A simplified model for the analysis of COVID-19 evolution during the lockdown period in Italy

A simplified model applied to COVID-19 cases detected and officially published by the italian government, seems to fit quite well the time evolution of the disease in Italy during the period feb-24th - may-19th 2020. The hypothesis behind the model is based on the fact that in the lockdown period the infection cannot be transmitted due to social isolation and, more generally, due to the strong protection measures in place during the observation period. In this case a compartment model is used and the interactions between the different compartments are simplified. The sample of cases detected is intended as a set of individuals susceptible to infection which, after being exposed and undergoing the infection, were isolated ('treated') in such a way they can no longer spread the infection. The values obtained are to be considered indicative as they depend not only on the quality of the data collected, but also on the qualitative approximation of the curves modeled to the data and on the number of available data. The same model has been applied both to the data relating to Italy and to some regions of Italy (Lombardia, Piemonte, Lazio, Campania, Calabria, Sicilia, Sardegna), generally finding a good response and indicatively interesting values.


Introduction
After the so-called 'phase 1', characterized by social distancing and a general halt of all 'nonessential' activities, the curve of the new COVID-19 cases detected daily dropped sharply. This period, also called 'lockdown' started with the decree of the Italian government of 9 March and lasted until 4 May 2020.
The data of the daily detections of infected cases are available to everyone, starting from February 24th, 2020, on the website [1]. This allowed a comparison of the growth curve of the cases detected with a compartmental model of the Susceptible-Exposed-Infected-Recovered type (see [2], [4], [3], ) which provides the temporal evolution of the number of individuals belonging to the different compartments. The model has been suitably simplified to take into account the lockdown situation.
The period under consideration (from 24/2/2020 to 19/5/2020) is sufficiently limited to allow us to overlook the effects of births and deaths on the population, therefore in this context we consider a constant population of N individuals, with N = S + E + I + R.
In the initial phase, the spread of the infection was not controlled, thus allowing an exponential increase in the number of Exposed and Infected. Between March 4th and 9th drastic measures were taken which, starting from Lombardy, were extended to all italian regions. The measures in the lockdown period tend to limit the spread of the infection as much as possible by isolating individuals in homes as far as possible and, where not possible (essential services), by imposing individual protections and disinfection of common areas.
The lockdown period, simplifying, can also be thought of as a type of treatment for infective individuals. We can therefore imagine that infective individuals, detected daily, in general are 1 of 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint managed ('treated') in such a way that they cannot infect other susceptible individuals. This 'treatment' includes both hospitalization and isolation.
The lockdown period, characterized by strong social isolation, also allows us to neglect any potential transfers between the various departments beyond the required path S → E → I → T .
To avoid confusing the SEIR model with the lockdown approximation, in this context, we will call T (Treated) the compartment of infected people who are no longer able to transmit the pathogen.
In the simplified SEIT model, the relationship between the parameters are (ref. eq. 17, 19, 20)

SEIR compartmental models
The mathematical models for the study of the spread of infections are based on a subdivision of the population into 'compartments' and on systems of differential equations to represent their temporal evolution. The most common models are the SEIR models (Susceptible-Exposed-Infective-Recovered) described in [2], [4], [3].
In the SIR and SEIR models the variables are represented by the number of individuals in the different compartments over time (in this case on a daily basis)

S (Susceptible)
Number of uninfected individuals susceptible to pathogen infection. E (Exposed) Number of individuals who have been exposed to the pathogen but who, throughout the incubation period, are not spreading the infection.
Number of individuals infected and capable of transmitting the pathogen to others. R (Recovered) Number of individuals who have passed the infection and are no longer able to transmit the pathogen to others.
The evolution over time of the variables S, E, I, R can be modeled (see eq.1) considering the interactions between the different compartments and taking into due consideration the relationships that contribute to the increase the number of infected than those that only describe the displacements between the various compartments without increasing the I variable. (1) In the system of differential equations (1) the components X i are functions of the variables (E, I, R, S) andẊ their derivative over time.

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint The solution of (1) represents the evolution over time of the number of individuals in the different sectors.
Given a situation of equilibrium (DFE: Disease Free Equilibrium) represented by a set of values X 0 , the stability of the system can be studied by analyzing the contribution of infectious components close to the DFE state.
By construction, the S and R compartments do not generate infections, therefore the variables to be considered in the stability study near DFE areĖ andİ (2 ).
In order to study the stability near a DFE, it'is important to distinguish the 'new infections' from the movement between the departments, the differential equation (2) associated with each variable is divided into two components: By construction F(X) and V(X) represent: • F i : new infections.
• V i : transfers between compartments only.
From this representation, given the initial conditions S 0 , E 0 , I 0 , R 0 , and the interaction parameters between the various compartments that characterize F and V, solving the (1) the temporal evolution of the number of individuals in the different compartments is obtained.
To study the stability of the system near equilibrium conditions (ref. [2]) we consider a DFE as solution of the differential equations (1) and the perturbative impact of the two infectious variables using the two Jacobian matrices of size 2 * 2 relating to I and E (see eq. 2) .
The product F V −1 is known as 'next generation matrix' [2]. Each element F V −1 ij represents the number of 'secondary' infections of the i compartment due to a single infected of the j compartment assuming that the whole population is susceptible. Each element of the matrix is a 'reproduction number' which indicates how much an infectious person in one compartment affects the number of infected in another compartment.
The dominant eigenvalue (spectral radius) of the 'next generation matrix' F V −1 is known as the 'basic reproduction number' R 0 and represents the number of new cases generated, on average, from a single case during its infectious period. R 0 also represents the stability threshold of the system near an equilibrium state.
• R 0 < 1 : the equilibrium (DFE) is locally stable and diffusion does not extend; • R 0 > 1 : the equilibrium is unstable allowing the spread of the infection.

A simplification (SEIT) to model the lockdown
As already mentioned, we will consider lockdown as a form of treatment of the infection since all members of the I sector, also as a result of the lockdown measures and, more generally, of social isolation, distancing and individual protection, reduce their ability to spread the pathogen. In this case, to avoid confusion we will call T (Treated) the R (Recovered) variable of the SEIR model.

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint We also will simplify the interactions between compartments by considering only the forward ones S → E → I → T and we will consider the total number of individuals N = S + E + I + T constant. In this context, N represents a sample of individuals, with homogeneous distribution, susceptible to infection in an environment with constant transfer coefficients between compartments.
The transfers between compartments in this case are: We note that: • The transfer S → E is proportional to the product of the ratio of infected individuals (I/N ) by the daily contact coefficient α SE of the individuals in S.
• The coefficient α EI represents the daily passage of individuals from E to I. The value 1/α EI is basically related to the incubation time of the infection.
• the value α IT basically represents the speed of 'treatment' of the individuals in the compartment I, so that they can no longer transmit the pathogen, with consequent transfer to compartment T .
Recalling that, in this context, we consider N constant over time, we have thatṘ = N −Ė − I −Ṫ , and the system of differential equations results: The variables are ordered so that the first two are the 'infectious variables' (E, I).
To study the stability we will considerĖ andİ and wh have: in eq.(7) α SE (I/N )S it is the only element that contributes to the increase of I.
Considering the separation of variables in two groups 'infective' and 'transfer' (see eq. 3), the eq. (7) in vector terms can be written as: A trivial solution of equilibrium DFE is had for S 0 = N which corresponds to the initial vector: CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint The Jacobian matrices (see eq. 4) near the equilibrium condition X 0 (S = N ) for the infectious variables E and I are therefore: The eigenvalues of the next generation matrix F V −1 are: The greater eigenvalue (spectral radius) λ 2 represents the 'basic reproduction number' R 0 . In this simplified representation it is easy to see that, once the equilibrium condition (DFE) is disrupted, R 0 > 1 implies that the daily contacts rate α SE in S compartment, that will produces new infected, is greater than the number of infective individuals treated per day α IT . In this case, on average, an infected individual produce more than one new infected during his infectious period. As consequence the system moves away from equilibrium by extending the spread of the pathogen.
On the other hand, if α SE < α IT (R 0 < 1) then the number of new infected per day is less than the neutralization rate and the infection will not spread further.

Model results for Italy
The DFE equilibrium represented by the trivial solution (9) was disrupted (15) with an individual 'Exposed' within a sample of N individuals. All S 0 = N − 1 individuals are susceptibles to infection.
N is equal to the total number of cases detected in Italy from 24 Feb to May 19th [1] : N = 226699 individuals.
5 of 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint In this context we consider the data [1] as detected values (T data ) for the compartment T . In order to smooth the effect of the variability of the number of cases between the various regions, an initial disruption based on the percentage of cases detected should be carried out. Because, as already mentioned, the results are indicative, for simplicity the perturbation described by (15) is used for all simulations.
The first problem to be addressed, for the definition of the model, is the estimation of the parameters α SE , α EI , α IT for the transfer rate of individuals between the compartments . To determine them, a logistic function is used: where A can be easily found from the first value at t = 0: To use the (16) the following assumptions have been made: • in the growth phase, the normalized data can be approximated to a logistic curve (16). To estimate k we will refer to the day in which the data are growing with a variation within 20% of maximum growth speed (see fig. 1a); • the parameter α EI is the reciprocal of the incubation time τ which will be the only tuning parameter of the model.
• the αSE parameter is proportional to the ratio between α EI and the k parameter.
• the parameter α IT is proportional to the product between α EI and the parameter k.
With the above assumptions these relationships have been identified: where η is a unitary parameter introduced to make the product ηk dimensionless. For simplicity, we will call κ = ηk the dimensionless parameter that will be used later: The relationship between the three parameters will therefore be: Then, in this context, for R 0 we can write (ref. 14 and 18): From the available data it was easy to derive A and k (ref. Eq. 16). The only 'tuning' parameter is τ which has been modified on the basis of a qualitative approximation of the model to the data.

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint Based on these parameters and relationships (17), the simplified SEIT model described above has been set up.
The time evolution of the Treated compartment (T ) of the model has been then compared with the values T data available for COVID-19 Italia (ref. [1]) finding a good representation with a value of the tuning parameter τ = 6days (see tab. 1 and fig. 1).   The model and the data were 'synchronized' on values around 5% of their normalized values obtaining an indicative estimate of the day t 0 in which the diffusion began and of the two peak values for E and for I (see fig. 2).

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(b) E, I evolution and model parameters The graphs in fig. 3 show the trend of the variation over time of the cases detected and of the compartment T ( fig. 3a) and the difference between the model and the cases detected ( fig. 3b). We can note a clear difference between data and model following the end of the lockdown phase.  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint 5 The simplified model applied to some regions of Italy Table 2 summarize the application of the simplified model to some italian regions, the graphic details are in the dedicated sections.   10 of 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Piemonte
(a) Data and model synchronized.
(b) E, I evolution and model parameters

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Lazio
(a) Data and model synchronized.
(b) E, I evolution and model parameters

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Campania
(a) Data and model synchronized.
(b) E, I evolution and model parameters

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Calabria
(a) Data and model synchronized.
(b) E, I evolution and model parameters 14 of 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 5, 2020.

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint

Sardegna
(a) Data and model synchronized.
(b) E, I evolution and model parameters

of 17
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 5, 2020. . https://doi.org/10.1101/2020.06.02.20119883 doi: medRxiv preprint