Lack of reliable data, the other "Pandemic"

To understand the extent of the COVID-19 pandemic, it is helpful to look at how existing models work to estimate the burden of other infectious diseases. For example, the Centers for Disease Control and Prevention (CDC) in the United States can’t test everyone for influenza, so instead it leverages estimates and models. More specifically every year, the CDC quantifies the burden of influenza in the United States, leveraging a combination of weekly influenza surveillance data and a mathematical model to estimate the numbers of influenza illnesses, medical visits, hospitalizations, and deaths.

During the 2018-2019 influenza season, labs participating in the Virus Surveillance System, which includes both clinical and public health laboratories throughout the United States, contributed to virologic surveillance for influenza. During that season those laboratories tested a total of 1.23 million specimens. The CDC statistics show that approximately 220,000 of the tested specimens -17.8% - tested positive for influenza A or B.

Based on the above data and their internal model the CDC estimated the symptomatic influenza cases in the season to be in excess of 35 million. In other words, for every individual who tested positive for influenza there are approximately 160 individuals with influenza not being tested. Furthermore, the CDC estimated that the death rate for influenza ranged between 0.1-0.15%, approximately 34,000 to 52,000 deaths in the 2018-2019 season. 

When we look at COVID-19 we need to consider that the number of cases reported today, is based almost exclusively on the number of positive test results. Since COVID-19 has been proven to have a higher contagion ratio compared to influenza, it is reasonable to assume that in a pandemic environment the true number of cases could be several orders of magnitude higher than reported. Considering that the World Health Organization (WHO) is estimating the current death rate of COVID-19 to be 2-4% and it bases that estimate on the reported cases, we can infer that ratio to be overinflated. In fact, in the current phase, testing is not adopted enough to be comprehensive and it is characterized from higher-than-usual adverse selection.

In the small town of Vó in northern Italy, the mayor, directed the health authorities to screen the entire population of 3,305 for COVID-19 early in the outbreak and determined that 2.7% of the population tested positive to the virus. Vó was not in the epicenter of the outbreak so it is safe to assume that statistically, the penetration of the virus may have been lower than in cities that were highly impacted. Interesting to note: between Veneto and Lombardia (where Vó is located), and the adjacent region where the epicenter of the outbreak was, the total population is 14 million; considering the timing of the testing in Vó, it would be safe to assume that if the Vó infected percentage is applied to the two regions, we would be looking at approximately 380,000 infected individuals. This number rises to approximately 750,000 infected individuals considering that north of the Italian peninsula is where 90% of the reported COVID-19 cases have been found in Italy, thus far. As of March 27, in all of Italy, 80,589 COVID-19 cases have been reported as a result of diagnostic testing, with 8,215 deaths confirmed. 

The reported mortality rate, estimated at ~10%, seems overestimated due to reported cases being exclusively linked to tested cases, as opposed to a more credible estimated rate of infection, which would indicate a mortality rate at least ten times lower than reported estimates. 

To corroborate this thesis, deCode Genetics has run a study in Iceland that began on March 13th, 2020, recruiting both symptomatic and asymptomatic volunteers. The results of the first 5,571 diagnostic tests yielded 48 positive samples (0.86%), indicating that the prevalence of the virus is modest among the general population. Since the first confirmed case of COVID-19 in Iceland on February 28, a total of 890 cases have been reported, resulting in two deaths and 12 hospitalizations.  In terms of tests per one million inhabitants, Iceland has now tested 26,762, which is the highest proportion we are aware of in the world. 

If we consider the incidence of COVID-19 in a country like Iceland being at 0.86% (using the above as a proxy), and 2.7% in a small town like Vó in Italy, there is reason to believe that the rate of infection in the United States and worldwide is significantly higher than what projected today. 

In fact, considering the population of the United States and applying the conservative Iceland statistic, we could be in the presence of at least 2-3 million cases of COVID-19, of which many asymptomatic or with mild symptoms not requiring hospitalization.

The above does not make COVID-19 a non-issue; on the contrary, the consequences of the pandemic on the health systems are evident and reflected in the increasingly serious lack of intensive care beds and ventilators. As a result of this, some of the restrictive measures are needed to enable the system to cope with the ongoing and oncoming higher hospitalization rate. But it also leads to the conclusion that one of the main challenge we are currently facing is not only the virus itself, but the lack of capillarity in testing and the statistical study to assess the real infection rate, apparently higher than reported and the mortality rates that would result being significantly lower than reported. Ultimately, the limitation in data science and statistical epidemiology modelling is a much easier problem to solve and I encourage the many scientists working on COVID-19 to help on this matter. The consequences of not solving it now is an even higher price for our society and not due to COVID-19 itself but to an even more relevant problem, the lack of accurate data and models showing us where the real problem is.  

1.     Reed C, Chaves SS, Daily Kirley P, Emerson R, Aragon D, Hancock EB, et al. Estimating influenza disease burden from population-based surveillance data in the United States. PLoS One. 2015;10(3):e0118369.

2.     Rolfes, MA, Foppa, IM, Garg, S, et al. Annual estimates of the burden of seasonal influenza in the United States: A tool for strengthening influenza surveillance and preparedness. Influenza Other Respi Viruses. 2018; 12: 132– 137. https://doi.org/10.1111/irv.12486

3.     Centers for Disease Control and Prevention. Estimated influenza illnesses and hospitalizations averted by influenza vaccination – United States, 2012-13 influenza season. MMWR Morb Mortal Wkly Rep. 2013 Dec 13;62(49):997-1000.

4.     Reed C, Kim IK, Singleton JA, Chaves SS, Flannery B, Finelli L, et al. Estimated influenza illnesses and hospitalizations averted by vaccination–United States, 2013-14 influenza season. MMWR Morb Mortal Wkly Rep. 2014 Dec 12;63(49):1151-4.

5.     Jester B, Schwerzmann J, Mustaquim D, Aden T, Brammer L, Humes R, et al. Mapping of the US Domestic Influenza Virologic Surveillance Landscape. Emerg Infect Dis. 2018;24(7):1300-1306. https://dx.doi.org/10.3201/eid2407.180028

6.     https://nordiclifescience.org/covid-19-first-results-of-the-voluntary-screening-on-iceland/

7.     https://www.wsj.com/articles/is-the-coronavirus-as-deadly-as-they-say-11585088464

8.     https://www.cdc.gov/flu/about/burden/2018-2019.html