Odd Numbers: How Many Missing Covid- 19 Deaths?

New York Times, 5/5/2020

I’m a bit of a data geek, and since I’ve been stuck at home I have been poring over the published statistics on Covid-19. One thing that particularly struck me as strange was the surprisingly low number of Covid-19 deaths per confirmed case in Texas and Florida, compared with other states with large numbers of cases. What might explain that?

There are two measures of the deadliness of a disease. The disease mortality rate is the number of deaths divided by the actual number of infections. The observed case-fatality ratio is the number of deaths attributed to the disease divided by the number of confirmed cases (times 100, to express as percentage).

To compute the former, you need to know the true number (or at least a credible estimate) of infections, and that is still unknown. So at this point epidemiologists can only make informed guesses about the disease mortality rate. It is generally believed that the true number of Covid-19 infections is far larger than the number of officially confirmed cases.

So we are left with the observed case-fatality ratio, for which we have actual numbers provided by state health departments gathered from data provided by hospitals and other health officials. Here’s what these numbers show (based on data published in the New York Times as of 5/4) for states with the highest numbers of cases:

• New York           7.6  (i.e., 7.6% of confirmed Covid-19 cases ended in death)
• New Jersey         6.1
• Massachusetts   7.4
• Illinois                 4.1
• California           4.1
• Pennsylvania     5.4
• Michigan            9.4
• Florida                3.7
• Louisiana           6.7
• Connecticut       8.5
• Texas                   2.7

What would account for these rather large variations? Why would Covid-19 patients in Michigan be dying at more than 3 times the rate in Texas?

But there are problems with these numbers too.

The number of confirmed cases (the denominator in the ratio) depends to a large extent on how widely the population is being tested. The more you test, the more cases you will find–particularly cases that produce few or no symptoms. Testing in the US so far has been woefully inadequate in general, but there are wide differences in the rate of testing between states. Among states with large numbers of cases, New York has tested the most (as of 5/4) at 5.0% of the population. Massachusetts is at 4.5%, Louisiana at 3.8%, and New Jersey at 3.1%. At the lower end are Texas, which has tested only 1.3% of its population, California and Pennsylvania at 1.8%, and Florida at 1.9%.

One might think that we could be pretty sure that at least the number of deaths (the numerator in the ratio) would be accurate. After all, there is a body which can be counted and for which a death certificate must be issued listing cause of death. But the statistics depend precisely on what is listed as cause of death (COD), and as with the living, the dead are often not tested to see if the virus was present. If the COD is entered as, say, pneumonia or heart failure, that death will not show up in the Covid-19 figures.

Apparently, this has been a significant problem. New York reviewed causes of deaths that were not initially reported as Covid-19 since the start of the pandemic, and subsequently added more than 5,000 to the state’s official death toll from the disease. The Washington Post reported that the US recorded an estimated 37,100 excess deaths in the early stages of the pandemic in March and the first two weeks of April, nearly 13,500 more than were attributed to Covid-19 for that same period. The New York Times just published another analysis of “excess deaths” based on CDC data, but the data for most states only goes through early April and for some (like Texas) only through March. The Washington Post also reported that in Alabama one out of ten patients that died with Covid-19 were not listed as dying of Covid-19. How widely this is still happening is unclear.

Could some of this misreporting be deliberate? The Tampa Bay Times reported that Florida state officials “have stopped releasing the list of coronavirus deaths being compiled by Florida’s medical examiners, which has at times shown a higher death toll than the state’s published count. The list had previously been released in real time by the state Medical Examiners Commission. But earlier this month, after the Tampa Bay Times reported that the medical examiners’ death count was 10 percent higher than the figure released by the Florida Department of Health,” state officials began withholding this information.

Why would some officials want to low-ball the death count? Perhaps, like Trump, they are anxious to re-open their states and localities for business, and bigger death numbers make things look bad. Of course, there might be other more benign explanations.

But back to the observed case-fatality ratio, are there other things that might explain the rather wide differences between the states? There could be differences in the prevalence of underlying conditions or “co-morbidities” such as obesity, diabetes, hypertension, COPD, etc. There could be differences in degree of access to health care and insurance coverage. If certain states are harder hit, it may be more difficult to get admitted to a hospital there. It could be a function of where a state is on the Covid-19 timeline. Deaths typically lag behind a surge in Covid-19 cases, and some states in the middle of the country did not see cases emerge until weeks after states on the east and west coasts.

So in the light of all of these factors, why would the Covid-19 death toll in Texas be so much lower? It can’t be because health coverage in Texas is so great. In fact, according the Census Bureau, Texas has the highest percentage of uninsured residents in the entire country at 17.7%, which is worse than Mississippi. (Florida isn’t far behind at 13%.) Texans have a lower life expectancy (78.8 years) than New Yorkers (81.0 years) and residents of other northeastern states hit hardest by Covid-19. The median age in Texas is among the youngest (34.8 years), but not that much lower than neighboring Louisiana (37.2 years), which has much higher reported mortality from Covid-19. But then again Florida also has a low case-fatality ratio and a much older population (median 42.2 years). Perhaps the most plausible benign explanation is that Texas got hit later than other states, and is not as far along on the curve. With Texas among the states most eager to end social distancing restrictions, the numbers could soon change drastically.

Would more extensive testing make a difference in this measure of mortality? Perhaps, though if the number of confirmed cases increased as a result, it would actually lower the observed case-fatality ratio, unless there were also a comprehensive review of causes of death during the pandemic that raised the fatality number as well. But just looking at states with similar levels of testing and similar case numbers such as Illinois, California, Pennsylvania, and Michigan, it’s difficult to explain why Texas and Florida show much lower numbers of Covid-19 deaths.

It’s probably important to note that there are also major differences in the case-fatality ratios of different countries. Among the countries with major outbreaks, Belgium, the UK, and France have ratios of around 15%, which is much higher than the US as a whole at 5.8%. Again, it’s difficult to account for the differences, and harder to know if the numbers are actually counting the same things in the same way.

As I write this, the official US death toll from Covid-19 has just passed 71,000. The true number is almost certainly significantly higher. Let us hope that the real numbers are not being fudged to make the political optics better.