Updated Look at COVID-19 Impact on Reported Mortality
Recently, I have again been hearing questions raised from friends and the media regarding the reported number of deaths attributable to COVID-19 in the US. Most often, the question is whether the reported number is too high — with people commonly citing the example of an infected individual who dies in an automobile accident, and then the doctor cites coronavirus as a contributing factor on the death certificate. Similar to the analysis I wrote about in early July, I can conclude that the total number of deaths occurring in 2020 — regardless of cause — remains significantly higher than the pattern established in the reported data over the prior 20 years.
That is, regardless of how local medical personnel categorize cause of death, the total number of deaths clearly show that something has changed in 2020 — and, conservatively, the excess deaths from that change is at least 200k through mid-September, and probably more than 225k.
As always in these articles, I would first like to describe the data used and its potential limitations. The raw data for the number of deaths in a period is exported from the Data.CDC.gov repository. For this update, I was looking to split the data by age bucket — which somewhat limited what I could find for 2020 versus the July article, but expanded the data back to 1999 that could be used to establish historical patterns.
As before, one of the most important characteristics to note about the reported mortality data by the CDC is the lag in reaching stable results. The first chart here compares the overall number of deaths by week captured in the September 16th dataset versus the July 8th dataset — the primary in the prior article. It shows the reported number can have upward revisions going back months — although, the material upward revisions (say, over 5%) are typically within 5–6 weeks. The point is that I am comfortable to cite precise analytics of reported results here through the end of July, but am cautious to give anything besides broad estimates for the period after that (and I am very skeptical of any media sources referencing precise mortality statistics for August through September, at this point).
The next caution about the 2020 reported data in the repository is it comes from a file named “Provisional COVID-19 Death Counts by Sex, Age, and Week” which starts with the week ending 1 Feb 2020, rather than the beginning of the year. There is also no data for 2019. The advantage of using this data series rather than the weekly series by state used previously is, first, the Age breaks, and second, a longer history when combined with a monthly series by Age covering 1999–2018. To compare the current year data to the pattern established in the prior year data, I use the 26 weeks ending between 8 Feb 2020 and 1 August 2020, compared to the 6 months from February to July reported in the earlier data sets. The overall story is shown in figure 2 below.
This chart makes clear that 2020 is not in line with the prior pattern — especially when you look at the 2-year growth rate in number of deaths. Historically, the number of deaths has grown in the low single digits during most 2-year periods (2–3%), save for 2 occurrences when it was negative and 2 others when it was about 5%. The median 2-year growth rate over the 20 years is 2.55%, which is in-line to slightly faster than the population growth rate during the period (per US census bureau data).
The growth rate for the February to July period of 2018 to the corresponding period of 2020 is over 17% (and will likely go higher in future revisions) and appears to be a clear aberration versus the historical pattern. Some may point out that there is a leap day included in 2020, but I think that is a stretch to be a major driver of the increase. Something is causing a higher number of deaths this year — and the emergence of COVID-19 needs to be at the top of the suspect list.
But how many excess deaths could it be?
If we assume the growth rate for deaths in 2020 in the US was going to be 5% even without the pandemic — that is, a high-end assumption (say, due to the leap day plus a highly stressful and polarized political/cultural environment) — then, the actual growth rate implies there were 170k excess deaths for the 6-month period. For context, there were 1.6mm deaths in the US from all causes in the 6 months from February through July — so 1 out of 10 of those deaths should be labelled in the “excess” category. It is not a hoax — more people have died in the US this year than anyone should have expected.
Taking it further, the excess deaths through July corresponds to a weekly average of about 6500 — so, as of this writing (18 Sep 2020), it is highly likely there are over 210k more deaths than we should have expected based on the pattern established over the prior 20 years. And, if we change the 2020 assumption to a more typical 2.5% growth from 2018, there have already been more than 250k excess deaths in the current year.
Possibly more sobering is what the data shows about excess deaths by age group: the belief by many that the elderly — living in nursing homes and nearing the last stages of life — are the only group susceptible to the most severe impacts of the novel coronavirus is just not accurate. Every age group above 15 years old used for reporting by the CDC, shows a similar spike in overall deaths from all causes in 2020 versus the established patterns over the prior 2 decades. In fact, the next chart shows the highest age group (85+) did experience a clear increase in the current year growth rate versus the historical median — but, it is a small absolute increase relative to the 8 age groups shown. Based on 2020 spike versus historical median, the 35–44 age segment has been the most severely impacted, followed by 25–34 year-olds. That is, these age groups represent a disproportionate share of the excess deaths occurring this year.
I did not include the 3 child age buckets reported on by the CDC here, as they account for only about 1–2% of all deaths in the US in most years and, the calculated growth rates can be very volatile due to the small numbers. I can say that the overall number of deaths in the under-14 age groups in the Feb-July period of 2020 is lower than the corresponding period in 2018 — a glimmer of good news versus the other grim statistics cited here.
The conclusions are clear:
(1) It is not a hoax that there are significantly more reported deaths in the US this year than could have been expected before the outbreak of the COVID-19 pandemic.
(2) This increase is not isolated to the elderly.
