Corona Sweden – Excess deaths up until May 29th

Today, the government bureau of statistics, SCB, provided their weekly update on total deaths, that is, all deaths, not just Corona-related ones,  in Sweden, Jan 1st to May 29th.

SCB also provides data from 5 previous years, so that we can, by comparing the numbers for 2020 – the Corona year – with e.g. the average for 2015-2019, calculate the “Excess Deaths”, that is, how many more people have died this year compared to average.

Let’s start by looking at a day-by-day view of Excess Deaths:

SCB_excess_daily_2020-05-29

As can be seen from the graph above, the daily deaths started growing late March, to reach a peak in mid-April, with a peak of about 150 excess daily deaths, to be compared with an average daily death rate of about 260. However, from late April, the daily excess deaths have been going down, slowly but surely.

Also note that the last 4-5 days of the data are on the negative side – based on the numbers from previous weeks, we know that these negative numbers are going to be updated upwards later, so in the next set of graphs, I’ve removed the last few days, to obtain stable data.

The daily numbers are very noisy, so let’s look at a weekly aggregate, up including the week ending May 24th:

SCB_excess_weekly_2020-05-29

In the graph above we can clearly see that the number of excess deaths is going steadily down since mid- to late April, the weekly number of excess deaths was, for the week of May 24th, under 200.

Another thing to notice is that the year 2020 started with a “deficit” of deaths – up until mid-March, 2020 had less deaths than during the baseline years 2015-2019.

It was just the other day that I learned from one of Professor Michael Levitt’s presentations that for most of Europe, the winter season 2019/2020 had very few deaths caused by the normal seasonal flu, so I decided to look closer at the Swedish numbers, to see if I could se such a pattern. More about this in a moment.

But first, let’s look at the cumulative number of excess deaths for 2020, year-to-date:

SCB_excess_weekly_cumsum_2020-05-29

Above we can see that Sweden by mid-March had a “deficit” of about 1300 deaths for the year, which seems like a plausible evidence of that this season’s winter flu was very mild, with very few casulties.  Then, by late March, the number of deaths accelerated quickly, at the same time as Corona started accelerating.

At this point, that is, up until week of May 24th, Sweden has an excess of 3027 deaths, year-to-date.  But since the winter flu season starts before Jan 1st, let’s change the start of the period we are looking at, from Jan 1st 2020, to May 2019, so that we get a full year from May 2019 until now:

SCB_season_19_20

The graph above shows daily deaths from May 2019 to 21st of May 2020, in red, and the baseline average in orange.  Despite the data being noisy, we can see that during the winter months, this season’s deaths were lower compared to the baseline. Let’s look at the same data, aggregated weekly:

SCB_season_19_20_weekly

In the graph above we can see that the number of deaths this winter season was lower than during the baseline, with the largest difference during January to late March. That is, we had an accumulated death “deficit” by the time Corona came around.  Thus, by April, there were many more vulnerable people, people within the risk group, around, exactly when Corona started to spread in the society, not least among the elderly in the nursing homes.

So, by taking this  full “seasonal flu deficit” into account, the number of excess deaths goes from 3027 to 2276, which corresponds to 2% yearly increase thus far. That’s about twice the annual random  variability, so there’s definitely an amount of excess deaths present thus far.

But it will be interesting to see by end of year, how much of excess death Corona-2020 in fact brought – it’s fully possible that by Dec 31st the number of excess deaths for year 2020 will be within random variability, that is, ~1000 from baseline mean, which inevitably will lead to some very interesting questions concerning the proportionality of the closedown of the society…. 

Thus, the interesting question becomes: were the measures taken to fight Corona, such as Lockdown’s, closure of schools/businesses etc, postponement of otherwise critical non-Corona related medical interventions, with all their negative effects on societies, effects that most likely will last for a very long time, and impact many more people than Corona itself, really in parity with the severity of the Corona virus…?  Now, Sweden, as we all know, imposed in comparison with other countries, very limited restrictions on society, but regardless, the impact of Corona will be felt here for years to come. And for those countries that did close down hermetically, the consequences will most likely be far more severe, in terms of unemployment etc.  Thus, to paraphrase a catchy marketing slogan:

Was it really worth it…? 

[UPDATE: an interesting post on Corona’s impact on economy]

Posted in Data Analytics, Epidemics | Tagged , | Leave a comment

Lies, Damn Lies & “Scientific” Corona Models

For 15 years ago, Stanford Professor John Ioannidis published a paper “Why Most Published Research Findings are False”.

Well, things seem not to have improved during those 15 years, at least not when it comes to models predicting the evolution of the Corona virus:

  • The now infamous model from Imperial College London, headed by Professor Neil Ferguson, predicted that Sweden by early May would have 40.000 deaths caused by corona. As of today, June 1, we have 4403.
  • Mid-April, a team of researchers from Belgium presented a model predicting that Sweden, by end of May would have 100.000 deaths.

Lately, I’ve been following Nobel Laureate Professor Michael Levitt (@MLevitt_NP2013), who’s extremely critical of the majority of papers published about Corona, since it appears that researchers publishing rubbish like the two examples above get away with it, very few are held to any responsibility for what they are publishing. And in some cases, e.g. the now infamous paper from Imperial College London, has (and will for a foreseable time!) resulted in vast negative consequences for many countries/societies, not only UK, but many other countries, since it was taken as gospel, with resulting lockdowns.

Basically, it appears to me that decisions about Corona have been largely taken by fear and populism, not real science. Real science means that you look at the data, apply critical analytic thinking, and if your model doesnt concur with the data, there’s something wrong with your model (or the author of that model, i.e you).

You can listen to Professor Levitt in the interview below:

In the next post, I’ll look into fresh data published today on Swedish deaths Jan 1st to May29th, from the Swedish Government Bureau of Statistics, SCB, and what we will find there provides very strong support for that Professor Levitt is absolutely right about Corona.

Posted in Data Analytics, Epidemics, Research | Tagged , , | Leave a comment

Bayes Rule for Poets and other Mathematical Virgins* :-)

[ * The title alludes to a book titled “Higher Mathematics for Poets & other Mathematical Virgins” by Tönis Tönisson, a book that for many many moons ago got me to appreciate mathematics in depth]

If you are anything like me, you’ve probably seen, maybe even been using, but simultaneously also been slightly confused over Bayes’ Rule, i.e.

P(A|B) = P(B|A) * P(A) / P(B)

How does it actually work, and why does it work…?

Then, still if you are like me, you might have Google’d it, and found out that there’s a way to understand the rule by using  graphics and absolute numbers instead of probabilities (example below).

But until recently, I hadn’t found any good explanation that clearly connects the algebra with the graphic/absolute illustration.

But for  a while ago, I did find a brilliant illustration @ 3Blue1Brown on Youtube :

This guy does a great job in not only this example on Bayes Rule, but on lot of things about math and science.

In this post, I’ll elaborate a bit on this example, in context of testing, e.g. testing for a medical condition, such as antibodies for Covid-19.

But Bayes Rule is also relevant for many many other domains, such as airport metal detectors (those annoying things forcing you to strip everything out of your pockets), ignition alcolocks and Search & Rescue Operations.  To mention just a few. And of course, it’s the foundation for Bayesian Inference, which you can find a lot about on this blog.

First, let’s define a bit of terminology:

  • Incidence or Base Rate : the ratio of population that has the condition, e.g. the a decease, that we are testing for
  • Sensitivity : the ability of the test to correctly identify the existence of the condition in a subject (e.g. person under test) that has the condition
  • Specificity : the ability of the test to correctly identify absence of the condition in a subject that does not have the condition

So let’s do an example:

Let’s say we have a population of 100 people, and furthermore let’s say that the incidence or base rate of some condition – say infected by Corona virus – in the overall population is 10%, i.e 0.1. Let’s furthermore assume that we have a test procedure, e.g. anti-body testing, that has a Sensitivity of 0.90, meaning that the test is able to correctly identify 9 out of 10 of people infected as infected (‘True Positive’), and a Specificity of 0.80, that is, the test is able to correctly identify 8/10 of non-infected people as not infected.

(Now, I have no idea what the real Corona anti-body testing Sensitivity and Specificity numbers are, I just made the above numbers up, for ease of illustration. )

So now we might want to know two things:

  • Given that you’ve tested positive for Corona, what’s the probability that you actually have Corona ?
  • Given that you’ve tested negative for Corona, what’s the probability that you actually don’t have Corona ?

Turns out that most people, including students of medicine, when given this type of problem, get the answers wrong, terribly wrong : many seem to “anchor” on the impressive sounding 90% and 80% rates for Sensitivity and Specificity, and therefore “guess” that the answer must be around those numbers.  Which it’s not, particularily if the Incidence/Base Rate is very low (which it is not in my example, for simplicity).

In order to answer the two questions, we can use Bayes Rule from above. Let’s replace the ‘A’ in the formula with Infected, and ‘B’ with Evidence, that is, the outcome of the test. Thus, Bayes Rule now reads:

P(Infected | Evidence) = P(Evidence | Infected) * P(Infected) / P(Evidence)

In English: “Probability of being infected given positive test is equal to Probability of positive test given you are infected * Probability being infected] / Probability of a positive test”

We immediately know two of the terms of the equation:

  • P(Infected) is our Incidence or Base Rate, that is, 0.10 in this example
  • P(Evidence | Infected) is our Sensitivity, 0.90

However, we do not immediately know P(Evidence), and of course, we do not know P(Infected|Evidence), which of course is what we ultimately want to know.

How can we figure out P(Evidence)…?

Well, ‘Evidence’ in our example is the number of positive test results. But the problem is that positive test results are of two kinds, True Positive and False Positive.

The True Positives are easy: take the proportion of population that is infected, that is, Population * Indicence, which give you the subset of all infected, and multiply that number with Sensitivity, which we know, and which equals P(Evidence | Infected).  So in our example we get 100 * 0.10 * 0.90, which gives us 9 True Positives.

What about the False Positives, how can we get them…?

In a similar way: take the proportion of the population that is *not* infected, and multiply by 1 – Specificity:

False Positives = Population * (1 – Incidence) * (1 – Specificity) ; which gives us 18 False Positives.

In total thus, we have 9 + 18 == 27 people that will test positive, but only 9 of these 27 actually have the infection.  So, the probability for you being infected, after having received a positive test result, is 9/27 or 1/3 or 33%.  Which is the answer to our first question.

Thus, the thing to notice here is that what Bayes Rule really expresses, is the ratio of True Positives to (True Positives + False Positives), that is:

P(Infected|Evidence) = True Positives / (True Positives + False Positives)

This can be easily seen by drawing up a 10 x 10 grid, representing our population of 100, and splitting the grid so that 90% (those not infected) of the population is above the split, and 10% (those infected) are below:

sensitivity_specificity

So, in the grid above, the bottom 10 cells represent the 10% infected of the population. Of these 10, the test will correctly identify 9 (since the test Sensitivity is 90%), leaving one person as a False Negative, or Type-II error.

The 18 ‘O’s on the top represent the False Positives (Type-I error), occuring since the Specificity of the test is 80%, which means that out of the 90 non-infected, the test correctly identifies 72 as not carrying the infection, but 18 as False Positives.

So, what I’ve found is the easiest way to understand Bayes Rule is as follows (in the context of the example) : thinking of the ‘P’s not as probabilities, but as Proportions. Lets break down the different terms of the equation into proportions:

  • population * Incidence  : gives you the proportion infected (the 10 cells at the bottom of the grid)
  • population * Incidence * P(Evidence | Infected) : gives the proportion of infected correctly identified by the test as infected ( the nine ‘X’s, the True Positive)
  • population * (1 – Incidence) : gives you the proportion of population not infected
  • population * (1 – Incidence) * (1 – Specificity) : gives you the proportion of the not infected falsely identified by the test as positive (18 False Positive)

So we can write our Baysian Rule as:

P(Infected | Evidence)  =

population * Incidence * P(Evidence | Infected)

divided by:

population * Incidence * P(Evidence | Infected) + population * (1 – Incidence) *                                                        P(Evidence | ! Infected)

where P(Evidence | ! Infected) is equal to 1 – Specificity

In Python:

# using Bayes Rule

# probability of having decease given that test tells so

# p(inf|ev) = p(inf) * p(ev|inf) / ( p(inf) * p(ev|inf) + p(~inf) * p(ev|~infected) ) 
# that is: true positives / (true positives + false positives)

p_inf = incidence
p_not_inf = 1 - p_inf
p_ev_given_inf = sensitivity 
p_ev_given_not_inf = 1 - specificity

p_inf_given_ev = (p_inf * p_ev_given_inf) / (p_inf * p_ev_given_inf + p_not_inf * p_ev_given_not_inf)

I’ll leave answering question number 2 above as an exercise for the reader.

Posted in Bayes, Epidemics, Math, Probability | Tagged , , , | Leave a comment

Corona Sweden – Expected vs Actual Deaths – an anomaly

Sweden is different : not only in “strategy”, but also with respect to the pattern of weekly deaths:

while all other countries that have reached the inflection point (at least of the handful I monitor closely) show – after having hit the inflection point –  monotonical decrease of the number of of deaths, Sweden does not. Let’s look at e.g Italy:

Italy_violinplot_weekly

When Italy reached the point where deaths started going downwards, the trend has continued monotonically downwards.  The same pattern is true for Spain, US and all other countries I’ve looked at.

Sweden, on the other hand, has a different pattern, which can be described by “down-up-down-down-up” – that is, the trend is not consistently downwards.  Let’s look at the numbers:

Sweden_violinplot_weekly

As can be seen below, the weekly number of confirmed seems to have stalled on a plateau – the number is not growning, but nor is it going significantly down.

For the number of deaths, there is a similar pattern – after the peak in late April, the weekly number of deaths have gone down only slowly, and not monotonically, as for the other countries.  Now, a possible explanation for the increase the past week might be that the week before was a 4 day public holiday, so maybe the reporting lags more than usual. But even so, the downward trend is much slower than for the other countries, let’s again compare with Italy:

 

weekly_increments_Sweden

Comparing Italy’s graph below to Sweden’s above clearly demonstrate the huge difference in the patterns.

weekly_increments_Italy

So there is indeed cause for concern. However, before anyone jumps into any conclusions about the “strategy” having failed miserably, I strongly recommend reading this post : the “strategy”, that everybody seem to be so concerned with, is, IMO, just a tiny fraction of the explanation on why Sweden has failed in it’s fight against Corona : there are confounding factors way more important than the “strategy” itself, factors that are caused by political decisions over the past 2-3 decades, and only now, by Corona, are brought clearly into light.

Posted in Bayes, Data Analytics, Epidemics | Tagged , , | Leave a comment

Understanding Exponential Growth

 

Nobel Laureate Professor Michael Levitt explains exponential growth in context of the Corona pandemic.

 

Posted in Math | Tagged , | Leave a comment

From Spectator : Norway health chief: lockdown was not needed to tame Covid

Lockdown unnecessary…? 

Posted in Epidemics | Tagged | Leave a comment

Corona True Mortality Rate & percentage infected

It’s been a while ago since I ran my Markov Chain Monte Carlo simulation on true mortality rate, so I did so last night. This time only accounting for the data from the past 30 days, reason being a hypothesis of mine that now, at least in those countries that I monitor, the numbers have stabilized to some kind of “steady state”, that is, we are well past the inflection point.

Summary of Inference Results:

True Mortality rate for the countries I monitor (US,Sweden,Germany,Spain,Italy) ends up remarkably consistently around 0.3-0.8%. Of course, there’s a lot of uncertainty (which you can observe from the Long Tails in the graphs below) but 50% HPDI resides in this interval, which you can read as there being 50% chance that true mortality falls in this range.  If this is indeed the case, then what we are dealing with, Corona, has a mortality on par with a severe seasonal flu.

Ratio infected of population varies between 1.3 to 15 (!) percent, with Germany in the lower end, and Spain at top. Typical ratio infected/population resides around 4-5%.

That is, we are still far from Herd Immunity, except perhaps Spain and Italy.

if these number are anywhere even close to being accurate – which I believe they are, since almost all anti-body tests results presented over the past few weeks end up in the same ball park – then it’s going to be interesting to see how long this pandemic is going to lockdown the world.  Several years…?

For those countries that have opted on strict lockdowns, with no vaccine in sight, and being far from herd immunity, it’s going to be very hard to decide to return back to anything resembling normal life, because the risk of a backlash is huge.

Of course, as always when dealing with averages – and all the reported numbers on Corona are averages, nation by nation – it’s very likely that there’s a huge amount of variability within different countries/regions/cities etc.

 

As an end note, this interview with a former manager of population statistics at SCB, the Swedish Government Agency for Statistics, concur with my  predictions on Corona, e.g. that the total Corona death toll, i.e the total number of people dead where the primary cause for the death was Corona, will not exceed 10.000.

Posted in Bayes, Data Analytics, Epidemics | Tagged , , | Leave a comment

The Swedish Corona Strategy Explained

Once in a while you can find something worthwhile in social media. Just the other day, out of the blue, came a post on Facebook flying by, a post that in a few paragraphs managed to explain, in a very clear way, the for so many so seemingly weird Swedish strategy for fighting Corona.

What I’ve been trying to convey on this blog (among other things), is that what we are dealing with wrt Corona, is characterized by Uncertainty, i.e what Donald Rumsfeld called “Unknown Unknowns” (and was rediculed for). And as we all know (right…?) uncertainty is very different from probability – in the former case, we don’t know enough to even assign probabilities to the various outcomes – heck, we don’t even know what outcomes there are (“unknown unknowns”).  So, the only viable way forward in situations where uncertainty rules, is to proceed with caution, trying to identify the risks, and mitigate them as best as we can, with the resources we happen to have available.

I’ve written extensively on this topic, two of those posts can be found here or here.

And the Facebook post that came flying by the other day, did an excellent job of explaining just that.

I managed to get in touch with the author, Johan Lundström, and asked for his permission to post an English translation here.

Now, I’m not a certified translator (I’m basically just a Grumpy Old Man, not certified in anything except BS! 😉 so in case you find what follows linquistically weird, then Mea Culpa, not Johan’s.

Here goes Johan’s excellent analysis:

DISCLAIMER

This is my current best understanding of the Swedish Corona strategy and the
alternatives Sweden had, and didn't have.

Before you complain that I have no medical background, this analysis does not
depend on medical expertise - what this analysis is about is game theory and
general logical reasoning.

THE ANALYSIS

We may identify five possible strategies for the Covid-19 epidemic. The
boundaries between them are fuzzy, so anyone who wants more or less alternative
strategies are welcome to stir the pot - this analysis will be kept at a
conceptual level.

POSSIBLE STRATEGIES

1: "Hold the Border"

Through testing and entry restrictions, we try to stop the virus from getting a
foothold in the country at all. This is what all countries would have preferred
to do, but unfortunately it assumes that you already have what you need in
place. That you are ready for the virus. Any virus. Awareness of SARS and that
China is always lying seems to be factors that would have helped. Being an
island seems like a big advantage. This was never something that Sweden could
have opted for nor succeeded in, unless doing something quite dramatic such as
closing the borders hermetically at the beginning of February, before the virus
even got started. Examples of hold-the-border countries: Taiwan, South Korea,
New Zealand, Iceland.

2: "The Hammer and then Hold the Border"

The idea here is that you've already got the virus in the country so you can't
hold the border as above. Instead, you strike with a hard lockdown until you get
the infection rate under control, after which you can let testing and tracking
take over. You may need to make occasional lockdowns, perhaps locally, whenever
infection is accelerating. China seems to have succeeded in this so far, and
many countries that ran a tougher lockdown - like Denmark - have had this
approach. The big question is whether it works, in the long run. If it does, then
you'll have a winning strategy. Otherwise, move on to strategy 3.

3: "The Hammer and the Dance"

Like the above, you start by knocking down the infection rate properly, but then
the goal is to keep it at low levels through reasonably high restrictions,
perhaps intermittent ones, such that R≈1. You will then have an ongoing spread
of infection at a low level, and hope to be able to hold out until the vaccine
arrives. The risks of this strategy are obvious - the cost of keeping R close to
one can be high and the need to hold the restrictions can last for very long,
during which you'll burn political & trust capital while you are on it,
(especially if you change the rules often), you do not know when that vaccine
actually comes, if ever. It is likely that many of the countries that have opted
for a tougher lockdown will end up here. It is also fundamentally this strategy
that Sweden actively has rejected.

4: "Flatten the Curve".

This strategy assumes a realization that you will not get rid of the virus in
any other way than hoping for herd immunity to be achieved as quickly as
possible, in combination with increased population awareness, understanding and
cooperation, volontary social distancing, perhaps combined with moderate
restrictions. Furthermore it assumes that a vaccine will not arrive in any near
future. Thus, the plan is to achieve herd immunity in orderly forms. The curve
needs to be "flattened" - thus becoming longer and flatter - in order not to
overload the healthcare system. This is the strategy you choose when you simply
do not believe in variants 1-3. It is also, however strange it may sound, a risk
mitigation strategy (I will return to that). Sweden is the flagship of this
strategy. (Before anyone says that the FHM denies that herd immunity *is* their
strategy, let's face the facts: by all available evidence: it *is* the Swedish
strategy! Otherwise what Sweden is doing is _ completely incomprehensible_ if
you remove the herd immunity aspect. Furthermore, FHM have on several occasions
referred to what can't be anything but herd immunity, far too many times for
that not being a key component of the chosen strategy)

5: "Let it burn"

Let's face it, a viable strategy is to do nothing, or very little. This means a
rapid and extensive disease phase and higher death rates, but then the
outbreak is over and done with in a few months. It is a completely unacceptable
strategy in developed countries, but in many poor countries it can actually be
quite reasonable, and often, the only option available - a disease that kills
one percent of the population, the old and fragile, and then is over and done
with. In these cases, with the limited options available, the results of this
strategy will likely not be worse than other epidemics or wars.


THE SWEDISH WAY:

So why has Sweden decided on its strategy, that is, by all available evidence:
strategy 4 ? As can be seen above, which strategy you chose depends on what
premises you have. For Sweden, the following considerations seem reasonable:

Strategy 1: Impossible, we were not aware enough, fast enough nor prepared
enough to handle this type of crisis.  [BLOGGER COMMENT: As is now a popular
excuse from our leading politicians: "we were a bit naive!" :-)]

Strategy 2: Not doable, either because we did not think it
would work long-term, or because we believe it will not work financially.

Strategy 3: Impractical and uneconomical. The idea is that the effects of this
type of long-term lockdown will ultimately be far worse than the direct effects
of the virus itself.


We can hypotetically elaborate the premises a bit more:

a) An acceptable vaccine will not arrive until after the next winter, at least
we can not depend on a vaccine arriving earlier than that;   AND

b) The costs and consequences of trying to keep the infection down in the long
run are prohibitative. 

What will ultimately determine whether the Swedish strategy was good or not is almost entirely
about whether these two premises _ are correct_ (both need to be correct for the strategy
to be a good one). Let me give you two scenarios.

Scenario 1: Astra-Zeneca's vaccine (or a vaccine from some other company, pick
your choice) turns out to work well, and there will be a billion doses available
in September-October. The countries that hammered down the infection by strict
lockdowns and endured until mass vaccination of the population will be the
"winners". Sweden will stand there like an idiot, after having got three to four
times more deaths than necessary. Should this scenario materialize, heads will
roll in Sweden.

Scenario 2: It takes two to three years to get a good, reliable vaccine. Holding
the border either will not work, or then will be paralyzingly expensive. All
countries must either give up strategies 1-3 and accept 4 instead, or take
sky-high financial costs, orders of magnitude worse than those accumulated up
until now. Sweden, on the other hand, has already achieved herd immunity before
the turn of the year 2020-2021, has not taken many avoidable deaths (total death
toll maybe 12000-15000), and has saved lots of money and some suffering not even
trying to fight a doomed fight. The Swedish strategy will be seen as a success.

Note that scenario 2 is worse - even for Sweden! - than scenario 1, and that
this is only a narrative about how good the Swedish _strategy_ finally turns out
to have been. That's what I meant by risk minimization - in any case, it won't
be _ worse_ than scenario 2 (except for unlikely scenarios where no effective
immunity is developed at all, but in that case we have nothing else to do but
discuss fundamentally new strategies - those of saving the civilization from
extinction instead).

There are some very valid arguments for actually choosing - as Sweden appears to
have done - a risk mitigation strategy when you are dealing with uncertainty,
i.e you do not know anything about the underlying probabilities of the events. 

Furthermore, the antibody tests of the past few days might be seen as drawing a
shadow over the Swedish strategy, but will likely not be a death sentence for it
: there may be half as many infected as hoped, but it fundamentally means that
we can expect somewhat higher death rates and longer time in the execution of
strategy 4, not that the strategy won't work.

One final point is that if one is to criticize the Swedish strategy, one must at
least make sure to do so on reasonable grounds. It is _not_ a valid criticism of the
strategy that we currently have among the highest death rates in the world - on
the contrary, it is an *inevitable consequence* of shutting down softer than
others and working on herd immunity.

That is what is expected even when the strategy in fact _ works_. Even in
scenario 2 above, where the Swedish strategy has been a success, it will look
like this.

 Valid arguments include:

- that the FHM apparently cannot make forecasts (although it is more of an
argument against them than against the strategy itself)

- that FHM seem to avoid questions about the strategy and the bad outcomes

- the bad outcomes, the very high mortality rates

- that testing is generally outrageously bad

- that mortality in elderly care has become excessive when the elderly
are systematically denied care and the spread of infection exploded

- believing that Sweden is wrong about premise a) or b) above (if just one of
turns out to be wrong, the strategy failed.

 

Posted in Epidemics | Tagged , , , | Leave a comment

Corona Sweden : Excess Deaths (all causes) Year-to-Date going down

SCB.se has just published an update on this year’s Excess Deaths (compared to a baseline of 2015-2019) up until May-22.

SCB is the government bureau of statistics, and these numbers represent *all* deaths, not just those accounted to Corona. Thus, by comparing this year’s current number of deaths to the baseline, we can establish “Excess Deaths”, which is shown in the graphs below.

Daily Excess Deaths:

SCB_excess_daily_0522

As mentioned in a previous post, April was the deadliest month in 20 years, but now, since a few weeks back, the number of excess daily deaths is clearly going downwards.

It’s also important to notice that before the outbreak of Corona, this year’s death rate was lower than the base rate.  This will become more clear when we look at the cumulative number of Excess Deaths below:

Cumulative Excess Deaths:

SCB_excess_daily_cumsum_0522

As can be seen from the cumulative graph, until late March, Sweden had a “deficit” of about 1400 deaths. Thereafter, the number of daily excess deaths has been increasing, until mid-April.

From mid-April the growth of excess deaths has been slowing down. On the graph above it looks like the peak, at about 2600 excess deaths, has now been passed.

However, from previous experience with these data sets, we know that the last 5-7 days of the data, that is, in this case, data from about may 18th onwards, will most likely be updated upwards. Therefore, let’s cut out the last 7 days of the data, and look at data up until May 17th, and do so by looking at the weekly totals:

Weekly Excess Deaths:

SCB_excess_weekly_0522

Here we can see that the number of excess deaths per week is clearly going down, even after discarding the unreliable data of the past 7 days.

Finally, let’s look at the total number of excess deaths week by week, still discarding the last 7 days:

Cumulative number of Excess Deaths:

SCB_excess_weekly_cumsum_0522

As can bee seen here, the growth rate of excess deaths has been slowing down starting early May, and it will be very interesting to see the what the trend looks like after next week.  My guess is that the downward trend will continue.

In summary:

  • up until May 17th, Sweden had 2637 excess deaths for the year, compared to baseline of 2015-2019
  • total number of deaths 2020 Y2D is 5% larger than that of the baseline years

 

 

 

Posted in Data Analytics, Epidemics | Tagged , | Leave a comment

‘Nothing can justify this destruction of people’s lives’

From Spiked:

‘Nothing can justify this destruction of people’s lives’

Posted in Epidemics | Tagged | Leave a comment