Professional Documents
Culture Documents
net/publication/339915690
CITATIONS READS
10 12,248
2 authors:
Some of the authors of this publication are also working on these related projects:
Understanding the defects in functioning of the calcium channel that lead to diabetes View project
All content following this page was uploaded by Morten Gram Pedersen on 20 April 2020.
* pedersen@dei.unipd.it
Abstract
The COVID-19 disease caused by the coronavirus SARS-nCoV2 is spreading rapidly across the globe since
its outbreak in China. Italy was the first seriously affected Western country, and the first to implement
drastic measures in an attempt to contain the disease. It is therefore necessary to analyze the Italian data
carefully, and in particular to investigate if and how the limitations in societal activities affect the disease
dynamics. We propose to model COVID-19 dynamics with a SIQR (susceptible – infectious – quarantined
– recovered) model, rather than a SIR model, since the confirmed positive cases correspond only to a
fraction of the total infectious population, and these individuals are isolated and thus do not transmit the
disease. We estimate model parameters as far as possible by fitting model expressions to official Italian
data. We predict that at the beginning of the epidemic the number of unidentified SARS-nCov2-positive
individuals was ∼10 times the number of confirmed cases. Model simulations indicate that the mild
restrictions imposed in Italy during the first two weeks of the outbreak have had negligible effects on
the disease dynamics, but later more drastic measures have managed to reduce the basic reproduction
number R0 below 1. We finally simulate how the disease will evolve when current limitations are lifted
under various assumptions on the effect of relaxation of the restrictions.
Author summary
The COVID-19 pandemic is currently being contained with tight restrictions imposed by the governments
in most Western societies, and the current question is how to relax these limitations without allowing the
epidemic to reemerge. Italy was the first seriously affected Western country, and the first to implement
such drastic measures. We analyze the Italian data with a SIQR (susceptible – infectious – quarantined
– recovered) model in order to quantify how limitations in societal activities has affected the disease
dynamics in Italy. We find that the mild restrictions imposed in Italy during the first two weeks of
the outbreak have had negligible effects on the disease dynamics, but later more drastic measures have
managed to reduce the basic reproduction number R0 below 1. We finally simulate how the disease will
evolve when current limitations are lifted under various assumptions on the effect of relaxation of the
restrictions, and find that the current situation is very fragile since even modest increases in the infection
rate will permit the disease to resurge.
Introduction 1
The COVID-19 disease due to the SARS-nCov2 coronavirus is spreading rapidly across the globe since 2
its outbreak in China, and was declared a pandemic by the WHO on March 11, 2020. After the first 3
severe patient was brought to the hospital of Codogno, Italy on February 20, 2020, and subsequently 4
tested positive for SARS-nCov2, a rapidly increasing number of patients have been identified, initially 5
1
in Northern Italy and later in the rest of the country and Europe. Italy is the most affected European 6
country, with more than 145.000 confirmed cases and nearly 19.000 COVID-19 related deaths, and was 7
the first to implement drastic measures in an attempt to contain the disease. Analyzing the Italian data 8
carefully may therefore provide important insights into the epidemiology of COVID-19, and in particular 9
to investigate if and how the limitations in activities affect the disease dynamics. 10
Simple mathematical models of infectious diseases are useful for providing insight into disease dynamics, 11
and compared to more complex models, can be fitted to data with a minimum number of assumptions on 12
model parameters [1, 2]. However, even simple models should respect that nature of the data. There is 13
thus a compromise between using a parsimonious model but sufficiently complex to be based on correct 14
underlying assumptions. 15
In our setting, to fit the data on identified SARS-nCov2 positive cases, a SIQR [3] is appropriate. In 16
this model, infected individuals may be isolated, entering the “quarantined” subpopulation Q, so that 17
these individuals no longer transmit the disease. Since Italian positive cases have been put in isolation 18
(in hospitals or at home) immediately, the revealed data of active cases thus correspond to the number of 19
individuals in state Q. It would not be correct to confront the infectious subpopulation of a SIR/SEIR 20
(susceptible – infectious – [exposed] – recovered) model with the recorded data, since it is unknown how 21
many infectious but undetected individuals are circulating in the population. As will be clear in the 22
following, besides being conceptually clearer, the explicit quantification of the quarantined individuals 23
will allow us to provide estimates of the infectious cases at the very beginning of the epidemic outbreak 24
as well as of the number of undetected infectious individuals during the growth phase of the epidemic. 25
Methods 26
We use a SIQR model [3] to describe COVID-19 dynamics in Italy. The model equations are 27
dS
= −βSI/N, (1)
dt
dI
= βSI/N − (α + η)I, (2)
dt
dQ
= ηI − γQ, (3)
dt
dR
= γQ, (4)
dt
where R models SARS-nCov2-positive, isolated individuals that recover or die from the disease. We do 28
not explicitly model the number of recovered or deceased non-identified COVID patients, but only the 29
rate α with which these patient become non-infectious. Further, N is the total number of individuals in 30
the population, assumed constant since we are studying the early phase of the epidemic. 31
Since there is evidence that the SARS-nCov2 virus can be transmitted in the absence of symptoms [4,5], 32
we do not include an explicit exposed-but-noninfectious (E) state, i.e., we do not consider a SEIQR 33
model [6]. Further, from the Diamond Princess cruiseship and from the Italian village Vo’ Euganeo, it 34
has been found that ∼50% of nCov-positive individuals do not develop symptoms [7, 8]. We assume 35
that such positive but asymptotic individuals can transmit the disease, i.e., the I state includes both 36
individuals that will not develop symptoms, cases that did not develop symptoms yet, and symptotic 37
patient that still have not been tested positive and isolated. 38
So far a relatively small fraction of the Italian population has been found positive for COVID-19. 39
Thus, we are still in the early phase of the epidemic where S ≈ N , and as well known, the number of 40
dI
≈ (β − (α + η))I, i.e., I(t) = I0 exp{(β − (α + η))t}, (5)
dt
where I0 is the number of infectious individuals at the beginning of the outbreak t = 0. We allow β to be 42
2
105
2
1.8
1.6
1.4
Total Cases
1.2
0.8
0.6
As mentioned, the number of individuals that have been found SARS-nCov2 positive and put in 44
isolation does not correspond to I but to Q + R. From (5) it follows that Q + R follows 45
d(Q + R)
= ηI = ηI0 exp{(β − (α + η))t}. (6)
dt
Assuming that the infection rate changes from β1 to β2 at time T1 , and from β2 to β3 at time T2 we 46
obtain 47
Q0 + R0 + ηI ρ1 t
ρ1 e
0
−1 , t ≤ T1 ,
Q + R + ηI0 eρ1 T1 − 1 + ηI0 eρ1 T1 eρ2 t − 1,
T1 < t ≤ T2 ,
0 0 ρ1
(Q + R)(t) = ηI0 ρ1 T1
ηI0 eρρ21 T1 ρ T (7)
Q0 + R0ρ +T ρρ1 (Te −T ) − 1 + ρ2 e −1
2 2
1 1e 2 2 1
+ ηI0 e
eρ3 t − 1 , t > T2 ,
ρ3
We fit this expression to the Italian COVID-19 data from February 22, 2020 through April 15, 2020 49
(Fig. 1). Further, γ was fit by averaging the number of recovered/deceased cases (∆Ri = Ri − Ri−1 ) 50
divided by the number of active cases on the corresponding day (Qi−1 ) over the three periods [0, T1 ), 51
[T1 , T2 ) and [T2 , April 15 2020). For three periods we estimate, respectively, γ = 0.041, 0.038, 0.023/day 52
(SE 0.007, 0.006, 0.002/day), indicating that isolated and hospitalized patients during the early phase 53
recover or die in approximately 25 days, which is consistent with the estimated median time-to-death 54
To identify the other individuals parameters we use previous findings. It has been estimated that the 56
average incubation time is ∼5 days [10, 11] and the duration of the milder cases of disease it 5-10 days [4]. 57
We assume an average time of duration from infection to recovery or death of non-isolated cases of 10 58
3
days, corresponding to a rate of 0.1/day. If a fraction δ of infectious individuals is tested positive and 59
Similarly η related to the time until patients are tested positive and isolated, but also to the fraction 61
of all infectious individuals that are tested positive. These are mostly symptomatic patients, which we 62
assume are isolated a few days after the incubation time is over and first symptoms appear, i.e., after 63
∼10 days. Letting δ denote the fraction of infectious individual entering Q, we obtain η = δ × 0.1/day. 64
Since ∼50-75% of the population is asymptomatic [7, 8], but some milder cases may also go unnoticed 65
and not end in isolation, we assume that δ = 1/3 of infectious individuals are tested after an average of 66
Results 68
Our chosen values for α and η yield the initial infection rate β1 = 0.292/day and consequently the basic 69
falling between 2 and 4 [10, 12–15]. Further, we obtain an estimate of the number of infectious individual 71
at the moment of the outbreak of I0 = 66.3/0.033 ≈ 2000, although with a large uncertainty because of 72
the large standard error on the estimate of ηI0 , and because the calculated I0 obviously depends on the 73
value of η. Based on the dynamics before the outbreak (δ = 0), we can estimate that the first infectious 74
case appeared in Italy log(I0 )/(β − 0.1) ≈ 40 days before the outbreak, i.e., around January 12, 2020. 75
We then simulated the dynamics of the SIQR model with the estimated parameters and found that 76
the Italian data was very well fitted (Fig. 2). We note that our results indicate that the early Italian 77
containment measure (school closings, hygiene indications, etc.) had negligible, if any, effect on the 78
disease dynamics until T1 falling on March 10, 2020. The lockdown of the Northern regions on March 8, 79
2020, which was followed by complete lockdown of Italy within a few days, lowered the infection rate 80
β by approximately one third. However, only later around March 20, 2020, did R0 become less than 81
one and the number of infected individuals started to decline. This corresponds reasonably with the 82
more stringent control measures and closure of all non-essential work places introduces around this date. 83
Indeed March 20, 2020, was the last day before the wider lockdown of non-essential companies. 84
The model predicts that, during the first phase but after the initial transient, for each patient in 85
quarantine, according to the official statistics, approximately seven infectious individuals are present in 86
the populations (Q/I ≈ 7). From (3) and (5) it can be shown analytically that following the transient 87
and during the exponential growth, I/Q = (γ + β − (α + η))/η = (0.041 + 0.216)/η ≈ 7. In other 88
words, the relation between Q and I follows directly from the fit to the data up to the factor η, while 89
it is independent of the assumptions on the other individual parameters. Considering that η is largely 90
unknown, range η ∈ (0.01, 0.05) used above, we obtain that I is 5–25 fold larger than Q during the early 91
We then simulated different scenarios that may be the result of relaxing the drastic measures imposed 93
during March, 2020 (Fig. 3). The beginning of this so-called ”phase 2” in Italy is currently planned to 94
begin May 3, 2020. If the current measures are maintained, keeping R0 ≈ 0.75, there will still be more 95
than 50.000 active cases in Italy by the end of June 2020. With a modest relaxation of the restrictive 96
measure, keeping R0 below 1, the number of cases still decreases, more slowly, reaching approximately 97
80.000 by end June 2020 for R0 = 0.95. If R0 becomes greater than one as a result of reopening of 98
parts of the society, the number of cases starts to increase again, being almost constant at ∼100.000 for 99
R0 = 1.05 or 1.1, but reaching more than 200.000 cases by the end of June if R0 = 1.3 during “phase 2”. 100
From our analysis it is clear that the decrease in slope seen in the log-transformed data (Fig. 1) is not 102
due to the measures due to school closings etc. imposed during the first weeks after the outbreak in Italy, 103
but a direct consequence of the model structure. Stronger measures were needed to lower R0 through a 104
reduced transmission rate β in order to relieve, in particular, the pressure on intensive care units and 105
4
105
2
Quarantined
1.8 Quarantined+Recovered
Recovered
I from model
1.6 Q from model
Q+R from model
1.4 R from model
Tc
1
1.2 Tc
2
Cases
0.8
0.6
0.4
0.2
hospitals [16]. Currently R0 is below one and the number of active cases is peaking (Fig. 2). However, 106
reopening of the society must be done very slowly in order to avoid that the epidemic reemerges (Fig. 3) 107
With our model approach, we estimate that the number of infectious individuals I is approximately 7 108
times larger than the number of isolated patients Q (Fig. 2). We were also able to estimate that there 109
were ∼2000 of infectious but undetected individuals in Italy at the time of the outbreak around February 110
21, 2020. Only when a patient with severe symptoms was hospitalized and tested for SARS-nCov2, and 111
the first infected person died from COVID-19 on the following day, wide testing and isolation efforts 112
started. By backward interpolation, we estimated that the first infectious individual (“patient zero”) 113
appear in Italy around January 12, 2020. This estimate suggests that COVID-19 was present in Italy 114
even earlier than a presumed ”patient zero”, suggested to be a German citizen linked to a cluster in 115
Munich [5] visiting Italy around January 25, 2020 [17]. 116
In conclusion, general reopening of the Italian society will likely cause the epidemic to resurge. It is 117
therefore paramount that most of current measures regarding social distancing, hygiene precautions and 118
other limitations are maintained even when the current lockdown is lifted and for many months to follow. 119
5
105
2
Quarantined
Current trend, R 0 =0.77
1.8
R0 =0.95
1.6 R0 =1.05
R0 =1.1
1.4
R0 =1.3
Tc 1
1.2
Tc 2
Cases
1 Tc 3
0.8
0.6
0.4
0.2
Fig 3. Model prediction of effects of lifting current restrictions. The figure shows simulations of the
SIQR model as in Fig. 2, but assuming that β, and hence R0 , is increased on May 3, 2020 (vertical blue
dash-dotted line).
Acknowledgments 120
Funding: this work was supported by MIUR (Italian Minister for Education) under the initiative 121
123
Authors contributions: MGP and MM conceived research and discussed all results. MGP developed 124
the theoretical framework and performed parameter estimation. MM developed the simulation framework 125
and compared real data to model results. MGP wrote the paper. MM commented drafts and approved 126
128
References
1. Hethcote HW. The mathematics of infectious diseases. SIAM review. 2000;42(4):599–653.
doi:10.1137/S0036144500371907
6
5. Rothe C, Schunk M, Sothmann P, Bretzel G, Froeschl G, Wallrauch C, et al. Transmission of
2019-nCoV infection from an asymptomatic contact in Germany. New England Journal of Medicine.
2020;382:970-971. doi:10.1056/NEJMc2001468
6. Jumpen W, Wiwatanapataphee B, Wu Y, Tang I. A SEIQR model for pandemic influenza and its
parameter identification. International Journal of Pure and Applied Mathematics. 2009;52(2):247–
265.
7. National Institute of Infectious Diseases Japan. Field Briefing: Diamond Princess COVID-19
Cases, 20 Feb Update. Last accessed March 12, 2020. Available from: https://www.niid.go.jp/
niid/en/2019-ncov-e/9417-covid-dp-fe-02.html.
8. Day M. Covid-19: identifying and isolating asymptomatic people helped eliminate virus in Italian
village. BMJ. 2020 Mar;368:m1165. doi:10.1136/bmj.m1165
9. Zhou F, Yu T, Du R, Fan G, Liu Y, Liu Z, et al. Clinical course and risk factors for mortality of
adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet. 2020.
doi:10.1016/ S0140-6736(20)30566-3
10. Li Q, Guan X, Wu P, Wang X, Zhou L, Tong Y, et al. Early Transmission Dynamics in Wuhan, China,
of Novel Coronavirus-Infected Pneumonia. N Engl J Med. 2020. doi:10.1056/NEJMoa2001316
11. Lauer SA, Grantz KH, Bi Q, Jones FK, Zheng Q, Meredith HR, et al. The Incubation Period of
Coronavirus Disease 2019 (COVID-19) From Publicly Reported Confirmed Cases: Estimation and
Application. Ann Intern Med. 2020. doi:10.7326/M20-0504
12. Zhou T, Liu Q, Yang Z, Liao J, Yang K, Bai W, et al. Preliminary prediction of the basic
reproduction number of the Wuhan novel coronavirus 2019-nCoV. J Evid Based Med. 2020
Feb;13(1):3–7. doi:10.1111/jebm.12376
13. Zhao S, Lin Q, Ran J, Musa SS, Yang G, Wang W, et al. Preliminary estimation of the basic
reproduction number of novel coronavirus (2019-nCoV) in China, from 2019 to 2020: A data-
driven analysis in the early phase of the outbreak. International Journal of Infectious Diseases.
2020;92:214–217. doi:10.1016/j.ijid.2020.01.050.
14. Liu Y, Gayle AA, Wilder-Smith A, Rocklöv J. The reproductive number of COVID-19 is higher
compared to SARS coronavirus. Journal of travel medicine. 2020. doi:10.1093/jtm/taaa021
15. Lai A, Bergna A, Acciarri C, Galli M, Zehender G. Early phylogenetic estimate of the effective
reproduction number of SARS-CoV-2. Journal of medical virology. 2020. doi:10.1002/jmv.25723
16. Remuzzi A, Remuzzi G. COVID-19 and Italy: what next? Lancet. 2020. doi:10.1016/S0140-
6736(20)30627-9
17. Kupferschmidt K. Genome analyses help track coronavirus’ moves. Science. 2020 March;367:1176–
1177. doi:10.1126/science.367.6483.1176