Science - USA (2022-05-06)

GRAPHIC: KELLIE HOLOSKI/

SCIENCE

science.org SCIENCE

facts’’ about transmission and in-
tervention. These are qualitative
“rules of thumb” that flow from
the quantitative output of mech-
anistic models (see the figure). It
is these qualitative understand-
ings that drive most of the scien-
tific, political, and social conver-
sation around infectious diseases
in moments of deep uncertainty.
When these accurately describe
the state of the world, such styl-
ized facts are invaluable. When
they do not, there are serious
consequences for public health
and the public’s confidence in it.
In the earliest moments of
the pandemic, when data were
sparse, modelers reached back
to the 2003 SARS outbreak to
understand whether it might be
feasible to stem the global spread
of this new—but related—patho-
gen ( 3 ). As data from Wuhan and
other early outbreaks, such as
on the Diamond Princess cruise
ship ( 4 ), became available, ana-
lysts began to estimate the value
of the basic reproduction rate R0
(the number of cases generated by one case in
a susceptible population) and its responsive-
ness to intervention through the time-vary-
ing reproduction rate, R(t). One such analysis
found that transmission rates in Wuhan de-
clined over the course of early 2020, with a
decisive dip that brought R(t) below the criti-
cal threshold of 1 after the institution of mo-
bility restrictions in February 2020 ( 5 ). This
and other analyses from the same period ( 6 )
suggested that lockdowns and other drastic
contact-limiting interventions could be ef-
fective in stemming the tide of transmission.
These insights shaped much of the global re-
sponse to SARS-CoV-2 in 2020.
Whereas early analyses were characterized
by a paucity of data, more information has
become available with each wave of the pan-
demic to suggest what might happen next.
For example, in a recent meta-analysis, the
COVID-19 Forecasting Team used data from
more than 3000 serological surveys con-
ducted in 76 countries to characterize age-
and region-specific infection fatality rates
(IFRs) associated with SARS-CoV-2 infection
( 7 ). This type of “historical” data is critical for
parameterizing models that can meaning-
fully capture spatial and demographic varia-
tion in the risks of infection and death. Such
information is critically important for cap-
turing the complex landscape of SARS-CoV-2
population immunity as the slow transition
to endemicity continues ( 8 ).
As the amount of SARS-CoV-2 data has in-
creased, newly entrenched stylized facts that

shape policy and behavior may make the re-

sponse to shifts in the pandemic less nimble.

These can ossify into “common sense” that

provides a sense of control and predictabil-

ity while actually crowding out important

new information ( 9 ). For example, a con-

sensus that emerged from years of research

and discussion that respiratory viruses did

not reliably transmit over distances greater

than the 6-foot buffer used in social distanc-

ing guidelines was used to dismiss compel-

ling observational evidence of an important

role for longer-range airborne transmission

of SARS-CoV-2 ( 10 ). As the volume of stylized

facts about SARS-CoV-2 transmission rates,

vaccine effectiveness, and IFRs continues to

grow, the risk that new pieces of conventional

wisdom will become disconnected from fast-

changing realities increases.

Mental models also fail when they narrow

the set of questions modelers think to ask:

The role of racial discrimination and eco-

nomic exploitation in amplifying the toll of

COVID-19 among those who are financially

and socially marginalized, and among people

of color subjected to racial discrimination,

was clear from the earliest moments of the

pandemic ( 11 ). But the mechanistic models

used as tools of rapid response at this time

were focused largely on the pathogen ( how

infectious is it?) and not the brutally unequal

human systems governing who is exposed

and when ( 12 ). This has sparked a wave of

new models that directly incorporate the

mechanistic drivers of social inequality in in-

fection and death ( 13 ). These al-

low quantification of the impact

of social and economic policies

on the burden of disease ( 14 ).

In the future, the rapid deploy-

ment of high-quality models that

adequately account for the joint

social and biological drivers of

infection will be critical to an-

ticipating and addressing these

inequities. This requires an im-

mediate and concerted effort to

increase socioeconomic, racial,

and geographic diversity in the

modeling workforce.

Mechanistic models have

also played a contentious role

in forecasting disease outcomes

over the short to medium term.

Although the predictive qual-

ity of disease forecasting has

increased in recent years, these

models necessarily reflect his-

tory-based assumptions about

future transmission. As a result,

even “ensemble” forecasting

approaches that use multiple

models with differing assump-

tions are often blindsided by

rare, highly consequential events such as the

emergence of SARS-CoV-2 variants with dif-

fering transmission characteristics, or rapid

shifts in human behaviors, such as vaccina-

tion, masking, reduced air travel, and social

distancing ( 15 ). Splitting the difference be-

tween using as much information as possible

to suggest what might happen next, while

acknowledging and quantifying the myriad

deep uncertainties associated with a global

pandemic of an evolving pathogen, must be a

top priority of preparedness for future global

infectious disease emergencies. j

REFERENCES AND NOTES

1. J. R. C. Pulliam et al., Science 376 , eabn4947 (2022).


2. J. Lessler, D. A. T. Cummings, Am. J. Epidemiol. 183 , 415
   (2016).


3. C. Fraser, S. Riley, R. M. Anderson, N. M. Ferguson, Proc.
   Natl. Acad. Sci. U.S.A. 101 , 6146 (2004).


4. J. Rocklöv, H. Sjödin, A. Wilder-Smith, J. Travel Med. 27 ,
   taaa030 (2020).


5. A. J. Kucharski et al., Lancet Infect. Dis. 20 , 553 (2020).


6. S. Flaxman et al., Nature 584 , 257 (2020).


7. COVID-19 Forecasting Team, Lancet 399 , 1469 (2022).


8. J. S. Lavine, O. N. Bjornstad, R. Antia, Science 371 , 74 1
   (2021).


9. T. Greenhalgh, Interface Focus 11 , 20210017 (2021).


10. L. Hamner et al., MMWR Morb. Mortal. Wkly. Rep. 69 , 606
    (2020).


11. W. N. Laster Pirtle, Health Educ. Behav. 47 , 504 (2020).


12. J. Zelner et al., PLOS Comput. Biol. 18 , e1009795 (2022).


13. K. C. Ma et al., eLife 10 , e66601 (2021).


14. A. Nande et al., Nat. Commun. 12 , 2274 (2021).


15. E. Y. Cramer et al., medRxiv
    10.1101/2021.02.03.21250974 (2021).

ACKNOWLEDGMENTS

The authors thank K. Broen, R. Naraharisetti, and R. Trangucci

for feedback on this manuscript.

10.1126/science.abp9498

INSIGHTS | PERSPECTIVES

Previous

epidemics

Lessons, data,

models

Epidemiological

surveillance data

Cases, hospitalizations,

genetic data

Social and

biological information

Survey data, severity

information, mobility

Models

Stylized facts (Qualitative)

Rules of thumb

Overall patterns

“Flatten the curve”

Predictions (Quantitative)

Fo re ca sts

Counterfactual

simulations

Spatial distributions

Seasonal patterns

Exponential growth

Surge

Basic reproduction number, R 0

Modeling epidemics

Epidemiological models translate quantitative data and assumptions about the

nature of infectious diseases into both qualitative “stylized facts” and quantitative

predictions that have proven to be critical drivers of the public, political, and scientific

conversation around severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2).

580 6 MAY 2022 • VOL 376 ISSUE 6593