In the coming days, I expect that the American public will see a blitz of (mis)information with the general theme being that coronavirus is not as “deadly” as we think. It is important to understand the inherent scientific biases that underlie studies used to support this argument.

The Los Angeles area prevalence study which I have critiqued both in previous posts as well as in the Los Angeles Times, has now been published in the Journal of the American Medical Association (The JAMA Network). It has undergone peer review and is significantly watered down from the original press release; the authors have had to acknowledge selection bias and the questionable accuracy of their antibody test kits. However, they couldn’t resist one dig saying “fatality rates based on confirmed cases may be higher than rates based on number of infections.”

Be assured that additional studies such as this will be forthcoming, to much fanfare (Dr. John P. A. Ioannidis from Stanford will most likely be at the forefront of press and television interviews). When reading or listening to these reports, it is important to make a distinction between CFR (Case Fatality Rate) and IFR (Infection Fatality Rate). Experts presenting their case that coronavirus isn’t “that bad” will seek to use these interchangeably when, in fact, they are quite different.

Case Fatality Rate (CFR) is the death rate among those with known disease. In other words, these are patients that have been tested, tested positive, and then tragically died as a direct result of their COVID-19 infection. The testing that they underwent is better validated and more accurate than antibody testing (this technology has been used since the 1980s). CFR is then calculated by dividng the number of deaths by the number of cases. Some choose to calculate it by the number of deaths by the number of recovered cases. Worldwide, the CFR is either 6.4% or 16.4% depending on how you calculate it. Given the time lag required to deem a patient ‘recovered’ I prefer the first calculation.

Infection Fatality Rates (IFR) is the death rate among those who test positive for COVID-19 but may not necessarily have had symptoms (asymptomatic), had such mild disease that it did not come to the attention of a medical professional or was mistaken for another illness (e.g. influenza or a common cold). The testing that is performed to discover infections has, thus far, been far less accurate than the gold standard PCR testing, with the primary error being a false positives. The Los Angeles study, for instance, reports test sensitivity of 82.7% meaning that 18% of the positive tests are incorrect. Conversely, the specificity is 99.5% which means that only 0.5% of the negative tests are incorrect. This is a significant issue. Antibody studies will typically claim a IFR of 0.5% or less.

So how do we reconcile a CFR of 16.4% with a IFR of 0.5%? The short answer is that you cannot because the comparison is truly apples and oranges. Case fatality rates based on recovered cases is undoubtedly an overestimate as not all cases are followed to recovery (termed ‘lost to follow-up’). Infection Fatality Rates are undoubtedly an underestimate as they have been based on flawed antibody detection technology, biased samples (only 50% of individuals who consented to antibody testing actually participated in Los Angeles), and lastly due to right censoring.

Wait, right censoring? I know, to throw a statistical concept into the the third to last paragraph of a post is really unfair. But I guarantee that nobody in the press will cover this source of error. Here’s the issue – people who ultimately succumb to their coronavirus infection do so some time after they become infected – sometimes weeks to months afterwards. However, every seroprevalence study simply uses number of deaths the day that the survey was completed. To be truly accurate, these studies need to wait 4-6 weeks to get an accurate count of just how many cumulative deaths have occurred in the community, before they can report the true IFR. The term right censoring comes from the definition of censoring which is “to officially suppress.” The right refers to the X-axis of a plot over time – or in other words these studies have suppressed events that would have occurred later (to the graphical right) in time. (Side note: there are statistical techniques to account for right censoring but the seroprevalence study authors are simply too lazy to do such).

So how should we approach this attempt to frame COVID-19 as “not that bad”? First, view statements by experts like Dr. Ioannidis with skepticism. When you hear that “estimates of infection fatality rates inferred from seroprevalence studies tend to be much lower than original speculations made in the early days of the pandemic”, think about that critically in terms of how that estimate may be biased. Also, look at the language that these experts use – specifically the term “speculation” when in true scientific discourse they should say “estimate” or “point estimate.” Lastly, if you doubt that COVID-19 is really that deadly, I highly recommend reviewing an outside source – specifically the excess death analysis by the Financial Times (link below). Given concerns that reported Covid-19 deaths had not been capturing the true impact of coronavirus, the magazine began tabulating deaths above the historical average. Thus far in 2020, the United States has had 52,300 deaths which are in excess of those expected in a typical year, 24% higher in fact. So, yes, coronavirus really is that deadly – and don’t let anybody (even a Stanford Professor) convince you otherwise.