An Update on Cohort-Based Trends

By October 23, 2020Commentary

Update:  The chart wasn’t quite right, it started with the first week that the state reported data, in which they just cumulated prior events.  I have replaced it with one without that week, which is more accurate and creates a better scale.  I also corrected my error in pasting the wrong column with Minnesota reported death data instead of CDC data.  Now you can more clearly see that over the last few months, on a cumulative basis, that is for all deaths, hospitalizations and cases, the rates have steadily declined at a slow clip.  That is true with the rate for the current cohort as well, although that understandably is slightly lumpier.  If you look at the table, look at the cumulative death rate.  If this were a CFR, that is a rate of deaths only for reported cases, it is 2.1%, which is a pretty significant rate, but if the state and CDC are right, and there are ten times more total infections, the IFR is .21%, which is down at flu level, and again, we don’t track and attribute flu deaths the way we do CV-19 deaths.  Now look at the current rate, not for the last week, I think a number of deaths still aren’t in for those, so go back almost a month to the week ending September 26.  The current death rate for that cohort is 1.15%, now assume ten times as many infections and you are down at .12%, and that is flu level.  Anyway, I hope this analysis give you a sense of what has actually been happening to sets of cases in Minnesota over time.

I have a couple of times attempted to sift through the Minnesota data to create a cohort-based trend analysis, particularly of hospitalizations and deaths as a function of cases.  I know this is complicated, but I want to explain what I did and why I did it, because many of you may be able to determine if I made any mistakes in construction or calculation.  You can just skip to the end and look at the tables and charts and get the idea.  Several pieces of data have undergone a reporting change in Minnesota, some make it harder, some easier to do this analysis.  I am taking a slightly different approach to this analysis than that used in the earlier ones.  The point of the exercise is to see if groups of cases over time have different outcomes.  A cohort is a week, because that is how you can get deaths by date.  Since the ultimate outcome is death, I start with those.  This time I am using the CDC week of death numbers, not Minnesota’s date of report, because the CDC approach is obviously more accurate.  So I start with the deaths and work backward to hospitalizations and cases, using the current CDC  and/or Minnesota parameters for lags.  You could start with cases for a week and work forward, but because of the median periods to hospitalizations and deaths, and because the only way to get deaths by date of death is the CDC data, it works better to start with deaths and go backward.

The median number of days from symptom development to hospitalization is age dependent.  For 18 to 49 year-olds it is 6 days, with an interquartile range of 3 to 10; for 50 to 64 year-olds it is 6, with a range of 2 to 10 and for those over 64 it is 4 days, with a range of 1 to 9.  65 plus year-olds account for a 42% percent of Minnesota hospitalizations and the 50-64 group is another 25%, so I am going to weight hospitalizations to use 5 days as the lag from symptom development.  The CDC further says that the median days from symptom onset to death is 15 days for 18 to 49 year-olds, with an interquartile range of 9 to 25; 17 days for 50 to 64 year-olds, range of 10 to 26 days, and 13 days for those 65 and over, range of 8 to 21.  Deaths in the 65 and over group are 72% of all deaths in Minnesota, and the 50 to 64s are another 10%, so I am going to weight this to 14 days.   In all cases it looks like the median is probably close to the mean, or average.

Hospitalizations this time are not daily census but new admits by day, which is an improvement.  Cases are from cases by date of specimen collection but because the state in the weekly report now indicates that the median lag from symptom development to positive result by specimen collection date is 2 days (the CDC says it is 3 days nationally, but let’s go with the local number) we need to push the cases period start forward by that additional time to be consistent with the estimates of time to hospitalization and death from symptom onset (without the range, it is hard to know if the median is close to the average, but it is the best we have and probably close).   (In other words, people who had symptom onset in the 7 day period ending 14 days before the date of the death-reporting week end, don’t show up as a case for two more days, on average, in the Minnesota case reporting, so we need to instead end the 7 day case period 12 days before the date of the death-reporting week end.)  The CDC date by death numbers are by week end, so we will use hospitalizations for the 7 day period ending 9 days before the date of the death-reporting week end.  And we will use cases from the 7 day period ending 12 days before that date.

And as before, there is both a cumulative trend line, week by week, for the entire epidemic up to and including the then-current week, and the current trend, which is just the trend for the then-current cohort, each week being a cohort.  I will do and publish a second chart in the next day or so with the testing and cases normalized, as I did in the analysis within the last week on testing and cases, using the average number of tests per week in the last few weeks and multiplying that by the positivity rates for a week.  You can draw your own conclusions from the chart.  Week 1 is March 28, the last week is the one ending October 10.  Remember these are the rates of deaths and hospitalizations for a cohort, which is a set of cases is a seven day period.  Look at the long slow steady tailing off, even with the massive rise in testing and cases.  The death rate in Minnesota is currently under 1%.  If we were detecting all infections it would likely be much lower.

Week EndingDeaths (CDC)Deaths--CumulativeCases--CumulativeCases (Seven Day Period Ending 12 Days Earlier)Hospitalizations (Seven Day Period Ending 9 Days Earlier)Hospitalizations--CumulativeHospitalization Rate--CurrentHospitalization Rate--CumulativeDeath Rate--CurrentDeath Rate--Cumulative
28-Mar1112179174232913.22%16.20%6.32%6.70%
4-Apr25373952167410334.26%26.08%11.57%9.37%
11-Apr458282543012723029.53%27.88%10.47%9.94%
18-Apr79161133450915338330.06%28.71%15.52%12.07%
25-Apr147308201568120358629.81%29.08%21.59%15.29%
2-May1554633213119826184721.79%26.36%12.94%14.41%
9-May16763062493036373122012.29%19.52%5.50%10.08%
16-May142772103174068448166811.01%16.17%3.49%7.48%
23-May164936144724155428209610.30%14.48%3.95%6.47%
30-May1561092190904618515261111.15%13.68%3.38%5.72%
6-Jun151124323662457244930609.82%12.93%3.30%5.25%
13-Jun90133327074341230233628.85%12.42%2.64%4.92%
20-Jun68140129693261915135135.77%11.83%2.60%4.72%
27-Jun541455319402247288380112.82%11.90%2.40%4.56%
4-Jul44149934439249912739285.08%11.41%1.76%4.35%
11-Jul32153137695325626841968.23%11.13%0.98%4.06%
18-Jul38156941095340020444006.00%10.71%1.12%3.82%
25-Jul34160345584448923546355.24%10.17%0.76%3.52%
1-Aug36163950232464827949146.00%9.78%0.77%3.26%
8-Aug45168454976474430652206.45%9.50%0.95%3.06%
15-Aug46173059738476229155116.11%9.23%0.97%2.90%
22-Aug57178764178444030958206.96%9.07%1.28%2.78%
29-Aug36182368638446029461146.59%8.91%0.81%2.66%
5-Sep50187373460482229764116.16%8.73%1.04%2.55%
12-Sep56192978723526325466654.83%8.47%1.06%2.45%
19-Sep47197682479375622068855.86%8.35%1.25%2.40%
26-Sep56203287358487923671214.84%8.15%1.15%2.33%
3-Oct57208994014665636174825.42%7.96%0.86%2.22%
10-Oct402129101289727538678685.31%7.77%0.55%2.10%

 

Join the discussion One Comment

  • Harley says:

    Sure looks like a pretty clear trend line to me. Unfortunately, the public health – government – health consultant industrial complex sees otherwise. Or at least until a few days after the election.

    I’m hoping there is a big November surprise brewing.

Leave a comment