Update May 9th: Archiving this document, it’s had no updates since end of March.
The purpose of this document is to collect parameters from expert analysis that can be fed into rough epidemiological models of SARS-CoV-2.
Update April 4th: I haven’t been updating this doc very frequently; there seem to be plenty of new (& better?) sources that are not included here.
Table of Contents
Growth Rate 2
Growth rate 2
R0 2
Serial Interval 3
R0-SI pairs 3
Infectious period 4
Incubation period (time before symptoms) 4
Latent period (time before infectiousness) 4
Doubling time 5
Severity 5
Case Fatality Rate (CFR) 6
CFR by comorbidity 7
Infection Fatality Rate (IFR) 7
Detection rate: Documented vs. undocumented/asymptomatic cases 8
Import rate 9
Time periods 9
First cases (date, amount) 9
Time until going to hospital 9
Time until severe/critical 10
Hospital stay 10
ICU stay 10
Time to death 11
Attack rate / Infection prevalence 11
Neurological effects 12
Growth Models 12
Growth BOTECs by various folks 12
Note on Growth Rates and R0 12
What are the models most sensitive to? 13
Growth Rate
Growth rate
* Note on growth rates & R0
* Ben Phillip’s model has updated growth rates & doubling times by country
R0
Note: R0 depends on a lot of things, especially mitigation measures and estimates of how much and how long people are infectious.
* Flaxman et al (Imperial, 29 Mar) Updated Rt estimates for various European countries
* “Across all countries we find current estimates of Rt to range from a posterior mean of 0.97 [0.14-2.14] for Norway to a posterior mean of 2.64 [1.40-4.18] for Sweden, with an average of 1.43 across the 11 country posterior means, a 64% reduction compared to the pre-intervention values”
* Abbott et al has nice (updated 20 Mar) graphs tracking effective R0 in various countries, averaging around 2-2.5
* WHO-China Joint Mission report says 2-2.5 in absence of interventions
* Liu, Gayle, et al: 3.28, IQR 2.12-4.44, median 2.79
* Analyzed 12 studies from 1 Jan to 7 Feb, took median
* Rocklöv et al (Diamond Princess, which has a population density 4x Wuhan)
* R0 14.8 initially and then declined to a stable 1.78
* Zhang, Diao, Wu (Diamond Princess) 2.88. Has some model of mitigation effects on R0.
* Klausner et al (Israel) - Israel took strong measures initially, but religious holiday (Purim) put them on exponential curve, such that R0 as of March 20 was 2.1
* Used SI distributions from Nishiura, Tapiwa, and Zhao papers.
* Jung et al: Modeled two scenarios using Chinese data through Jan 24. R0 of 2.1 and 3.2 respectively. Range of 1.6–4.2.
* Rovetta and Bhagavathula (Italy, looking at cases March 1-14) R0 of 3.51. Ended up underestimating growth - KS thinks because their initial case estimates were too low.
* Du et al Looked at chinese infection pairs, got a serial interval of 3.96 and used Li et al’s growth rate to calculate an R0 of 1.32
* Tindale et al (6 Mar): 1.97 in Singapore, 1.87 in Tianjin
* Shim et al (Korea): R0 of 1.5
* Li, Pei, Chen et al (16 Mar, China 10-23 Jan) R0 of 2.38
* Zhang et al: Basic R0 of 3.6 in early stages. Models R0 reduction over time due to Chinese interventions
* Mizumoto et al (China): R0 of 5.2 before Jan 23 interventions, 0.58 after.
* Sanche et al (11 Feb) R0 of 4.7-6.6
* Liu, Hu, et al (13 Feb) R0 of 4.5 for mainland China up to Feb 7
* R0 of 6.9 before Jan 9, R0 peak of 8.8
* Tuite & Fisman (5 Feb): R0 of 2.3
* Hermanowitz et al (4 Feb): R0 of 2.4-2.5
* Wu et al (31 Jan, Wuhan) R0 of 2.68
* Li et al (29 Jan): R0 of 2.2
* Read et al (27 Jan): R0 of 3.11
* Liu, Hu, Kang et al (26 Jan): R0 of 2.9
Serial Interval
Seems more likely that SI calculated from pairs is accurate, vs. population-level data
Useful list of sources & discussion
* Du et al (13 Mar, 468 Chinese pairs): 3.96 days, SD 4.7-5 days. Negative serial intervals “suggest the possibility of transmission from asymptomatic or mildly symptomatic cases”
* Tapiwa et al (8 Mar) 5.2 days Singapore, 3.95 Tianjin
* Nishiura et al (27 Feb, 28 & 18 pairs, public data). 28 pairs = 4.0 median SI ±2.9SD. 18 high confidence pairs = 4.6 median SI ±2.3SD
* Zhao et al (25 Feb, Hong Kong, 21 chains, 12 pairs). 4.4 Mean 3.0 SD
* Tindale et al (6 Mar): mean 4.56 days Singapore, 4.22 days Tianjin
* Bi et al (4 Mar, N=391, Shenzhen cases): 6.3 days, SD 4.2 days. Note Figure 2(B)
* Tuite & Fisman (5 Feb): 7 days
* Ping et al (6 Mar, N=57) 6.37 mean ±4.15 SD
* You et al (17 Feb, 71 chains) 4.41 mean ±3.17SD
* Li et al (29 Jan): 7.5 days
* R0-SI pairs
It seems that when papers are looking at population-level data, they often derive R0 from SI [citation needed], thus it doesn’t make sense to average R0 & SI independently, unless you are including the R0-SI pairs from each paper. An attempt to do that below:
* Du et al (Cases outside Hubei) (Uses growth rate from Li et al to derive R0)
* SI 3.96
* R0 1.32
* Li et al (Wuhan, prior to Jan 22)
* SI 7.5
* R0 2.2
* Bi et al (Shenzhen)
* SI 6.3
* R0 0.4
* Zhang, Diao, Wu (Diamond Princess)
* SI 7.5
* R0 2.88
* Serial interval distribution is required for R0 estimation, and there was insufficient information about cluster cases for serial interval estimation. Therefore, we assumed the serial interval of COVID-19 on the ship was equal to that of COVID-19 in Wuhan, China, with a mean of 7.5 days and a standard deviation of 3.4 days (Li et al., 2020)
* Zhuang et al (Italy, Korea). Assumes gamma-distributed SI of 4.5 with 3.1 SD, and one seed infection.
* R0 of 2.6 in Korea (w/ data from 20 Jan - 1 Mar, assuming 31 Jan start)
* R0 of 2.6 in Italy (w/ data from 5 Feb - 5 Mar, assuming 5 Feb start)
Infectious period
* You et al (17 Fe