Content uploaded by Lasse Fridstrøm
Author content
All content in this area was uploaded by Lasse Fridstrøm on May 16, 2020
Content may be subject to copyright.
A surprisingly simple way to assess the
reproduction number
of any national corona epidemic
Lasse Fridstrøm
May 15, 2020
Page
KORMOD
Ridiculously simple model: only information used is the daily incidence
of new positive coronavirus tests.
The prevalence of infectious people is the sum of all new infections (the
incidences) 3 to 9 days back (7-day period).
This is tantamount to assuming a rectangular infection-to-infection time
distribution with mean 6.5 days.
The daily incidence rate = incidence/prevalence = new infections per
infectious person per day
Reproduction number (R) is 7 times daily incidence rate.
Alternatively, gamma distributions with mean 6.5 or 4.8 days are used
for infection-to-infection time, as suggested by the Imperial College in
their March 30 report or by Ferretti et al. (2020), respectively.
2
Page
Rectangular or gamma distributed infection-to-infection time
3
Page
Incidence rate in Norway under three different assumptions
regarding infection-to-infection time distribution
4
Page
Problem: underreporting
Only a small share of infections are discovered by testing. There
is measurement error (underreporting).
But the error affects both numerator and denominator in
incidence-to-prevalence ratio.
If the relative measurement error is constant, the ratio
incidence/prevalence, and hence R, is correctly estimated.
If the relative measurement error changes fast, sizable bias may
arise.
Under normal conditions, bias will be moderate.
5
Page
Reporting probability: four numerical examples
6
Page
Sensitivity analysis: true incidence rate by degree of reporting
7
Page
R estimates for
UK, USA, Sweden, Norway, Germany
Data source: Miscellaneous
8
Page
R estimates for
USA, India, Brazil, Russia, Turkey, Chile
Data source: Covid-19 Dashboard of Johns Hopkins University
9
Thanks for your attention!
Stay safe!