🇸🇪 Sweden's Excess Mortality, Calculated via four Different Methods

No statistical significant excess mortality detected by any method.

Here are the four most common methods to calculate excess mortality:

Last Value (Naive)

Five pre-pandemic seasons, use last value as baseline/expected.

2) Average / 3) Median

Three pre-pandemic seasons, use mean/median as baseline. Three periods are common, as mortality tends to decrease over time, hence we do not want to artificially inflate the average/median using too many earlier periods.

4) Linear regression

A trend line, that takes into account the last ten seasons.

As you may have noticed, I used:

- ASMR (Age-Standardize Mortality Rate) which is superior to CMR (Crude Mortality Rate), b/c it normalizes changes of the age structure over time (and between populations).

- Midyear Season, e.g. July 2020 - June 2021 has been commonly found a less-noisy period, than calendar years. Lastly, we can summarize the resulting excess mortality (forecasted baseline/expected value - actual), and the corresponding prediction intervals.

Excess Mortality

The last column indicates, if in any period, statistical significant excess mortality was detected. That is, a value which would lie outside the forecasted range.

That is not the case, hence none of these robust models has detected any significant excess mortality!

--> No statistical significant excess mortality detected by any method.

Sources:

- Most of the methodology I have used here, is described in the book: https://otexts.com/fpp3/prediction-intervals.html

- I have re-implemented the fable prediction methods in pure TS/JS, so that we can now use them via MortalityWatch.