Covid Sweden – status 2020-08-03

SCB.se just released their weekly preliminary stat’s on deaths all causes.

However, before looking at that data, let’s take a look at how Covid has evolved in Sweden, based on the numbers from Johns Hopkins University:

weekly_increments_Sweden

Something remarkable seems to have happened in June…: The number of cases goes way up, but the number of deaths continue downwards… Any guesses on why this is…?

My take on it is “what you test is what you get”, i.e. the number of cases is fully dependent on the amount of testing, and testing in Sweden has varied a lot over the past 5-6 months.

Let’s take a look at the same data, but with a different presentation, based on an analysis I abandoned a couple of months ago, when I realized that the numbers on confirmed cases and covid-registered deaths really don’t tell us the true story… The only metric that has real value in understanding the effects of Covid is All Cause Deaths, which we will look at in a moment.

So, in the below graph, I’ve run a Bayesian Inference on expected number of deaths, based on the number of confirmed cases. Each green “blob” represents a probability distribution for expected number of weekly deaths – rotate the blobs 90 degrees in your head to see that they are probability distributions – and the red dot represents the actual number of deaths for that week.  The x-axis values are weeks, starting when the number of confirmed cases hit 1000.

The basic idea of the model is thus that it “predicts” how many deaths per week there “should” be, given the number of confirmed cases, while inferring probability distributions for mortality rate & ratio asymptomatics within the population:

Sweden_violinplot_weekly

In the graph above, we can see that early on in the pandemic, the model predicted fewer deaths than the actuals (red dots), but from the 12th week – around June – the model predicts far more deaths than actually occurred.  Thus, when the amount of testing goes up, the number of confirmed cases go up, but deaths don’t necessarily follow.

The main takeaway from this is (again!) that focusing on cases really doesn’t give us relevant information for making decisions on virus mitigation policies. Neither is the number of “Covid-deaths” very useful, because as we now know, what gets registered as “death by Covid” in fact in most cases is death *with* Covid, i.e. an individual with multiple severe comorbididies, where Covid is the “last of thousand cuts”.

So, for the past few months, I’ve pretty much abandoned analyzing the Covid-related numbers, and instead exclusively focused on All Cause Deaths as a metric for how a country is doing, in terms of overall health care, elderly care and mortality.

All Cause Deaths

In what follows, please keep in mind that the numbers on all cause deaths from scb.se are *preliminary*, and we know from previous weeks that typically, the last 7-10 days or so of the dataset very plausibly will be slightly adjusted.

First, let’s look at weekly deaths  and compare this year with a baseline of 2015-2019, which also gives us the “Excess Deaths”:

scb_poisson_weekly

Now, since the numbers are preliminary, and we know that the last 7-10 days very likely will get revised, let’s just pretend the last data point is not there, i.e we skip the last 7 days of data: still, the trend is clear : already during the 3 weeks before the last, the number of weekly deaths has been below average, and for the two latest weeks, the number of deaths has been below the Credible Interval, in fact, touching upon “all time low” for the period of 2015-2019! Currently, Sweden has about 3300 “Excess Deaths”, but considering that 2019/2020 flu season left us with a “death deficit” of about 2600, it’s now clear that actual “Excess Deaths” are less than 1000.

Next, let’s compare deaths per million, stacked by month, Y2D 2020, to 1990-2019:

scb_1990_2020_deaths_per_M_stacked

Now, while still keeping in mind that these data only run to July 29th for 2020, *and* that the last 7-10 days might get slightly revised, it’s still reasonable to state that thus far,  all cause deaths in Sweden for 2020 Y2D are… quite “Normal”… nothing extraordinary…! 

Full year 2020 prediction for deaths

In what follows, I’m assuming that deaths for the rest of the year will grow proportionally as they have grown from Jan 1st to July 22, which while considering the very high number of deaths in April+May is an unlikely outcome, but might serve as an indicator for the plausible “high end” number of deaths for 2020.

The chart below shows for 2020  – under the assumptions stated – the expected number of deaths per million, per age group, conditioned on age and population size, compared to previous years:

pymc_cond_age_year_per_M_stratified

It’s quite clear from the graph that expected deaths for full year 2020 – under the stated assumptions – will be well in “normal” range, nothing remarkable, nothing extraordinary, and first and foremost: no cause for panic, disaster and doom & gloom…! Despite all the media headlines.

[graphs with better resolution can be found here ]

[UPDATE 2020-08-03 – I just noticed that twitter handle @HaraldofW updated his analysis on Sweden, here’s a link to it ]

 

 

 

 

 

 

 

 

 

About swdevperestroika

High tech industry veteran, avid hacker reluctantly transformed to mgmt consultant.
This entry was posted in Bayes, Data Analytics, Epidemics, MCMC, Pandas, Probability, PYMC and tagged , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s