BREAKING: Harvard Epidemiologists Find that Extreme Social Distancing Could Lead to More COVID-19 Deaths than the DO NOTHING Alternative

Harvard University epidemiologists published a fascinating article in Science magazine yesterday.  I am going to delay posting a deeper dive into the Imperial College and University of Washington models so that I can devote my time today to getting the word out about the implications of this new Harvard model.

Here is what the Harvard researchers did:

• They ‘calibrated’ their model using historical data (2014-2019) from two strains of the common cold in the U.S.  This makes a lot of sense, because the common cold is a type of coronavirus, so it is reasonable to expect similarities.
• They simulated outbreaks (i.e. infection peaks) based on a range of ‘effectiveness’ of various non-pharmaceutical interventions (NPIs).  Rather than complicate the model by trying to estimate the effectiveness of specific NPIs (such as school closures, stopping mass gatherings, etc.), they characterize the sum-total of all NPIs in effect at any given time.  The ranges of ‘effectiveness’ they simulated were 0% (do nothing), and 20%, 40%, and 60% (which I am referring to as mild, moderate, and high levels of social distancing, respectively).  For comparison purposes, they point out that attenuation of normal cold and flu viruses in the summer in New York is 40% (i.e. the spread of such viruses naturally declines during the summer months -- higher temperatures, higher humidity, people spending more time outdoors; the degree of attenuation is different for different geographic locations, but in New York, this is equivalent to a reduction in R0 of 40%).  I would guess that our current levels of social distancing fall into the extreme category, considerably greater than 60%; however, that is just a guess on my part.  I would love to see some scientific evidence tying specific NPIs to a resulting decrease in R0.

Here is what they found (note that I am using the word ‘deaths’ in lieu of ‘critical infections’ because the two should be directly and proportionally related):

• Highly effective social distancing merely pushes the number of peak deaths to a later time.
• If that later time coincides with the natural seasonal increase in the fall / winter months, the number of peak deaths might end up larger than the ‘do nothing’ alternative (see Figures 5C and 5H).
• In the absence of seasonality, the net effects of highly-effective social distancing on total deaths (compared to the ‘do nothing’ alternative) are negligible (see Figures 4F, 4G, 4H, and 4I), and the more extreme the levels of social distancing, the closer the final outcome is to the ‘do nothing’ alternative.  Relating this to the figures, note that the green line (high social distancing) is closest to the black line (do nothing), followed by the blue line (moderate social distancing), then the red line (mild social distancing).

• In the presence of seasonality, even if the delayed peaks are shorter, the cumulative number of deaths might still be higher (see Figure 4G).  This is because the area under the curve is greater, even though the peak of the curve is shorter.
• In the presence of seasonality, in 1 out of 5 scenarios, implementing high social distancing (i.e. reducing the infection rate by 60% or more) for 4 weeks results in the SAME number of deaths as the DO NOTHING alternative (see Figures 5A and 5F).
• In the presence of seasonality, in 3 out of 5 scenarios, implementing high social distancing for 8 weeks, 12 weeks, and 20 weeks results in MORE deaths than the DO NOTHING alternative (see Figures 5B, 5G, 5C, 5H, 5D, and 5I).
• In order to adequately respond to this virus, “Longitudinal serological studies are urgently needed to determine the extent and duration of immunity to SARS-CoV-2” (p. 1).  In other words, as I have been saying for almost a month, we cannot formulate truly reliable models until we have much better information regarding the total number of people who have already been exposed to the virus (and have developed antibodies), which we can only determine by serologic testing.

There’s more to say about the Harvard model, but I will leave it at that, for now.

The big takeaway (for me) is that governments need to reopen businesses and relax social distancing restrictions IMMEDIATELY or they face the very real likelihood that they will end up causing MORE total COVID-19 deaths compared to the DO NOTHING (from the beginning) alternative.  Get the word out!

NOTE:  The above takeaway is mine (not the Harvard researchers).  I doubt they would make such a bold statement, but it is perfectly in line with what their results tell us.  After all, they clearly stated that “Preventing widespread infection during the summer can flatten and prolong the epidemic but can also lead to a high density of susceptible individuals who could become infected in an intense autumn wave” (italics added, p. 16).  It also bears noting that the Harvard researchers provide a separate trigger-based analysis (similar to the Imperial College model except Harvard relies upon cases as their on/off trigger whereas Imperial College uses critical cases).  I will try to incorporate a discussion of that analysis in a future post.

Below are the relevant figures.  Here is the legend:

• For Prevalence Charts (A, B, C, D): Solid = infections; Dashed = critical infections (for seasonal charts, i.e. Figure 5, note that critical infections are nearly identical to infections, but with a slight lag)
• For Cumulative Infections Charts (F, G, H, I): Solid = cumulative infections (for seasonal charts, i.e. Figure 5, note that cumulative critical infections will be nearly identical to cumulative infections, but with a slight lag; cumulative deaths should be directly and proportionally related to cumulative critical infections)
• Blue Window = Portion of time all relevant NPIs are in effect
• Black = Do Nothing (i.e. no NPIs)
• Red = 20% Reduction in R0 (i.e. mild social distancing)
• Blue = 40% Reduction in R0 (i.e. moderate social distancing)
• Green = 60% Reduction in R0 (i.e. high social distancing)

Here are the charts with no seasonal considerations (Figure 4):

FIGURE 4 (A, B, C, D, F, G, H, I)

Here are the charts considering seasonal effects (Figure 5):

FIGURES 5A, 5F

FIGURES 5B, 5G

FIGURES 5C, 5H

FIGURES 5D, 5I

Steve Trost is Associate Director of the Institute for the Study of Free Enterprise and can be contacted at trost@okstate.edu.  He has a bachelor’s degree in engineering from MIT, a master’s degree and PhD in engineering from Oklahoma State University and a PhD in entrepreneurship (also from OSU).

Follow Dr. Trost on twitter: @TrostParadox

Disclaimer: All comments, observations, and statements presented herein represent the opinions of the author and in no way reflect the views of Oklahoma State University or the Institute for the Study of Free Enterprise.