving effect is still there (Figure 7B) - it has to be stressed that this may not apply to risk groups! Data obtained in laboratories and longitudinal studies of COVID-19 convalescents (like Huang and colleagues) point towards strongly declining ability to kill the bug by serum antibodies roughly 6 months past infection. With this in mind, I'm gonna get the third dose as soon as possible and can recommend the same to everybody.

Perspective: what we can do better

What is also obvious from my limited and by far incomplete model: full vaccination rates of 65% like in Austria nowadays save many lives but are too less to prevent rampage of the health care system. What went wrong? this is my personal diagnosis:
  • No vax movement is politically sexy. We have a growing and well organized community of vaccination critics and denialists with an overlap with the esoteric, extreme right and populist scene, which in turn has a potent political representation in the Federal bodies and Parliament. For this reason, a large subset of the unvaccinated group is not prone to any rational argumentation, campaigns and won't change their minds facing the dramatic situation in Austrian hospitals. The only argumentation they may follow comes from above - one should consider including the 'open-minded' parts of their political leadership in the pandemic management.
  • Vaccination as personal choice. This is still a widespread opinion in the Austrian society. Vaccination is strongly associated with infant conditions and protection of one's offspring or, like in case of the highly popular anti-TBE (German: FSME) vaccination, with protection of oneself. Hence, individuals who perceive themselves as a risk group have a strong interest at the SARS-CoV-2 vaccination. People do not consider vaccination as a way to protect their society, economics or health care system. It has to be stressed: it is the only way to get our society out of the vicious circle of the pandemic
  • Mass testing. A year ago, the former Austria's chancellor came up with the idea to test the whole population for SARS-CoV-2 and eradicate the disease this smart way. You may read in my blog, there's no chance to fight off the pandemic this desperate way. Nevertheless, Austria became a 'test republic', with millions of free antigen and PCR assays made every day - even now, when you may get a free CoV jab in a supermarket, church or rock concert. The dubious idea called '3G' (Getestet, Geimpft, Genesen, in English: tested, vaccinated or recovered) put the momentarily negative test results on the par with a sustained anti-SARS-CoV-2 immunity. The colloquial opinion of many shaped by numerous campaigns and advertisements is that testing protects you from the disease. Or even worse, it's better than vaccination, cause you're sure you are not infected and an element of an etiquette preceding personal and business meetings. The effect is, we conduct roughly 2 million PCR tests daily and could not relieve our heavy burdened hospitals. The politics has to stop this expensive hogwash and promote the vaccination instead!
But last at not least, the data presented herein have a positive side. They show clearly, that two thirds of us made the right decision and took an important personal effort to save thousands of lives and bring us closer to the end of the pandemic: they got two tiny jabs in their shoulder. I hope, more will follow!


Popular posts from this blog

On inflammation, machine learning and lung recovery after COVID-19

Fact check: is Omicron less dangerous than Delta?