US All Cause Mortality: 2020-22 vs. 2017-19

I don’t trust COVID death statistics. Although I have used them at various times in the past, I always recognized the weakness and inconsistency of this data due to varying and often incredibly liberal definition of a “COVID death”. Going forward, whenever possible, when talking about deaths I will look to use all-cause deaths in my analysis.

Recently, I began to examine all-cause deaths in US states, comparing 3 year periods: 2020-2022 vs. 2017-2019. This clearly aligns to the 3 years prior to COVID vs. the 3 years of COVID. All cause deaths are obtained from:

• Weekly Counts of Deaths by State and Select Causes, 2014-2019

• Weekly Provisional Counts of Deaths by State and Select Causes, 2020-2023

For each state, I summed the all-cause deaths across each 3 year period. Next, we’d like to adjust for population changes during these 6 years. To do so, I used the population of the middle year of each 3 year period. I.e., for the deaths between 2017-2019, I used the 2018 population of that state to obtain all-cause deaths per 100k population, and for 2020-2022 I used the 2021 population to obtain that figure. In this way, I could then take the ratios to determine what increase or decrease in all-cause deaths per 100k population occurred in a state between these successive 3 year periods.

As an example, here are the relevant numbers for Oregon, showing all-cause deaths increased by 15% per capita from 2017-19 → 2020-22

With this as background, we can now look to see whether all-cause mortality changes from pre-COVID to COVID era has been significantly impacted by vaccination levels. Is it appropriate to look at all-cause deaths when discussing the vaccine? If (1) COVID is expected to be a significant cause of death in the past 3 years, and (2) COVID vaccines have a significant impact on COVID deaths, then it is not unreasonable to expect COVID vaccines to potentially have a significant impact on all-cause deaths.

---

Note: What follows looks very similar to a short discussion from Mathew Crawford’s post today. He showed an analysis from the Society of Actuaries of Excess Death vs. Vaccination level for Q3 2021. Please consider this post to hopefully be a more robust look at the same question. I believe by looking at 3 year periods, we reduce the possibility of a seasonal impact or some other anomaly skewing the analysis due to a short time period.

---

We will start with a plot showing the change in per capita all-cause deaths vs. vaccination level (per New York Times tracker) for each state (circles are sized by population):

Looking at the national level, there appears to be a statistically significant (p-value=0.003, R^2=0.17) relationship showing higher vaccination levels correlate to a smaller increase in all-cause deaths in the COVID period.

Broken down into the 4 major regions, the correlation pretty much disappears. None of the four regions show statistical significance (p-values of 0.70, 0.56, 0.21, and 0.88 from top to bottom). You can also find states at opposite ends of the spectrum in terms of vaccine levels, but virtually identical all-cause death increases (North Dakota-Minnesota, Pennsylvania-Connecticut, Alabama-Virginia, Wyoming-New Mexico).