key: cord-0806380-fsrdu4tq authors: Allieta, M.; Allieta, A.; Rossi Sebastiano, D. title: COVID-19 outbreak in Italy: estimation of reproduction numbers over two months toward the Phase 2 date: 2020-05-18 journal: nan DOI: 10.1101/2020.05.12.20076794 sha: 50683837569c30ddae7e9a4b4580ccc08cc44d13 doc_id: 806380 cord_uid: fsrdu4tq After two months from the first case in COVID-19 outbreak, Italy counts more than 190,000 confirmed positive cases. From the beginning of April 2020, the nationwide lockdown started to show early effects by reducing the total cumulative incidence reached by the epidemic wave. This allows the government to program the measures to loosen lockdown restrictions for the so called "Phase 2". Here we provided the reproduction number estimation both in space and in time from February 24th to April 24th, 2020 across two months into the epidemic. Our estimates suggest basic reproduction number averaged over all the regions of 3.29, confirming that epidemiological figures of the SARS-CoV-2 epidemic in Italy are higher than those observed at the early stage of Wuhan (China) outbreak. Based on the SARS-CoV-2 transmission dynamics reported here, we gave a quantitative evaluation of the efficiency of the government measures to low the reproduction number under the unity (control regime). We estimated that among the worst hit regions in Italy, Lombardy reached the control regime on March 22nd followed by Emilia-Romagna (March 23th), Veneto (March 25th) and Piemonte (March 26th). Overall, we found that the mean value of time to reach the control regime in all the country is about 31 days from the February 24th and about 14 days from the first day of nationwide lockdown (March 12th). Finally, we highlighted the interplay between the reproduction number and two demographic indices in order to probe the "state of activity" of the epidemic for each Italian region in the control regime. We believe that this approach can provide a tool in the management of "Phase 2", potentially helping in challenging decision to continue, ease or tighten up restrictions. After the first COVID-19 case was diagnosed in Lombardy, Italy, on February 20th, 2020, [1] the novel coronavirus rapidly spread across the country leading to a dramatic spike in the number of new positive cases and deaths. To minimize the likelihood that people who were not infected come into contact with people who had contracted the disease, the Italian government imposed a series of progressively more strict social distancing measures which culminated in a national lock-down announced on March 11th, 2020. [2] Around two months from the first case and more than 190,000 confirmed positive cases later, from the beginning of April, the effect of the nationwide lockdown started to achieve some level of success and the number of new infections began to smoothly decrease. These early signs of a slowdown of the COVID-19 pandemic in Italy provide a comforting picture of the outbreak's stabilization which is driving the government to periodically review its lockdown measures in view of the so called "Phase 2", i.e. the period during which citizens will have to live together with the virus as of all the industrial sector, including the nonessential economic activities, will start to reopen. However, since the regional differences in the number of new positive cases has been reported to be huge, with the Northern regions of Italy (namely Lombardia) being most affected, the establishment of the proper precautions to plan the "Phase 2" is a truly complicated task. The planned restrictions and permissions that will be applied could thus vary from region to region. In this context, the systematical estimation of key epidemiological parameters, for each region can provide insight into the speed at which the disease had spread and will give a useful tool to figure out if a differential approach at the regional level on the measures to apply for "Phase 2" is feasible to keep down the transmission of SARS-COV2. At the beginning of epidemic and during the lockdown phase, Italian Government and the mainstream of the local and national mass media have been emphasized the relevance of the basic reproduction number (R0) , i.e. the average number of secondary cases generated by a single primary case in a theoretically fully susceptible (100%) population, as the most important and informative parameter to monitor the epidemic trends. Obviously, R0 has an undoubted relevance since when R0 > 1 the infection may spread in the population and more R0 is large and deeper would be the interventions needed to control the epidemic. On the other hand, if R0 < 1, on average the infectious individual infects less than one person and the epidemic falls in a so called "control regime" where it will not be sustained, and it will die out. Nevertheless, R0 is not the only parameter that affect the impact and the spreading of All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05. 12.20076794 doi: medRxiv preprint the disease over a population which may largely result even from several demographic and epidemiological factors. In this communication, we provided an estimation of the basic reproduction number R0 for all the Italian regions by the cumulative confirmed COVID-19 cases continuously updated and made public at the website of Dipartimento della Protezione Civile. [3] In addition, we estimated the time dependent reproduction number Rt, which is the average number of secondary cases generated by an infectious individual at time t. We linked Rt related to the last date of our period of observation (April 24th, 2020), with two demographic and epidemiologic indices in a simple three-dimensional array in order to highlight the "state of activity" of the epidemic for each Italian region. We provide a useful tool in the management of "Phase 2", potentially helping in challenging decision to continue, ease or tighten up restrictions. The official demographic data of the resident population, the surface and the population density updated on January 1st, 2019, for each Italian region and Italy were taken from the Italian National Institute of Statistics (Istituto Nazionale di Statistica, ISTAT) and reported in The official data of COVID-19 epidemic in Italy was taken from the task force of the Dipartimento della Protezione Civile. Cumulative data are available at various aggregation levels, namely national, regional and provincial and are accessible on Github. [3] Data for the analysis were considered from February 24th to April 24th, 2020. In this period, we collected the daily cumulative number of confirmed positive cases (N), the number of "active" confirmed positive cases (NA), i.e., the number of infected people living not recovered from COVID-19, and the "density of infected people" (DA), calculated as NA/surface and expressed like the population density as number of persons/Km 2 , for each Italian region and Italy. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. To obtain the estimation of reproduction number we use the maximum likelihood estimation (ML) method which assumes that the number of secondary cases caused by an index case is Poisson distributed with an expected value R. Given then observation of (N0 , N1 ,..., Nt) incident cases over consecutive time units, R is estimated by maximizing the following loglikelihood function [5] : i is the distribution of the generation time corresponding to the distribution of the serial interval, i.e. the time between when a person gets infected and when they subsequently infect another other people, calculated at time i within the assumption that the incubation period does not change over the course of the epidemic [6] . We consider that the distribution of the serial interval was expected to follow a gamma distribution with mean (±SD) of 6.50 ± 4.03 days as reported by the Imperial College COVID-19 Response Team [7] . We note that this value agrees very well with gamma distribution with mean 6.6 days (95% CI, 0.7 to 19) recently determined from the analysis of 90 observations of individual serial intervals in 55 clusters in Lombardia (Italy) [8] . To estimate R = R0 , the LL(R) function must be calculated over a period where epidemic curves showed exponential growth. As a first guess, to select this time window we used the simple procedure described by Obadia et al. [5] In brief, we computed the function over a range of possibile time periods by determining the deviance r 2 statistic for each iteration. Largest r 2 corresponds to the time window over which the ML model best described data. To evaluate the time dependent reproduction number Rt we adopted the method developed by Wallinga and Teunis. [9] The transmission probability (pij) of individual i being infected by individual j at ti, tj onsets, respectively, can be described mathematically as: [5] The net reproduction number Rj is then then sum of all pij involving j as the infector = ∑ and it can be averaged over all cases with same date of onset as All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05. 12.20076794 doi: medRxiv preprint Finally, since Rt are computed by averaging over all transmission networks compatible with observed incidence data, no assumption is made about the time dependence of the epidemic unlike, for example the exponential growth in the well-known Bayesian approach. [5] , [9] We believe, hence, that this model is particularly suitable to estimate the reproduction number in the post-peak period where the transmission is expected to decrease. All the above data analyses were performed using the R0 package [5] as implemented in statistical software R. [10] 3.Results infection in Italy shows that the exponential growth period may take place during the first 15-20 day from the national epidemic onset (February 24 th , 2020). Table S1-S2 (Supporting Information) show demographic and epidemiological data, respectively. As it is widely known, Table S2 shows that COVID-19 epidemic affected (and is affecting) harder the northern Italian regions, with N=16859 and NA=89384 on April 24th, i.e. more than 80% of the cases of the country (with 54,7% of the Italian resident population), if we aggregate epidemiological and demographic data of the northern regions (Lombardia, Piemonte, Veneto, Emilia Romagna, Liguria, Valle D'Aosta, Trentino-Alto Adige) plus Marche and Toscana regions. Furthermore, in Lombardia region epidemic had a huge spread, with N=71256 and NA= 34368 on April 24th, i.e. more than one third of the cases of the country (with 16.7% of the Italian resident population). In the top of the panels of Figures S1-S3 (Supporting Information) we reported the incidence data for all the regions plus Italy. Initial inspection of the datasets shows again that the exponential growth period may take place during the first 15-20 day from the relative All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint epidemic onset. It should be noted that for the evaluation of R0 in the initial outbreak stage, we considered data from February 24 th , up to March 18 th , since data in a wider range can be affected by the national lockdown on March 11 th . In Figure 2 (a) we showed R0 values obtained for SARS-COV-2 in all the regions and in Italy. Table 1 reports the same data represented in Figure 2 (a), compared with those obtained by Riccardo et al., [2] D'Arienzo et al., [12] Distante et al. [13] . According to our ML estimation, (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In this work, we analyzed the time evolution of incidence of the SARS-CoV-2 epidemic for two months from onset, February 24 th to April 24 th , in all the Italian regions. We estimated the basic reproduction number (R0), by using the ML method in the early stage of the epidemic. In addition, we determined time evolution of this parameter across the two months of the observational period. Finally, we linked Rt, with two indices, the population density and DA, the latter representing the density of infected people in a region as recorded on April 24 th . Firstly, we point out that these data can be considered only an approximation of the actual epidemic dynamics. Indeed, the reported number of cases strictly depends on the number of swabs that are used for Covid-19 testing and can be biased by several factors like underreporting, delays in recording as well as errors in classification of cases. [13] Therefore, large data noise is general observed, especially at the regional level, which requires a careful inspection of the epidemic curve as well as data smoothing in order to avoid unrealistic reproduction number estimation. As described in the results, for the evaluation of R0 in the initial outbreak stage, we considered data from February 24 th , up to March 18 th . Data in a wider range can be affected by the national lockdown on March 11 th . This period agrees well with previous investigation where the same time window has been assumed as the infection period to determine R0 for the whole Italy. [14] Taking these preliminary considerations into account, our result of R0 = 3.22 for Italy is highly consistent with values obtained by fitting the exponential growth rate of the infection across a 1-month period. [12] Similar conclusion has been drawn for Northern regions transmission dynamics and the same results were found for the Southern regions. [12] In another work, Riccardo F. et al. [2] reported R0 ranging from 2.50 to 3.00 for six selected Italian regions (Lombardia, Veneto, Emilia-Romagna, Toscana, Lazio, Puglia). Despite these values are lower than R0 obtained here, a variability of ~ 0.5 for most of the regions is thus confirmed independently of geographical location. Again, Gatto et al. [15] , while including additional parameters like mobility and the spatial distribution of communities, determined a comparable initial generalized reproduction number R0 = 3.60. Overall, these data support the idea that epidemiological figures of the SARS-CoV-2 epidemic in Italy are slightly higher than those observed at the early stage of outbreak in Wuhan (China). [16] The initial large values observed resulted from a sudden increase of independent first reported infections which in many cases can be related to the so called "super-spreading" All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint events. Indeed, as observed for SARS outbreak, [9] in the early stage of the epidemic the time dependence of Rt shows a fluctuating pattern characterized by wide confidence interval raised by the initial low number of cases used in the calculations. In this context, the superspreading events cannot be necessarily triggered by a single infector, but it can be related to few people which are perpetuating an epidemic in the susceptible population. [13] Here we observed that most of the regions have faced "super-spread events" in the early stage of epidemic. and significant is the observation of such event in southern regions After the early stages, the Rt showed a decreasing trend which is likely to be affected by the temporal depletion of susceptible individuals (intrinsic factors) and by the implementation of control measures (extrinsic factors). [17] Both these factors slow down the growth rate of incidence and deeply affect the shape and time scaling of the epidemic peak driving Rt to fall below 1. [17] We found that the mean value of time to reach the control regime is about 31 days from the (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. Although the Italian Government's restrictive measures have proven to be of considerable utility in preventing even more devastating effects from the epidemic, the challenge in tackling "Phase 2" appears even more demanding. In this line, obtaining simple and effective indices to evaluate the state of activity of the epidemic seems mandatory: if the Rt index remains essential for understanding the trend in a given area, however it is not the only parameter to account for. Briefly, if we consider two areas with the same Rt, that of the two that has a population density and a higher percentage of infected people must be considered more at risk, monitored more carefully and potentially the target of more timely restrictive measures. The population density of a given area is clearly an Rt-independent risk factor for the development of an epidemic that spreads through human infection, although the population density of the different Italian regions may not be truly representative of the distribution of the population. Urban areas and in particular metropolitan areas (Rome, Milan, Naples) have a population density higher than the regional one. Furthermore, due to the peculiar Italian orography, some regions (for example Liguria, Valle D'Aosta, Trentino-Alto Adige) concentrate the population in a "habitable" area much less large than the total surface. While admitting its arbitrariness, DA (ie the number of infected people per Km 2 ) is in some way a representative parameter of how much the epidemic was active in the previous period and, above all, what is the generic risk of "meeting" a subject affection in a given area. Therefore, we suggest to associate a combined use of Rt with DA and population density to evaluate the epidemic risk of a specific area, in our case of the Italian regions. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint Caption: ( + ) Date of epidemic onset February 24 th ; (*) the original incidence data related to Trento and Bolzano were merged into a single region called Trentino-Alto Adige resulting in a geographical disaggregation of Italy into 20 regions. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Caption: Official data from the Dipartimento della Protezione civile available at https://github.com/pcm-dpc/COVID -19; (*) the original data related to Trento and Bolzano were merged into a single region called Trentino-Alto Adige resulting in a geographical disaggregation of Italy into 20 regions; N = aggregate number of infected people, NA = number of active infected people, DA = density of active infected people. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint FIGURE S1. Daily incidence as numbers of new cases from February 24 to April 24 2020 for COVID-19 outbreaks (upper panel) and the corresponding time dependent reproduction number (Rt) (lower panel) for the following Italian regions: (a) Abruzzo, (b) Basilicata, (c) Calabria, (d) Campania, (e) Emilia-Romagna, (f) Friuli V. G., (g) Lazio. In upper panels vertical bars are the incidence data whereas in lower panels black dots are the Rt mean values accompanying by grey vertical lines standing for 95% confidence intervals. In the same panel the horizontal solid line indicates the threshold value R = 1, above which an epidemic will spread and below which the epidemic is controlled. Days are listed from the onset February 24 th , 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint FIGURE S2. Daily incidence as numbers of new cases from February 24 to April 24 2020 for COVID-19 outbreaks (upper panel) and the corresponding time dependent reproduction number (Rt) (lower panel) for the following Italian regions: (a) Liguria, (b) Lombardia, (c) Marche, (d) Molise, (e) Piemonte, (f) Puglia, (g) Sardegna, In upper panels vertical bars are the incidence data whereas in lower panels black dots are the Rt mean values accompanying by grey vertical lines standing for 95% confidence intervals. In the same panel the horizontal solid line indicates the threshold value R = 1, above which an epidemic will spread and below which the epidemic is controlled. Days are listed from the onset February 24 th , 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. In upper panels vertical bars are the incidence data whereas in lower panels black dots are the Rt mean values accompanying by grey vertical lines standing for 95% confidence intervals. In the same panel the horizontal solid line indicates the threshold value R = 1, above which an epidemic will spread and below which the epidemic is controlled. Days are listed from the onset February 24 th , 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.12.20076794 doi: medRxiv preprint Covid-19: preparedness, decentralization, and the hunt for patient zero COVID-19 working group, Epidemiological characteristics of COVID-19 cases in Italy and estimates of the reproductive numbers one month into the epidemic The R0 package: a toolbox to estimate reproduction numbers for epidemic outbreaks Estimation in emerging epidemics: biases and remedies Estimating the number of infections and the impact of non-pharmaceutical interventions on COVID-19 in 11 European countries The early phase of the COVID-19 outbreak in Lombardy, arXiv Different epidemic curves for severe acute respiratory syndrome reveal similar impacts of control measures R: A language and environment for statistical computing. R Foundation for Statistical Computing Assessment of the SARS-CoV-2 basic reproduction number, R0, based on the early phase of COVID-19 outbreak in Italy Covid-19 Outbreak Progression in Italian Regions: Approaching the Peak by the End of March in Northern Italy and First Week of April in Southern Italy Int No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity Guida alla lettura e all'interpretazione dei dati COVID-19 COVID-19 and Italy: what next? Lancet (2020) Spread and dynamics of the COVID-19 epidemic in Italy: Effects of emergency containment measures 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 The effective reproduction number as a prelude to statistical estimation of time-dependent epidemic trends No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted