COVID-19 Impact on Reported Mortality

Over the past few weeks, a few people have begun asking me about the reported statistics on deaths in the US attributed to the COVID-19 pandemic. As with many issues these days, there are two divergent views. One school of thought is that the reported number of deaths caused by the novel coronavirus is overstated, as doctors can list the disease along with a number of additional causes – even when patients have underlying conditions that can cause death anyway. The other school of thought is that the reported number is understated, as some very ill individuals are fearful enough by the crisis coverage on the news that they are avoiding hospitals and other healthcare services – only to die suddenly at home – without ever being tested or diagnosed with the disease. I cannot resolve these two camps from the data I have reviewed but, I can definitely conclude that the total number of deaths in the US – from all causes – is significantly higher so far in 2020 than the patterns established over the previous 6 years. 

Before we get into conclusions, I first need to provide some caveats on the data used in this analysis. The source files I relied upon are from Data.CDC.gov – one of which is a static file providing reported deaths by week for each state for the years 2014-2018 and the other being a file that updates about weekly and provides provisional reported deaths by week for each state for the years 2019 and 2020.

The first thing to observe about the number of deaths reported in the US is there is some strong seasonality in the counts – as can be seen in the first chart below. Deaths are lowest in the July-August period and then rise markedly through December to a peak in January or February of the following year. As I have learned recently, this is primarily due to the seasonal flu – and late-2017 into early-2018 was a particularly bad season (near 80k deaths per CDC versus 25-55k over the prior 8 years).

The second observation I suggest retaining is shown in this next chart: the Jan-Feb peak does not hold for the most recent provisional counts for 2020. Instead, the reported deaths in April are clearly much higher than the smaller peak during the standard flu season. An additional observation to take away from this chart is the sharp drop-off apparent through May and continuing into June.    

 With apologies to Nate Silver, much of the drop-off from the April peak shown is “noise” – meaning it comes from a bias in the data collection process rather than representing true “signal” from the number of deaths across the country over the period. The third chart highlights the caveat in using the provisional reporting from the CDC. The columns show how the number of deaths reported for past weeks consistently gets restated as each generation is published. In fact, the chart shows that revisions between the data as published on June 3rd versus the data as published on July 8th can even go back 6 months. Admittedly, these are very small upward revisions to the results from January through March – but, the chart also shows that between the early June update to early July updates, there have been significant increases in the reported numbers for back in May. Additionally, each update shows no data reported for the most recent 10 days to 2 weeks, along with major upward revisions going back 4-5 weeks. 

The point is that I am now very skeptical of any news sources purporting to show mortality statistics that reach June or later – especially, when they include unambiguous proclamations that mortality from the disease has been solved in recent weeks. 

The chart below shows the scale of the increased number of deaths in April and May. The April numbers (blue bars) are unlikely to be revised materially at this point and average near 20k higher each week than seen in the April weeks for the prior 6 years. In fact, as shown in the first chart, the US had never before suffered a week in the series with more than 70k reported deaths. The prior high had been 68,320 in 2018 for the week ended January 13th – the peak mortality of that severe flu season. Then, all 4 weeks in April 2020 surpassed 70k, with the new high set the week ended April 11th at 78,569. The implication is the total incremental deaths in April 2020 is about 70-80k versus the prior years.

The total number of deaths in May 2020 (orange bars), while lower than April 2020, is still much higher than May for each of 2014 through 2019 – again, representing an increase averaging over 7k per week in the current year versus the each of the prior years. Additionally, the May 2020 results will likely still see some upward revisions – especially, the last week. The implication is the total incremental deaths in May 2020 is about 30k versus the established pattern.  

In total, 2020 has seen more than 100k additional deaths in these 2 months alone. As of this writing, the number of deaths in the US attributed to COVID-19 is reported at 134k on the Johns Hopkins dashboard. So, is it possible that we will find an 20-30k additional reported deaths across the last week or two in March (when the increase versus prior years appears to have begun) plus, all of June plus, the first week or so in July? It is likely to be at least that high. Let’s call it a wash between the two methods, so far. That is, using the number directly reported as due to COVID-19 versus, estimating the amount from the 2020 increase in overall deaths from all causes, reach the same conclusion that we have suffered 130-140k additional deaths during the first few months of the COVID-19 outbreak – despite other secondary effects of the lock-down (such as, less driving and other work activity) helping to reduce the number of deaths across the country.  

The callous question remaining is whether the increase shown over these few months merely reflects a slight acceleration of deaths that would inevitably occur over the next few months from the pre-existing conditions often cited along with the most severe cases from COVID-19. At best, the question can only be answered in hindsight – say in 6-9 months, if we see a reversal of the first-half result and the second-half shows a lower reported number of deaths than the prior 6-year pattern established. Even then, the statistics can still be muddled if the spread of new cases remains high through year-end – as then, the disease could be accelerating deaths from mid-2021.  

Next steps for me is to examine the data by region. Controlling for population, the NE7 has had lower case volume than the other regions for about 30 days now. The region clearly had high mortality statistics in April into May. If the case volume continues to remain low for the NE7 region and simple acceleration of inevitable mortality is the explanation for the April-May increase, we should be able to see this in the data over the next 2-3 months.  

Whether it does or not, I am still not sure the answer will reassure me in any way. Analyzing mortality statistics in detail is a harsh pastime during a pandemic – probably any other time, as well.