Expected Death Analysis

Covid-19 is causing a higher death rate than normal. There has been some controversy as to whether covid-19 was the cause of death, even if it was present in the deceased. To avoid this controversy, it's possible to just look at the number of deaths in a particular week and compare it with the number the same week over the past few years. The graph below uses data from the CDC ( JSON). In particular, the week of reported deaths and the percentage of expected deaths columns are used. The percent of expected deaths compares the death count in the particular week to the average for that week in 2017 throgh 2019. As can be seen in the data through March 14, this number is pretty stable. However, after March 14, we see the percentage of expected deaths increase dramatically. Towards the end of the dataset, the percentage of expected deaths is very low. This is because the CDC records the actual date of the death, but may not receive that report for several weeks. Their analysis of 2015 through 2016 data showed that less than 25% of the deaths were reported within the first few weeks. At least 75% of the deaths are reported within 8 weeks. Therefore, data should not be considered complete until at least 8 weeks after the date of death, and even then could be up to 25% low.

Data not complete until at least 8 weeks after date of death

Remember to look back at least 8 weeks for accurate data. Looking long term, since we are all going to die some day, the coronavirus has "moved up" the deaths of a large number of people. How far were they moved up? If they are moved up within the view of this analysis, we expect the % of expected deaths (blue line on graph) to go below 100%, and the accumulated percentage of expected deaths to settle back down to around 100%.

Reporting Propagation Delay

As noted above, it can take weeks for the death data to be reported to the CDC. It may be possible, however, to predict final numbers based on early results. If reports to the CDC take the same path week to week and there is not large variation in delay from certain geograhic areas that have large numbers of deaths, we could predict final numbers based on early numbers. If, for example, we find that 10% of the total deaths are reported by the Monday following the Saturday ending the week we are analyzing, we could predict the final number by multiplying the initial number by 10. As we move farther and farther past the end of the week being analyzed, the "correction factor" gets smaller and our estimate of the final number becomes more accurate.

To test this, I have started gathering numbers for the week ending June 13, 2020 starting June 15,2020. I'll add additional weeks as they become available. If this prediction method works, we should see consistent numbers for the "Percent of Max" for each of the days after the end of the week (called "Delay Days"). The analysis is here.

Average Age at Time of Death

I have not yet found a source for this data on a weekly basis. It would be interesting to see how this varies over time. There may be some seasonal variation. How much? There is likely to be an overall decrease in the average age at time of death due to covid-19, but how much of a decrease?

Comments to harold@hallikainen.org