Note: the website is optimized for computer viewing; in case of mobile or tablet viewing landscape mode is recommended.

Work group

Project coordinator

Prof. Dario Gregori, PhD, Head of the Unit of Biostatistics, Epidemiology, and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli Studi di Padova. LinkedIn

App and R package development

Corrado Lanera, PhD, Unit of Biostatistics, Epidemiology, and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli Studi di Padova. Head of the Laboratory of Artificial Intelligence for Medical Sciences LinkedIn

Epidemiological modeling

Prof. Paola Berchialla, PhD, Department of Clinical and Biological Sciences -- Università degli Studi di Torino LinkedIn

Prof. Dolores Catelan, PhD, Department of Statistics, Computer Sciences, Applications "G. Parenti" (DISIA), Università degli Studi di Firenze LinkedIn

Predictive models

Danila Azzolina, PhD, Department of Translational Medicine -- Università del Piemonte Orientale LinkedIn

Ilaria Prosepe, MSc, Unit of Biostatistics, Epidemiology, and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli Studi di Padova. LinkedIn

Modeling in Environmental Epidemiology and Pollution

Prof. Annibale Biggeri, MD, MSPH, MSB, Department of Statistics, Computer Sciences, Applications "G. Parenti" (DISIA), Università degli Studi di Firenze LinkedIn

Prof. Cristina Canova, PhD, Unit of Biostatistics, Epidemiology and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli studi di Padova. LinkedIn

Elisa Gallo, MSc, Unit of Biostatistics, Epidemiology and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli studi di Padova. LinkedIn

Francesco Garzotto, MSc, Unit of Biostatistics, Epidemiology and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli studi di Padova. LinkedIn

Geospatial Analysis e Cartographic Representation

Prof. Francesco Pirotti, PhD, Department of Territory and Agro-Forestry Systems (TESAF) CIRGEO - Interdepartmental Centre of Geomatics Research. LinkedIn

Analysis of mortality

Prof. Corrado Magnani, PhD, Department of Translational Medicine -- Università del Piemonte Orientale, Novara. LinkedIn

Daniela Ferrante, PhD, Department of Translational Medicine -- Università del Piemonte Orientale, Novara. LinkedIn

Risk communication

Giulia Lorenzoni, PhD, Unit of Biostatistics, Epidemiology, and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli Studi di Padova. Head of the Laboratory of Clinical Epidemiology and Digital Health LinkedIn

Nicolas Destro, MA, Unit of Biostatistics, Epidemiology, and Public Health of the Department of Cardiac, Thoracic and Vascular Sciences and Public Health -- Università degli Studi di Padova. LinkedIn

Navigation instructions

Website structure

  1. Home: This page.
  2. Highlights: Main considerations for the Veneto region.
  3. Epidemic: Dynamic and interactive time series.
  4. Principal indices: Regional and national models and predictions.
  5. Issues: Link to report issues with the website.
  6. Info and sources: Sources description, Use License and software used for the app development.
  7. Latest metrics: Choose a region (or 'Italy', default) to visualize the most relevant metrics based on the latest data. Data is updated daily, usually at 6pm (local time), by the Italian Civil Protection.

How to use the dynamic graphs

The majority of graphs on this website are dynamic. The user can choose:

1. the information he/she wants to visualize: while looking at one graph the user can visualize further details by clicking or hovering the mouse cursor over the different points/curves; zoom on some areas of interest by clicking on the semi-transparent buttons +/- at the top right of the graph (or by selecting the area with the mouse cursor). If multiple information is reported on the same graph (e.g. multiple regions o measures), it is possible to exclude part of the information by clicking on the right legend items or visualize only one piece of information by double-clicking on it. It is moreover possible to save each graph independently by clicking on the semi-transparent camera button. By clicking on the small-house button the original version of the graph is restored.

2. which and how much information is to be processed and shown: Whenever cells appear above the graph, the user can decide to add or remove regions, provinces or metrics (from those available when the pane is selected).

Impact of testing on hospitalizations

The Veneto region and the Piemonte region followed two different testing policies. In order to estimate the impact of a more wide testing policy we compared these two regions.
In particular we tried to apply to the Piemonte region the model embraced by the Veneto region.

Figure 1 shows that, based on the number of confirmed cases, the two regions should register more or less the same number of hospitalizations: it the figure here below this is represented by the green and the red curve, which are almost overlapping.
Still, if we take a look at the real data we can see that the Piemonte region (red dots) actually registers a much higher number of hospitalizations then the Veneto region.

If we include in the model the number of swabs performed throughout time, as shown in Figure 2, we can see that the Veneto model can predict well the number of hospitalizations in the Piemonte region, hence explaining the difference observed in Figure 1.

Methodology

We used a Poisson model with a natural spline of degree 3 on the days and a swabs x days interaction. We used the resident population as an offset.

Preliminary analysis of the overall mortality in 122 municipalities of the Veneto Region from the 1st of March to the 21 of March 2020.

The National Institute of Statistics (Istat) made available on its website (https://www.istat.it/it/archivio/240401) the mortality data of 1084 Italian municipalities, with data updated to the 21st of March 2020. 1.1

As it is possible to see on the Istat website, the municipalities that take part in this analysis are the ones that counted at least 10 deaths in the period 1 January 2020 to 28 March 2020 and that registered a rise in mortality of at least 20 % in the first 21 or 28 days of March 2020. The selection criteria introduced by ISTAT causes an overestimate of mortality. For this reason the numbers we present here must be seen as the highest forseeable values.

Overall mortality is a strong indicator as it has low susceptibility to errors or discrepancies in assessments and it accounts for both the mortality caused directly by a specific pathology and the mortality caused in an indirect way, as for example by difficulties for people who suffer from different pathologies in accessing the hospital services.

Moreover, the overall mortality is not affected by diagnostic questions or difficulties in coding cause of death and is therefore a useful foundation on which we can build an accurate estimate of the effects of the COVID-19 epidemics.

Data is made available by Istat in different tables, which can be accessed and downloaded from the official website. The tables allow an immediate reading and can also be used for further analysis. We hence used the data in order to make some descriptive analysis, that are essentially presented in the form of graphs, in order to illustrate the trend in overall mortality by geographic area, sex, age and time period.

These are some preliminary analyses that aim at sharing information during times of emergencies, that will be improved and explored further in the coming weeks. In particular the current goal is only to give a reasoned presentation of the absolute values and the change percentages. Further analyses will be conducted in order to reach a better modelling of the trend and to improve the indices of the confidence intervals.

These analyses want to answer to the following questions:

  • What is the entity of the observed mortality change if we compare the period from the 1st to the 21st of March 2019 to the period from the 1st to the 21st of March 2020?
  • How the mortality change distributed by sex, age and province of residency?
  • If we also consider previous years, starting from 2015, can we observe relevant change throughout the different years? And again, what is the distribution by sex, age and province of residency?
  • Starting from which week of the year is it possible to observe change of the overall mortality?

Note on aggregation and numerosity of data: some variables were grouped into wider categories, as indicated in the analyses results.

How much did the overall mortality change from last year (1-21 March 2019 vs 1-21 March 2020)? How is the mortality change distributed by sex, age, and province of residency?

The percentage change in mortality (1-21 March 2019 vs 1-21 March 2020) was estimated by region, sex and age aggregated data. In this case, aggregation by province only includes the aforementioned municipalities made available by Istat. Data was categorized as in the table provided by Istat (https://www.istat.it/it/files//2020/03/Tavola-sintetica-decessi.xlsx). Age categories are: 65-74, 75-84, over 85.

The percentage change is defined as:

change% = 100 * ( deaths2020 -- deaths2019 ) / deaths2019

This index appears in the original table, computed for each municipality. In order to reduce statistical variability given by random fluctuations, which is rather high in those municipalities with a low number of inhabitants, we computed the percentage change on a regional level. Total deaths by provincies and percentage change (from 2019 to 2020) are shown in the table.2

The analysis was also conducted separately by age class and sex, and results are presented in the following graphs (Figure 1 and Figure 2)

Figure 1: Change percentage by age and province. 1-21 March 2019 vs. 1-21 March 2020.

For a correct reading of the percentage change it is necessary to remember that number of total deaths is very different from one province to another as the sample size can change quite a lot. In some provincies the mortality change seems quite important, but it is actually given by a small difference in terms of number of deaths (Table 1).

Figure 2: Change percentage by sex and province. Periodo 1-21 March 2019 vs. 1-21 March 2020.

Considering the data on mortality starting from 2015, what is the entity of the change registered throughout the years by age and province of residency?

The data provided by Istat allows to analyse the mortality trend starting from 2015. Data can be found at https://www.istat.it/it/files//2020/03/dati-comunali-settimanali-ANPR-1.zip. Further analyses will be conducted in the coming weeks to better explore mortality between 2015 and 2020.

Deaths in all municipalities in the Istat database belonging to the same province were summed together in order to obtain the number of deaths by province. The graph here below (Figure 3) shows the number of total deaths by province from 2015 to 2020.

Figure 3: Number of deaths by province in the period 1-21 March from 2015 to 2020.

The graphs here below (Figure 4) show how mortality changed from 2015 to 2020 by province and age. Deaths in all municipalities in the Istat database belonging to the same province were summed together in order to obtain the number of deaths by province. Age classes were defined as follows: under 64 (putting together the classes 0-14 and 15-64 of the original table), 65-74, over 75.

It is necessary to keep in mind that the graphs only show absolute numbers, hence differences between provinces are mainly due to different sample sizes.

Figure 4: Number of deaths by province and age in the period 1-21 March from 2015 to 2020.

In which week of the year is it possible to notice change in the overall mortality?

Data regarding the 122 municipalities of the Veneto Region, as presented in the table at https://www.istat.it/it/files//2020/03/dati-comunali-settimanali-ANPR-1.zip for the period of time that goes from the 1st of January to the 21st of March, can be helpful to answer this question. The data in the table is divided into time slots of 7 days, except for the period from the 1st to the 10th of January; this period was therefore excluded from the analysis. The following graphs (Figure 5) present the trend by age and province. The graphs report on the horizontal axis the date that represents the beginning of each time slot.

Figure 5: Number of weekly deaths by age and province from the 12th to the 21st of March 2020.

Notes

1 For further information on data collection see the Istat methodology.
2 2.If the index is equal to 100% it means the mortality has doubled.

Preliminary analysis of the overall mortality in 1084 italian municipalities from the 1st of March to the 21 of March 2020.

Prof. Corrado Magnani
Department of Translational Medicine -- Università del Piemonte Orientale

The National Institute of Statistics (Istat) made available on its website (https://www.istat.it/it/archivio/240401) the mortality data of 1084 Italian municipalities, with data updated to the 21st of March 2020. 1.1

As it is possible to see on the Istat website, the municipalities that take part in this analysis are the ones that counted at least 10 deaths in the period 1 January 2020 to 28 March 2020 and that registered a rise in mortality of at least 20 % in the first 21 or 28 days of March 2020. The selection criteria introduced by ISTAT causes an overestimate of mortality. For this reason the numbers we present here must be seen as the highest forseeable values.

The data collection includes 6.177.016 men and 6.496.805 women, distributed according to residence. Representation is different in the different regions (Table 1).

Table 1: Residents in the 1084 Italian municipalities as of the 1st of January 2019.

Overall mortality is a strong indicator as it has low susceptibility to errors or discrepancies in assessments and it accounts for both the mortality caused directly by a specific pathology and the mortality caused in an indirect way, as for example by difficulties for people who suffer from different pathologies in accessing the hospital services.

Moreover, the overall mortality is not affected by diagnostic questions or difficulties in coding cause of death and is therefore a useful foundation on which we can build an accurate estimate of the effects of the COVID-19 epidemics.

Data is made available by Istat in different tables, which can be accessed and downloaded from the official website. The tables allow an immediate reading and can also be used for further analysis. We hence used the data in order to make some descriptive analysis, that are essentially presented in the form of graphs, in order to illustrate the trend in overall mortality by geographic area, sex, age and time period.

These are some preliminary analyses that aim at sharing information during times of emergencies, that will be improved and explored further in the coming weeks. In particular the current goal is only to give a reasoned presentation of the absolute values and the change percentages. Further analyses will be conducted in order to reach a better modelling of the trend and to improve the indices of the confidence intervals.

The analyses want to answer to the following questions:

  • What is the entity of the observed mortality change if we compare the period from the 1st to the 21st of March 2019 to the period from the 1st to the 21st of March 2020?
  • How the mortality change distributed by sex, age and region of residency?
  • If we also consider previous years, starting from 2015, can we observe relevant change throughout the different years? And again, what is the distribution by sex, age and region of residency?
  • Starting from which week of the year is it possible to observe change of the overall mortality?

Note on aggregation and numerosity of data: some variables were grouped into wider categories, as indicated in the analyses results.

How much did the overall mortality change from last year (1-21 March 2019 vs 1-21 March 2020)? How is the mortality change distributed by sex, age, and region of residency?

The percentage change in mortality (1-21 March 2019 vs 1-21 March 2020) was estimated by region, sex and age aggregated data. Remember data only includes the municipalities provided by Istat. Data was categorized as in the table provided by Istat (https://www.istat.it/it/files//2020/03/Tavola-sintetica-decessi.xlsx). Age categories are: 65-74, 75-84, over 85.

The percentage change is defined as:

change% = 100 * ( deaths2020 -- deaths2019 ) / deaths2019

This index appears in the original table, computed for each municipality. In order to reduce statistical variability given by random fluctuations, which is rather high in those municipalities with a low number of inhabitants, we computed the percentage change on a regional level. Total deaths, number of municipalities included in the measurement for each region and percentage change (from 2019 to 2020) are shown in the table.2

The analysis was also conducted separately by age class and sex, and results are presented in the following graphs (Figure 1 and Figure 2)

Figure 1: Change percentage by age and region. 1-21 March 2019 vs. 1-21 March 2020.

All regions show increased mortality for the two oldest age classes. Moreover, the majority of regions also present an increased mortality between 65 and 74 years. For a correct reading of the percentage change it is necessary to remember that number of total deaths is very different from one region to another as the sample size can change quite a lot. In some regions the mortality change seems quite important, but it is actually given by a small difference in terms of number of deaths.

Figure 2: Change percentage by sex and regions. Periodo 1-21 March 2019 vs. 1-21 March 2020.

Considering the data on mortality starting from 2015, what is the entity of the change registered throughout the years by sex, age and region of residency?

The data provided by Istat allows to analyse the mortality trend starting from 2015. Data can be found at https://www.istat.it/it/files//2020/03/dati-comunali-settimanali-ANPR-1.zip. Further analyses will be conducted in the coming weeks to better explore mortality between 2015 and 2020. Anyhow, a descriptive analysis is still useful to notice the low variability of the number of deaths between 2015 and 2019.

Deaths in all municipalities in the Istat database belonging to the same region were summed together in order to obtain the number of deaths by province. The graphs here below (Figure 3) show the number of total deaths by region from 2015 to 2020. Regions were aggregated in two groups in order to enhance the readability of the graphs. The two groups are based on an Istat classification that splits regions in "North Regions" and "South Central Regions and Islands".

Figure 3: Number of deaths by regions in the period 1-21 March from 2015 to 2020.

The graphs here below (Figure 4) show how mortality changed from 2015 to 2020 by region and age. Age classes were defined as follows: under 64 (putting together the classes 0-14 and 15-64 of the original table), 65-74, over 75.

It is necessary to keep in mind that the graphs only show absolute numbers, hence differences between regions are mainly due to different sample sizes.

Figure 4: Number of deaths by regions and age in the period 1-21 March from 2015 to 2020.

In which week of the year is it possible to notice change in the overall mortality?

Data presented in the table at https://www.istat.it/it/files//2020/03/dati-comunali-settimanali-ANPR-1.zip for the period of time that goes from the 1st of January to the 21st of March can be helpful to answer this question. The data in the table is divided into time slots of 7 days, except for the period from the 1st to the 10th of January; this period was therefore excluded from the analysis. The following graphs (Figure 5 and 6) present the trend by age, sex and region. As in the previous analyses, age is divided into classes (same as before) and regions are divided into two groups (same as before).

In this study it is possible to notice increased mortality starting from the week of the 1st of March, especially in those regions that are most affected by the epidemics (especially the Lombardy Region). Remind that the municipalities included in the analysis are the ones made available by Istat.

Figure 5: Number of weekly deaths by sex and region from the 12th to the 21st of March 2020.

Figure 6: Number of weekly deaths by region and age from the 12th to the 21st of March 2020.

Notes

1 For further information on data collection see the Istat methodology.
2 2.If the index is equal to 100% it means the mortality has doubled.

Comparative analysis between the Piemonte Region and Veneto Region of the epidemiological data relative to Covid-19 infection.

Analysis of data related to number of cases, number of tests, number of hospitalizations, number of ICU admissions and number of deaths from the 28th of February to the 30th of March in the Piemonte Region and the Veneto Region.

The two regions took different approaches to the use of the nasopharyngeal swab. The Piemonte Region followed right away the guidelines provided by the Italian Superior Council of Health, testing, during the early stage of the outbreak, only people who both presented clinical signs of Covid-19 infection and satisfied some epidemiological criteria (i.e. came in contact with someone who visited areas considered at risk). The Veneto Region, on its own initiative, decided to act differently: all people that showed symptoms of Covid-19 were tested, as well as people who came in contact with them. This different approach during the early stage of the epidemics (from 28th February to 8th March) drew attention to a higher number of cases in the Veneto Region (Figure 1).

On the 8th on March the so-called "Red Zone" was extended to the whole country placing Italy on lockdown, which led to a progressive reduction of circulation and everyday activities.

Although in the early stage of the outbreak (approximately one week) the Piemonte Region showed a lower number of cases, it is possible to see how Veneto gained in terms of hospitalizations. Looking at the curves that plot the number of hospitalizations in the two regions it is possible to notice a clear bifurcation, showing that the Piemonte region was heavily burdened with a significant increase in hospitalizations (Figure 2). The same observations can be made looking at the number of ICU admissions (Figure 3) and the number of deaths (Figure 4).

The parting of the hospitalization curves of the two regions around the 12th of March (when the Piemonte Region started to grow faster than the Veneto Region, though apparently having less cases) could be the effect of a smaller number of tests performed: people that were identified were way less than the real number, which also led to overestimate the fatality rate (deaths for the disease/people with the disease).The health policies implemented by the Veneto Region paid off. The Region decided to test more, probably motivated by the first outbreak in Vo' Euganeo, increasing, at least at the beginning, the chances of finding and isolating positive cases, hence protecting the most vulnerable age groups and reducing the spread of the virus even before the lockdown of the 8th of March. As the curves plotting the number of cases divided more and more markedly, the deaths curves also did so, with a temporal shift of 5/6 days.

In the current stage of the epidemics, this probably means that:

  1. patients who present symptoms but do not test positive are left at home and do not start any therapies right away;
  2. When the situations worsens, patients visit the ER where, if they are positive, they are hospitalized and can finally start a therapy;
  3. Still, they start therapy too late to prevent ARSD and, probably, the late use of tocilizumab is less likely to succeed;
  4. with consequent increase in terms of deaths.

Our analysis suggests that in order to modify this sequence of events, reduce hospitalizations and most importantly optimize ICU admissions it is necessary to increase the number of at-home diagnoses, extending testing to paucisymptomatic patients. This would allow not only to reduce the number of infected people but also to start an anti-inflammatory and/or antiviral at-home therapy under the direct control of trained doctors. The aim would be to prevent or at least mitigate and slow down the disease progression, downsizing the fatality rate of Covid-19. It is our belief that the integrated management of hospital and territory could be the winning strategy.

*The following people contributed to the interpretation of data: Carmen Fava, Giuseppe Saglio (Department of Clinical and Biological Sciences, Università di Torino) and Andrea Ricotti (Department of Public Health and Pedriatic Sciences, Università di Torino

Possible effect on hospitalizations of the health policies implemented by the Veneto region

This works aimes at giving a first impression of the possible effect of the health policies implemented by the Veneto region in order to contain the spread of COVID-19.

In order to understand whether the containing measures helped slow down the spread of COVID-19, a predictive model based on the data collected until the 12th of March was compared to what was actually observed.

Figure 1 shows that there was a slowdown after the 12th of March: this day represents an epidemic change-point.

Thanks to the comparison between the predicted and actual values it was possible to estimate some quantities:

  1. The number of avoided hospitalizations in the Veneto region as of the 27th of March: 800 (95% C.I. 755 -- 845) (Figure 2)
  2. Il rallentamento dell'evolversi della epidemia rispetto al previsto:
    • 3.64 (95% C.I. 3.12 - 4.16) days were "gained" in terms of hospitalizations as of the 24th of March (Figure 3)
    • Rallentamento dell'epidemia al 27 marzo pari a 97.91 ospedalizzazioni/giorno (95% C.I. 94.33 -- 101.48) (Figura 4)

Expected hospitalizations (bold green curve; the other two green curves indicate the 95% confidence levels) based on course of the epidemic as registered until the 12th of March. Actual values (red dots) observed in the following days.

Figure 2. Avoided hospitalizations in the Veneto region compared to what was expected from the data gathered until the 12th of March. The grey area indicates the 95% confidence interval.

Figure 3. Gained days, estimated by looking at the shift to the right of the curve (predicted vs observed). The grey area indicates the 95% confidence interval.

Figure 4. Slowdown of the epidemic velocity (predicted vs observed). The grey area indicates the 95% confidence interval.

Technical details regarding the estimation of the model

The estimation of the model was based on the number series of hospitalizations that were observed until the 12th of March. This day represents a change-point in terms of growth of the epidemics. This change in the number series was detected by a Bayesian Changepoint Detection Method (1). The polynomial regression model is based on a local approximation of the regression function (smoothing parameter equal to 0.75). The shape of the curve fits the quadratic trend of the early stage of the outbreak.

Recent studies showed that the curve of cases could be of a quadratic nature rather than exponential, especially in the early stage of the outbreak (2).

It is assumed that the hospitalizations growth rate is similar in shape to the cases growth rate.

References

  1. Barry D, Hartigan JA. A Bayesian Analysis for Change Point Problems. J Am Stat Assoc. 1993;88(421):309--19.
  2. Brandenburg A. Quadratic growth during the 2019 novel coronavirus epidemic. 2020.

Possible effect on hospitalizations of the health policies implemented by the Veneto region

This works aimes at giving a first impression of the possible effect of the health policies implemented by the Veneto region in order to contain the spread of COVID-19.

In order to understand whether the containing measures helped slow down the spread of COVID-19, a predictive model based on the data collected until the 12th of March was compared to what was actually observed.

Figure 1 shows that there was a slowdown after the 12th of March: this day represents an epidemic change-point.

Thanks to the comparison between the predicted and actual values it was possible to estimate some quantities:

  1. The number of avoided hospitalizations in the Veneto region as of the 24th of March: 561 (95% C.I. 533 -- 589) (Figure 2)
  2. How much the epidemic slowed down compared to what was expected:
    • 3.23 (95% C.I. 2.53 - 3.93) days were "gained" in terms of hospitalizations as of the 24th of March (Figure 3)
    • the epidemic velocity registered a drop of 64.84 hospitalizations/day (95% C.I. 61.06 -- 68.63) (Figure 4)

Figure 1. Expected hospitalizations (bold green curve; the other two green curves indicate the 95% confidence levels) based on course of the epidemic as registered until the 12th of March. Actual values (red dots) observed in the following days.

Figure 2. Avoided hospitalizations in the Veneto region compared to what was expected from the data gathered until the 12th of March. The grey area indicates the 95% confidence interval.

Figure 3. Gained days, estimated by looking at the shift to the right of the curve (predicted vs observed). The grey area indicates the 95% confidence interval.

Figure 4. Slowdown of the epidemic velocity (predicted vs observed). The grey area indicates the 95% confidence interval.

Technical details regarding the estimation of the model

The estimation of the model was based on the number series of hospitalizations that were observed until the 12th of March. This day represents a change-point in terms of growth of the epidemics. This change in the number series was detected by a Bayesian Changepoint Detection Method (1). The polynomial regression model is based on a local approximation of the regression function (smoothing parameter equal to 0.7). The shape of the curve fits the quadratic trend of the early stage of the outbreak.

Recent studies showed that the curve of cases could be of a quadratic nature rather than exponential, especially in the early stage of the outbreak (2).

It is assumed that the hospitalizations growth rate is similar in shape to the cases growth rate.

References

  1. Barry D, Hartigan JA. A Bayesian Analysis for Change Point Problems. J Am Stat Assoc. 1993;88(421):309--19.
  2. Brandenburg A. Quadratic growth during the 2019 novel coronavirus epidemic. 2020.

Impact of statistical uncertainty on COVID-19 predictions

How to read and use the graphs

In the graphs the daily new confirmed cases (red dots), by region or for the whole country, vs the prediction of what will happen in the future if we assume logistic growth (black dots).

It is possible to see how changing the parameters modifies the predictions. The range given for each parameter is the 99% CI of the values that best explain the data gathered until now.

Varying the parameters for Italy (within the 99% CI) also results in a proportional change in all respective regional parameters.

Expected number of total cases in Alessandria

The estimation of the models are based on the data collected until the 19th of March; the models are used to predict the number of total cases for the next three days.

Scenario 1. The first scenario uses Local Polinomial Regression Estimation with smoothing parameter alpha equal to 0.75. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 2. The second scenario is based on a Generalized Additive Model (GAM) where a natural spline was used to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 3. The third scenario is based on a Poisson Model (using an offset to account for the place of residency) with a natural spline to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Expected number of total cases in Vercelli

The estimation of the models are based on the data collected until the 19th of March; the models are used to predict the number of total cases for the next three days.

Scenario 1. The first scenario uses Local Polinomial Regression Estimation with smoothing parameter alpha equal to 0.75. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 2. The second scenario is based on a Generalized Additive Model (GAM) where a natural spline was used to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 3. The third scenario is based on a Poisson Model (using an offset to account for the place of residency) with a natural spline to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Expected number of total cases in Novara

The estimation of the models are based on the data collected until the 19th of March; the models are used to predict the number of total cases for the next three days.

Scenario 1. The first scenario uses Local Polinomial Regression Estimation with smoothing parameter alpha equal to 0.75. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 2. The second scenario is based on a Generalized Additive Model (GAM) where a natural spline was used to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Scenario 3. The third scenario is based on a Poisson Model (using an offset to account for the place of residency) with a natural spline to account for non-linearity. The expected number of total cases for the next days are reported on the graph together with the 95% CI.

Possible effect on ICU admissions of the health policies implemented by the Veneto region

Figure 1. Observed vs actual number of people admitted to the ICU (fit loess 0.7 span, 2 degrees). The red curve represents the expected number of people that would have been in need of intensive care if no containment policies were implemented and it is estimated using the data gathered until the 4th of March; the green curve represents what was actually observed until the 18th of March projected 5 days into the future (until the 23rd of March). The estimate sets by default the permanence in the ICU at 10 days. By moving the slider it is possible to see how the predictions change if we assume a different average length of stay (7 to 28 days)

Possible effect of the health policies implemented by the Piemonte region

This works aimes at giving a first impression of the possible effect of the health policies implemented by the Piemonte region in order to contain the spread of COVID-19.

In order to understand whether the containing measures helped slow down the spread of COVID-19, a predictive model based on the data collected until the 9rd of March was compared to what was actually observed.

Figure 1 shows that there was a slowdown after the 9th of March: this day represents an epidemic change-point.

Thanks to the comparison between the predicted and actual values it was possible to estimate some quantities:

  1. 1. The number of avoided cases in the Piemonte region as of the 15th of March: 545 (95% C.I. 461 -- 629) (Figure 2)
  2. 2. How much the epidemic has slowed down compared to what was expected:
    • a. a. 2.43 (95% C.I. 1.59 -3.27) days were "gained" as of the 15th of March (Figure 3)
    • b. b. the epidemic velocity registered a drop equal to 95 cases/day (95% C.I. 84 -- 106) (Figure 4)

Figure 1. Estimated cases (bold green curve) based on course of the epidemic as registered until the 9th of March. Actual values (red dots) observed in the following days.

Figure 2. Avoided cases in the Piemonte region compared to what was expected from the data gathered until the 9th of March. The grey area indicates the 95% confidence interval.

Figure 3. Gained days, estimated by looking at the shift to the right of the curve (predicted vs observed). The grey area indicates the 95% confidence interval.

Figure 4. Slowdown of the epidemic velocity (predicted vs observed). The grey area indicates the 95% confidence interval.

Technical details regarding the estimation of the model

The estimation of the model was based on the number series of the cases that were observed until the 9th of March. This day represents a change-point in terms of growth of the epidemics. This change in the number series was detected by a Bayesian Changepoint Detection Method (1). The polynomial regression model is based on a local approximation of the regression function (smoothing parameter equal to 1.5). The shape of the curve fits the quadratic trend of the early stage of the outbreak.

Recent studies showed that the curve of cases could be of a quadratic nature rather than exponential, especially in the early stage of the outbreak (2).

References

  1. Barry D, Hartigan JA. A Bayesian Analysis for Change Point Problems. J Am Stat Assoc. 1993;88(421):309--19.
  2. Brandenburg A. Quadratic growth during the 2019 novel coronavirus epidemic. 2020.

Possible effect of the health policies implemented by the Friuli Venezia Giulia region

This works aimes at giving a first impression of the possible effect of the health policies implemented by the Friuli Venezia Giulia region in order to contain the spread of COVID-19.

In order to understand whether the containing measures helped slow down the spread of COVID-19, a predictive model based on the data collected until the 10th of March was compared to what was actually observed.

Figure 1 shows that there was a slowdown after the 10th of March: this day represents an epidemic change-point.

Thanks to the comparison between the predicted and actual values it was possible to estimate some quantities:

  1. 1. The number of avoided cases in the Veneto region as of the 14th of March: 118 (95% C.I. 102 -- 135) (Figure 2)
  2. 2. How much the epidemic is slowing down compared to what was expected:
    • a. 1.61 (95% C.I. 0.57 - 2.66) days were "gained" as of the 14th of March (Figure 3)
    • b. the epidemic velocity registered a drop equal to 44 cases/day (95% C.I. 40 -- 47) (Figure 4)

Figure 1. Estimated cases (bold green curve) based on course of the epidemic as registered until the 10th of March. Actual values (red dots) observed in the following day.

Figure 2. Avoided cases in the Friuli Venezia Giulia region compared to what was expected from the data gathered until the 10th of March. The grey area indicates the 95% confidence interval.

Figure 3. Gained days, estimated by looking at the shift to the right of the curve (predicted vs observed). The grey area indicates the 95% confidence interval.

Figure 4. Slowdown of the epidemic velocity (predicted vs observed). The grey area indicates the 95% confidence interval.

Technical details regarding the estimation of the model

The estimation of the model was based on the number series of the cases that were observed until the 10th of March. This day represents a change-point in terms of growth of the epidemics. This change in the number series was detected by a Bayesian Changepoint Detection Method (1). The polynomial regression model is based on a local approximation of the regression function (smoothing parameter equal to 1.5). The shape of the curve fits the quadratic trend of the early stage of the outbreak.

Recent studies showed that the curve of cases could be of a quadratic nature rather than exponential, especially in the early stage of the outbreak (2).

References

  1. Barry D, Hartigan JA. A Bayesian Analysis for Change Point Problems. J Am Stat Assoc. 1993;88(421):309--19.
  2. Brandenburg A. Quadratic growth during the 2019 novel coronavirus epidemic. 2020.

Possible effect of the health policies implemented by the Veneto region

This works aimes at giving a first impression of the possible effect of the health policies implemented by the Veneto region in order to contain the spread of COVID-19.

In order to understand whether the containing measures helped slow down the spread of COVID-19, a predictive model based on the data collected until the 3rd of March was compared to what was actually observed.

Figure 1 shows that there was a slowdown after the 2nd of March: this day represents an epidemic change-point.

Thanks to the comparison between the predicted and actual values it was possible to estimate some quantities:

  1. The number of avoided cases in the Veneto region as of the 12th of March: 348 (95% C.I. 322 -- 373) (Figure 2)
  2. How much the epidemic is slowing down compared to what was expected:
    • 2.4 (95% C.I. 2.05 -- 2.74) days were "gained" as of the 12th of March (Figure 3)
    • the epidemic velocity registered a drop of 15.91 cases/day (95% C.I. 11.99 -- 19.82), peaking on the 6th of March with 40 cases/day (Figure 4)

Figure 1. Estimated cases (bold green curve; the other two green curves indicate the 95% confidence levels) based on course of the epidemic as registered until the 2nd of March. Actual values (red dots) observed in the following days.

Figure 2. Avoided cases in the Veneto region compared to what was expected from the data gathered until the 2nd of March. The grey area indicates the 95% confidence interval.

Figure 3. Gained days, estimated by looking at the shift to the right of the curve (predicted vs observed). The grey area indicates the 95% confidence interval.

Figure 4. Slowdown of the epidemic velocity (predicted vs observed). The grey area indicates the 95% confidence interval.

Technical details regarding the estimation of the model

The estimation of the model was based on the number series of cases that were observed until the 2nd of March. This day represents a change-point in terms of growth of the epidemics. This change in the number series was detected by a Bayesian Changepoint Detection Method (1). The polynomial regression model is based on a local approximation of the regression function (smoothing parameter equal to 1.5). The shape of the curve fits the quadratic trend of the early stage of the outbreak.

Recent studies showed that the curve of cases could be of a quadratic nature rather than exponential, especially in the early stage of the outbreak (2).

References

  1. Barry D, Hartigan JA. A Bayesian Analysis for Change Point Problems. J Am Stat Assoc. 1993;88(421):309--19.
  2. Brandenburg A. Quadratic growth during the 2019 novel coronavirus epidemic. 2020.

National events

Instructions

Visualize/hide one or more measures on the graph by clicking on the items in the legend. Double click to only visualize the selected item.

Click on the autoscale button (the third one) to maximize the size of the graph.

Time series trend - cumulative events

Time series trend - daily new events

Regional events

Istructions

Add/remove computations for one or more region/measure by adding/removing it from the box.

NOTE: The number of tests per day can be visualized on the graph (by selecting the correct item in the legend) but it is hidden by default as it is off the scale (compared to the other measures).

Visualize/hide one or more regions/measures on the graph by clicking on the items in the legend. Double click to only visualize the selected item.

Click on the autoscale button (the third one) to maximize the size of the graph.

Time series by region

Cumulative events

Daily new events

Regional time series by event

Cumulative events

Daily new events

Provincial events

Add/remove computations for one or more region/measure by adding/removing it from the box.

Visualize/hide one or more regions on the graph by clicking on the items in the legend. Double click to only visualize the selected item.

Click on the autoscale button (the third one) to maximize the size of the graph.

Time series

Cumulative events

New events

Data Info

This app takes the needed data from the offical records that track the italian COVID-19 outbreak, on a national, regional and provincial level

Data are processed and made available by the Presidenza del Consiglio dei Ministri - Dipartimento di Protezione Civile (Italian Civil Protection Department) and licensed under CC-BY-4.0 as provided by the Ministero della Salute (Ministry of Health).

Data is usually updated daily at 6pm.

Further information about data attribution and availability can be found at the web page of covid19ita. .

Software

The app covid19ita was developed using R ver. 3.6.3 as an expansion package. The source code of the package and the app is freely available online on Github at https://github.com/UBESP-DCTV/covid19ita/ .

For the app development the following expansion packages were used: {shiny} ver. 1.4.0, {shinydashboard} v.0.7.1 and {golem} ver. 0.2.1.

The analyses were performed using functions of the following packages: {stats} ver. 3.6.3, and {gam} ver. 1.16.1

Our graphs were produced with the following packages: {ggplot2} ver. 3.3.0 and {plotly} ver. 4.9.2.

Note for R users

In addition to this app (that can be locally excecuted, after installing the package {covid19ita}, by running the code `run_app()`), the R package {covid19ita}, available on GitHub and licenced under CC-BY-4.0 , makes raw data available for direct use.

Principal indices

Deaths

Deceased over positive cases

Deceased over hospitalized

Recovered

Recovered over hospitalized

Self Isolation

Home confined over hospitalized

Growth rate (%) with respect to the previous day
(100% = no growth, >100% increasing, <100% decreasing)

Cases

Deaths

ICU

Intensive care

Regional percentage (fit loess, span = 1.5, degree = 2) of population that, respectively, has been admitted to the ICU (vertical axis) and has not been (as of today) hospitalized though being tested (this data was approximated by the total number of tests minus hospitalized patients), up to the day chosen in the slider. It is possible to see the entire evolution automatically by clicking play.

Table: regional evolution of bed occupancy with respect to the asymptomatic tests (weighted on the population)

Geographic distribution of the number of cases by province

Loading Geodata...