Was Less Really More?
This post from Jay Bhattacharya replays a recurring theme for him relating to the purported success of the “Swedish strategy”, with limited non-pharmaceutical interventions (NPIs) to limit the spread of COVID. Here, he asserts that “not locking down” would have saved lives in the UK.
A closer look reveals that he is claiming that fewer people died in Sweden during the COVID Pandemic than would have died if the pandemic had never happened. Stated another way, the NIH director wants us to believe that COVID was good for public health in Sweden.
This is obviously not true. The interesting thing is how did we get to a point where the head of the world’s largest medical research institution is insisting that it is.
That story goes back to the beginning of the pandemic.
Background
Jay Bhattacharya began arguing against public health interventions the moment pandemic began. The so-called Swedish strategy was the framework for the Great Barrington Declaration, with its “focused protection” strategy. As the pandemic continued, it became increasingly obvious that COVID was far more dangerous than he originally claimed and that Sweden was doing far worse than its Nordic neighbors in limiting COVID mortality. Over the course of the next year, he and his fellow anti-interventionists pivoted away from any discussion of COVID mortality and began to talk in terms of excess mortality, asserting that strict interventions resulted in a rise in deaths from other causes. The problem is that excess mortality is not measurable. It’s a made up number. And how you make it up determines the result.
Excess Mortality = Actual Mortality – Expected Mortality
That sounds reasonable, but the word, “expected” is doing some heavy lifting here. It’s a calculation based on past mortality. How you calculate it determines excess mortality. Here Bhattacharya has used a method that yields a negative excess mortality. That means Expected Mortality is greater than Actual Mortality. That implies that Sweden would have had more deaths without COVID than with it.
That means he either didn’t read the graph, didn’t understand the graph, or understands it but hopes we don’t.
First let’s take a look at how Sweden did in controlling COVID. Then we can consider the reasons Bhattacharya got it so wrong. And why he’s posting a graph of epidemiological data from a website run by a software engineer.
Key Assertions
- Sweden was effective as its peer countries in controlling the spread of COVID.
- If we look at longer term excess mortality rather than COVID mortality, Sweden did better than peer countries.
Summary of Evidence (as discussed below)
- When compared to true peer countries such as Norway, Sweden experienced much higher rates of cumulative COVID mortality.
- Excess mortality requires an estimate of mortality had the pandemic not occurred. The answer to this question depends heavily on how those estimates are made and raises serious question about the appropriateness of its use in this context.
COVID Mortality
Consider this graph from the, Our World in Data website, maintained by Oxford University
| Country | Approximate Chinese Visitors (2019) | Population Density (people/km²) | Approximate COVID Mortality (deaths per million) |
| United Kingdom | 860,000 | 276 | 3,400 |
| Sweden | 215,000 | 25 | 2,680 |
| Norway | 155,000 | 15 | 1,200 |
| New Zealand | 407,000 | 19 | 1,160 |
| South Korea | 6,023,000 | 530 | 700 |
Consider COVID deaths in a few comparison countries in terms of their population density, which affects transmission rates and the rate of and the annual influx of Chines tourists, which should correlate with initial COVID cases. Sweden did not fare as poorly as the UK, but the UK has far more Chinese visitors and ten times the population density. Norway is closely matched to Sweden. It had a moderately fewer Chinese tourists and 37% lower population density, but Sweden had 167% higher COVID mortality. New Zealand and South Korea had far more travel from China and South Korea has much higher population density, but both had much lower COVID mortality.
As the failure of the Swedish strategy became clear, Bhattacharya and his fellow anti-interventionists tried to
Bottom Line
Sweden fared far worse than Norway, the closest comparison in terms of climate, demographics, initial influx of cases, and population density. The data are consistent that the less stringent NPI’s in Sweden resulted in substantially higher COVID mortality.
Did Sweden fare better in terms of excess mortality?
Let’s look at the graph. At first glance, it seems to show Sweden with an increasing advantage over the UK in terms of excess mortality. But look again. The graph shows negative excess mortality in Sweden. What IS negative excess mortality.
Excess mortality refers to the actual number of deaths minus the expected number of deaths. In this case, expected deaths refers to deaths that would have occurred if there had been no COVID pandemic. So a negative excess mortality means THERE WOULD HAVE BEEN FEWER DEATHS WITH NO COVID PANDEMIC.
Imagine it is the end of 2019 and your task is to predict mortality for the next five years. That is expected mortality. The only thing you have to work with is past mortality. The solid line here is Sweden’s mortality from 2010 to 2019. Given the clear downward trend in these data, it would be reasonable to calculate a trend line and extend it forwards.
If you did that, the data from the period of the pandemic would look like this. The solid line of actual mortality exceeds the expected mortality for most of the pandemic.
But that’s not what they did.
Bhattacharya replaces the decade-long trend with an average of just three years — 2017, 2018, and 2019. That’s a bit like using your last three games to estimate your batting average.
It is a fundamental principle of statistics that more data gives you a more accurate estimate. If you think an average is a good predictor, why not use ten years of data?
If he had taken an average for ten years before the pandemic, actual mortality would have exceeded expected mortality throughout the pandemic, even during the deadly first year. That would have been hard to sell.
One there thing to note about the graph that Bhattacharya shared. For England, the historical mortality was relatively stable, so an average is a reasonable estimate of expected mortality. That explains its positive excess mortality.
There is one other critical issue about this graph. Note that Bhattacharya compares Sweden to the UK. Epidemiologists are very careful when comparing countries like this. There are many differences between them that could affect COVID rates but have nothing to do with public health interventions. With respect to COVID risk, the UK and Sweden are very different. The UK has far more air traffic than Sweden, including more than four times the flights from China. That means there were far more cases coming in early in the pandemic. The UK also has more than ten times the population density, meaning the disease spread far more easily once it arrived. So this comparison is completely inappropriate.
Bottom Line
The data clearly show the graph was using the wrong methods to make an inappropriate comparison that resulted in an implausible conclusion. The results raise serious questions about the use of expected mortality, which is a counterfactual with no ability to verify findings.