489 Days of Studying COVID in SWEDEN

Inspired by my friend Professor Michael Levitt’s recent analysis, “Lessons from 500 days studying COVID-19” , I thought I’d post a brief summary over what’s been going on in Sweden up until May 2021, as a follow-up to my summary of 2020 back in January.

First, a look at “Covid Impact” from start to YTD in terms of positive tests, Covid-ICU’s and Covid-attributed deaths, in proportion to population in each age group [using 2020 population count]:

As seen above, proportion positive tests is dominated by “the working ages”, that is, 20-60 years old, while ICU admissions are most frequent in 60-80 years old, and deaths are almost exclusively occuring in 80+.

Next, COVID events timeline:

According to the above timeline, Sweden has had two waves of positive tests, peaking January & April 2021, three waves of ICU admissions, April 2020, Jan & April 2021, and two waves of deaths, April 2020 and Jan 2021. Worthwhile to notice here is e.g. that deaths during the latest wave of positive tests and ICU’s were very low. Any thoughts on how come…?

Also worthwhile to notice that during the first, deadly wave during spring -20, the number of “cases” seem to have been very low… Any ideas why…?

We can also look at these COVID-events on a week-by-year basis:

The most remarkable thing to me here is the temporal similarity of ICU admissions: they peaked the same week of year both 2020 and 2021, with an almost identical pattern. Any ideas on why…? And of course we again see that the latest wave of ICU’s did not have a corresponding wave of deaths.

But tracking COVID-specific events is IMO not the best way to understand the impact of a pandemic. I prefer using All Cause Mortality (ACM), since that metric is not fuzzy, dependent on what counts as a COVID-{case/ICU/death}. All Cause Mortality is based on something very binary – either you are alive or dead. So, in what follows, we’ll look at various metrics based on ACM. Also, unless otherwise stated, I’m using average mortality 2015-2018 as baseline in all calculations below.

First, a look at weekly ACM for age groups -64,65-79,80-89,90+, 2015-2021 YTD:

Above, the twin peaks of spring 2020 and winter 2020/21 are clearly visible for all age groups except -64, where only the spring peak is (barely) visible, and where only 2 weeks are (slightly) above the baseline range for those weeks. We can get a better picture by looking at the same data, but on a monthly asop to weekly basis:

The observant reader should notice something interesting in the above graph: after the 2018 spring flu season, mortality for all age groups was basically at or below average until COVID hit the ACM numbers in April 2020. So, during the fall/winter of 2018, spring 2019, fall/winter 2019 and spring 2020 (until COVID struck in March/April), overall mortality was way below average.

Could this exceptionally low number of deaths during a period of almost two years have any impact on deaths during 2020… Particularly if a new,possibly human engineered potent virus comes for a visit…?

2020 and 2021 YTD – Normal or Abnormal ?

Let’s next look at “normality”, that is, expected number of deaths, vs. “abnormality”, actual, observed number of deaths, for the same period of 2015-2021 YTD:

One way to define normality is for the red plots in each of the four subplots above to basically overlay the black plot, that is, the observed deaths should be more or less equal to the expected deaths.

Let’s focus on 2018 and forwards : how many “abnormal” months do you see, and which way does the “abnormality” go…?

Of course it’s easy to get caught staring at the “twin peaks” of 2020, but if you look a bit more carefully at the previous years…?

Perhaps what I want you to see is easier to see in the next graph, showing the cumulative number of expected vs. observed deaths for each age group:

To me it looks like 2015,2016,2017 had observed deaths pretty much in line with expectation. But from 2018 spring onwards, at least a couple of the age groups started to show a “gap” between expected and observed deaths, and during 2019, that gap becomes really conspicuous for all age groups! No wonder though, since we saw from the previous graph that monthly deaths were never above the baseline, the expectation, for any age group, during entire 2019. Could this remarkable state of affairs have had any impact on the number of deaths 2020…?

Let’s give a name to this “gap” between observed and expected deaths : “EXCESS DEATHS”.

Now, when we know what we mean by “excess deaths”, let’s look at monthly cumulative excess deaths per year, still 2015-2021 YTD:

If you’d look at this graph but without any other context than the dates given on the x-axis, my guess would be that you would consider 2019 being the most exceptional year out of the 6 and a half years presented…? But for some reason, as soon as it becomes clear that the data on the chart is about excess deaths, the focus immediately turns to 2020, which indeed was an outlier, as can be readily seen on the above chart. But if 2020 is considered an outlier, what do we then call 2019…? And by the way… doesn’t 2021 at least look like the beginnings of yet another outlier year….?

We can combine the age groups and look at total monthly cumulative deaths for the same period of 2015-2021 YTD:

To summarize our findings thus far: while the years 2015-2017 ended up pretty much on expectation, and 2018 with a minor “death deficit”, 2019 ended up with an unprecedented death deficit. After that deficit, 2020 ended up with only a minor number of excess deaths (between 1K and 4K, depending on how you define your baseline, and the granularity of your age binning), despite the pandemic, and despite the fact that Sweden, “The World’s Cautionary Tale“, had no Lockdowns & no masks (at least until the government folded to international pressure in Nov. 2020).

We can also summarize our total yearly excess deaths, for each year 2015-2021 YTD:

Do you count your skiing days per Gregorian year or per season…?

But wait – there’s even more to (dis)cover.

We know that mortality in Sweden (I haven’t looked at other countries in enough detail to tell) is heavily seasonal, particularly for those older than 65 (have a look at the charts above). Basically, most deaths occur early in the year and late in the year. Pretty much the same as for my skiing: good skiing years I’m able to start skiing in November, and continue all the way until April. So there’s a strong correlation between my skiing and Swedish mortality…! Maybe I should give up skiing…?

So, mortality peaks in Nov-Dec and Jan-Mar. But wait – that’s actually two different years…! No wonder I count my skiing days not based on the Gregorian calender, but per season, with each season starting sometime late fall. It simply makes much more sense to count skiing days per season asop to calendar year, since skiing is so clearly an activity heavily impacted by seasonality.

And so is mortality, so why not look at seasonal mortality, instead of the standard yearly mortality…?

Exactly as another friend of mine, Professor Eyal Shahar, recently did in this brilliant article.

So, let’s use the same techniques as for the year based analysis above, but now instead of using year as our prime unit of time, we use season, defined as the period from October to September:

First, monthly seasonal cumulative age group mortality 2015-2021 YTD:

Can you see the outlier(s)….? I can’t. Let’s look at the seasonal expectation vs. observed:

A couple of things to notice: the season 18/19 had a clear death deficit every month, for each age group, and also the season 19/20 started out with a death deficit until May-2020, when COVID started to impact the numbers. Also worthwhile to notice is the last bar of each age group for the season 19/20 : the under 65’s as well as the 80-89 year olds (where most deaths typically occur) ended up with a death deficit, while the remaining two age groups, 65-79 and 90+, ended up with a barely noticeable excess.

Since the current season, 20/21 is not over and done with until end of October, the only thing I’m going to say about it is that despite the COVID wave of Jan 2021, things current season do not look very exceptional at all to me.

Finally, let’s combine the seasonal age group data, and look at the expected vs. observed deaths for seasons 15/16 – 20/21 YTD:

Spot the anomaly.

As a preemptive strike against those readers who I’m sure will state that “but your baseline of 2015-2018 should include 2019!” or “you should take into account the fact that mortality has over the past decades been gradually declining!” : I’ve covered why I’m using the average of 2015-2018 as the baseline several times before on this blog, and can’t be arsed to repeat that again. If you really want to know why I prefer 2015-2018 as the baseline, just do a search on “excess deaths” in the search box of this blog.

PS: I really should have written this post the day before yesterday. If so, I’d at least could have claimed to have a prime number in the title….