How a change in reporting policy created a dramatic change in counting deaths

Recent Changes in Florida’s Covid-19 Death Reporting

Recently we wrote about the increasing delays in reporting Covid-19 deaths in Florida.

As an example: deaths reported in the two-week period between July 21 and August 3, the 50^th, 75^th, and 95^th percentile of days between date of death and date of report was 5, 10, and 20, respectively.

For the week of August 23 to 29, these figures have increased to 12, 28, and 41 days.

This suggests that the number of deaths that occurred in mid-to-late July (and through August, most likely) was far higher than what “current” reports indicated, complicating inferences about the trends in deaths by day.

After reaching out to the Florida Department of Health about the dramatic change in reporting, they confirmed in an email that the official policy of how they reported deaths changed sometime between June and August.

From March through at least May, as soon as DOH learned that a covid-19 positive person had died from complications of the illness, whether from a hospital, the police, or other authoritative source, they updated the case line data to reflect that information.

Sometime later, the state would receive the official death certificate from the Medical Examiner’s office and THEN add the official date of death to their internal data.  

This time death-by-date-of-death data used to be represented on DOH’s now-bug-ridden dashboard as the number of resident deaths per date by county.

But two changes occurred during the summer:

First, DOH no longer allows users to view deaths by date of death for each county. The data is state-wide only.

Second, they stopped adding deaths when they were notified, and began waiting for the arrival of the death certificate before adding any information about the deaths to their count.

When asked when this change occurred and why, DOH responded, “Unfortunately, I do not have an exact date of the change,” followed by the stock response of “This is an emerging, rapidly evolving situation…”

The timing seems somewhat suspect as the change in reporting appears at the peak of deaths, perhaps to slow the number of deaths being reported and spread them out over a longer span of time. It could simply be a focus on clearing out the large backlog of deaths in July, or that the state’s scantly-staffed offices were overwhelmed. Maybe the change the Medical Examiner’s office made to allow doctors to sign off on death certificates plays a role here. Without answers from DOH, there’s no telling why or how it happened; we can only look at how the change played out in the data.

The figure above shows the cumulative count of deaths reported by date of death. The blue lines are deaths that occurred between July 20 and August 6; the red lines are deaths that occurred August 7 onward (omitting the most recent 10 days). Note that the date labels at the end of each line represent the current count (smoothed) by date of death in descending order from left to right.

We can see that the blue lines follow a relatively similar distribution. The change in reported deaths from one day to the next is gradual – a function of the trend in true deaths by date and the reporting process. In general, the distribution of deaths reported appears consistent for the blue period. Beginning with deaths on August 7, there is a significant difference in the distribution.

First, notice the large drop in reported deaths between 8/6 and 8/7. There is prominent periodicity in the counts by day but the half-cycle that begins on 8/4 and ends on 8/8 has an unexpectedly large drop.

Notice the change in distribution from 8/6 to 8/7. The red period starts flattening much earlier and rather than increasing linearly around day 14 – as happens in the blue period – it flattens altogether. It appears that even a third distinctive cluster may be forming (from 8/15 to 8/18).

Perhaps most important to see is the trajectory of the blue periods, which do not appear to be slowing; there are likely many more deaths to be counted from July and into August. For such a process, we would expect the count to level off and gradually approach the final value which is not observed.  

Some of the differences between these periods may turn out to be a true reduction in the number of deaths, and it is unknown how the volume of deaths affects the reporting distribution.

However, if reporting processes were to remain constant, we would not expect such a dramatic decrease in deaths in subsequent days. The significant and rapid change in distribution suggests a major overhaul in policy from the Florida Department of Health.

Whatever the cause, the effect has been an appreciable drop in recently reported deaths that do not appear to be entirely the result of deaths decreasing in this time. There is also an apparent stalling of deaths being reported for death dates between 8/7 and 8/13.

This information is important to keep in mind while sifting through the ‘deaths by date of death’ misinformation out there (namely, from the Florida Department of Health itself, but also from self-proclaimed pseudo-experts at “business” schools).

There has very clearly been a change in the reporting of deaths and to use old estimates (e.g. 90% of deaths are reported in 21 days) would seem misguided, leading to premature celebration and further poor decision-making by both politicians and the public.