Thread by @strijkers

The curious case of #Sweden

In contrast to other countries Sweden’s confinement measures against #SARS_COV_2 are softer. Their data are for some prove of irresponsible opportunism, for others that the rest overreacted.

Both are (for now) wrong. Long thread on data and nuance./1

Let’s start by debunking some “myths”:
1) and foremost, no, Sweden is not actively pursuing #Herd_Immunity as strategy.
According to Tegnell, Sweden’s state epidemiologist, herd immunity is a *potential* end state, not a potential response strategy.
So what is their strategy? /2

As in other countries their objectives are epidemic peak mitigation while having as little casualties as possible. Math models, as that of the Imperial College, showed millions of deaths if #corona is left unchecked (changing UK and US strategies)./3
imperial.ac.uk/news/196234/co…

So why isn’t Sweden doing the same?
Short answer: 1) mutual trust (public follows advice responsibly); 2) (often overlooked in int. press) they estimate #COVID19 mortality differently, as explained by Tegnell in ‘Dagens Nyheter’ /4
dn.se/nyheter/sverig…

dn.se/nyheter/sverig…

The gist: Tegnell argues that a strategy should also consider the long-term of a chosen response to the #coronavirus. While most models estimate direct #Covid_19 mortality over a period of months, Sweden includes estimates of indirect #Covid_19 mortality over a period of years /5

The Swedish health council in response to #corona doesn’t only include epidemiologists, virologists and microbiologists, but all sorts of medical and health experts to estimate the impact (direct and indirect) of different strategies in the long-term. /6

For example, what would the long-term effects of (repetitive) strict confinement be on health? Think of: cardiovascular problems due to inactivity and higher obesity, increased alcohol consumption, shortage of medicine, home violence, increased poverty-related deaths, etc. /7

By taking an interdisciplinary approach and looking long-term, the Swedish health board concludes that softer confinement is a better approach.
2) Which brings us to the second “myth”: It is not the case that ‘Sweden is open for business as usual’. /8

Schools are open, but only up to 15y; secondary schools (15-18y) and universities not; gatherings of +50 people are not allowed (sport events, concerts,...); working home stimulated; distance measures for bars and restaurants (e.g., fewer tables and closed when not respected) /8

Risk groups (elderly, retirement homes,...) under confinement, the strong advice to uphold social distancing. The latter relates to ‘mutual trust’, where Swedes are expected to follow health advice and act responsibly (the example Tegnell refers to: vaccination; /9

even though vaccination is not obliged in Sweden, only recommended, 98% of the Sweden follow the advice and are vaccinated).
The fact that Sweden, in contrast to other countries, follows an interdisciplinary and long-term approach is interesting, but imo also risky /10

It’s interesting, because they are likely right that #SARS_COV_2 related (indirect) deaths will occur and that those numbers increase with strict confinement.
However, as a scientist working with multi-dimensional & inter-disciplinary data, ‘error’ is what worries me:... /11

Adding independent variables to your model (other health issues, longer time period,...) may explain the variance of your dependent variable (i.e., what you measure/estimate, for example, deaths) better, but it may also add randomness and increase error (=over-fitting). /12

An example: you can make a model for explaining/estimating deaths in traffic (dependent variable) by looking at ‘speed’ (independent variable) and learn that from a certain speed, deaths markedly increase (for example >90km/h). Now you can add the variable ‘weather’.../13

This is interesting because it can give you additional info, namely an interaction between ‘speed’ and ‘weather’. You learn that the speed limit to minimize deaths in car accidents when looking at speed alone (90), needs to be lowered more (70) under bad weather conditions /14

Now one can add more variables (tire pressure, age, brand, nationality,...). The problem is that if you add stuff where independent data is uncertain (unlike speeding or weather, for which concrete numbers exist), probability of error increases and.../15

interpretation of multiple interactions becomes harder or even impossible; furthermore, because your data becomes sparser (spread out over more dimensions), you may capture random noise rather than relevant trends and underestimate critical variable contributions /16

Returning to the Swedish approach, one has to wonder on what these second-hand #coronavirus deaths are estimated. Is it true that strict confinement will necessarily lead to increases in obesity or alcohol consumption for instance? And even if so, how much? This is not clear. /17

If they get those indirect variables wrong, they may be widely underestimating the direct variable (death caused by #SARS_COV_2), with a potential catastrophic death toll in return.

What can we learn from their #coronavirus graphs? /18

You may have encountered graphs like this, where the less steep curve and relatively lower number of cases in Sweden is taken by some to argue strict confinement as in the rest of the world is an overreaction /19

However, comparing cases, even relative to population, is not useful since different countries adopt different testing strategies (Sweden performs few tests). Also, demographic (population density, household make-up) and cultural differences can explain the slower onset. /20

Comparing mortality is better, but also may give quite different comparisons:
Left: Sweden seems to be doing much worse.
Right: Sweden seems to be doing quite well.
What’s going on? /21

Also mortality has comparability problems: 1) onsets differ; 2) manners of counting differ. Sweden only counts confirmed #Covid_19 patients who died, *but* regardless of cause of death and groups together deaths in hospitals and retirement homes /22
experience.arcgis.com/experience/09f…

Imo, better comparative measures between countries are hospitalizations and used intensive care units (icu). Not perfect, but arguably with less systematic differences. When doing so between Nordic countries, Sweden is clearly in worse shape. /23

research.swedbank.se/PublicSubscrip…

However, according to Tegnell, this is still within their expected range and their objectives, as mentioned, are more long-term (note also that Sweden has about 3x more habitants than Denmark; taking into account that 85% lives in the south, density is roughly the double) /24

compared to other EU countries, Sweden is (for now) not doing worse (e.g., Belgium: 106/mil; Netherlands: 81/mil; France: 103/mil; Germany: 60/mil; Sweden: 82/mil), also when correcting for onset (closer to Belgium and France, but still) /25
coronatracker.com/analytics/

Another surprising observation is that Sweden *may* be stabilizing (reaching peak), which would be faster than several other EU countries (but a stabilizing curve doesn’t necessarily mean peak; could be local minimum driven by Stockholm data)./26

Take home:
- Are the Swedish data supporting claims that confinement in response to #coronavirus is a ‘cure’ which is worst than the disease?
No. Simply because Sweden *is* also taking confinement measures, just less strict and more based on public responsibility. /27

A poll showed more than 70% of the Swedes significantly changed their behavior in response to the recommendations. Add to that closed higher ed, prohibited 50+ gatherings, stricter for people at risk, etc., then one can hardly call it (as often done) a laisser-faire strategy. /28

In addition, their data (in particular people needing intensive care) looks more grim than it’s Nordic neighbors (likely the most reliable comparison)
- So it proves their strategy is irresponsible opportunism (economy before life)? Also no. /29

For one, decisions on public health in Sweden are made by the health institute, not politicians (by law). And they do follow a strategy, a long-term one. Also, while worse as their neighbors, the numbers did not (yet?) skyrocket as could be expected from math simulations /30

Some final thoughts:
- experts across the world are fighting the #CoronaOutbreak, so why isn’t anyone else following the Swedish model? Did they really not consider long-term estimates? Or did they and quickly disregarded it as unreliable and dangerous? /31

- And even if the Swedish approach works and they end up in the long run with less economic, social and (according to them) human damage, would this hold for other countries as well or are their demographic and cultural particularities a necessary condition? /32

In any event, the Swedish approach will be a critical case study over the coming weeks, months and years to “hold a mirror up to” the world’s pandemic response (reflecting beauty or horror).
/end.