Mapping Spatiotemporal Diffusion of COVID-19 in Lombardy (Italy) on the Base of Emergency Medical Services Activities

Round 1

Reviewer 1 Report

The research of this paper is very meaningful. But in terms of content, it lacks innovation. The data is relatively unitary. It is recommended to increase the amount of data. In addition, whether the model can be applied to other regions?

Author Response

The research of this paper is very meaningful. But in terms of content, it lacks innovation. The data is relatively unitary. It is recommended to increase the amount of data. In addition, whether the model can be applied to other regions?

 

Thank you for the interest shown in our paper. The two main innovations in our work are relevant to the use of emergency calls by citizens and to the dispatches of ambulances to track early evolution of the Covid-19 epidemic, rather than being based on official laboratory diagnosis. Also the level of geographic granularity of the proposed analysis, performed in districts with almost uniform resident population, more statistically sound when compared to the usual administrative subdivisions applied in similar studies, represents an innovation. To corroborate these points, an extensive analysis of the literature was conducted, yet no study proposing a similar framework was identified. If the reviewer has knowledge of other studies that apply comparable methodologies, please provide us the reference, as it will be our strong interest to make a comparison to improve our work. To better highlight the introduced novelties, these paragraphs were added to the discussion in lines 509-532:

-           Data availability: official diagnoses are limited by testing capabilities and available resources, in particular in the first phase of the epidemic [10] due to the novelty of the pathology, posing a strong limitation when used as input for spatiotemporal analyses [9,11,31,32]. By using calls to the emergency number 112 or the following ambulances dispatches, the need of having a massive testing capacity to detect an abrupt increase in possible positive COVID-19 patients is by-passed: these alternative data sources are collected daily, with no delays due to external structures and protocols (as positivity to tests), thus providing a daily updated reference for comparison. Moreover, this approach is patient-centric: it is the patient that, based on the perceived symptoms, calls the 112 number that, following a first phone interview to verify the patient’s symptoms, dispatches the ambulance. As both the emergency calls and the ambulance dispatches resulted, on a provincial level, highly correlated with the official casualties due to COVID-19, they could be considered as representative of the phenomenology of COVID-19 early diffusion on the Lombardy territory.

-           Geographic granularity: a localized early response is vital in mitigating the effects of epidemics spreading [13,14] by implementing interventions tailored on specific territorial needs and allowing a more accurate optimization of available healthcare resources. The proposed framework gives important insight considering specific geographical areas with similar numbers of resident population, that could be further refined to better highlight specific locations at risk where to suggest early lock-down preventive measures. To obtain sufficient granularity, clusters of homogeneous resident population (around 100000 citizens) where defined as a compromise between spatial resolution and the need of having enough events to trigger a visible increase of cases in a specific cluster. To do so, an original method to aggregate municipalities was developed and applied.

Concerning the increase in the amount of data, we would be more than happy to test the same framework on other territories, as the proposed analysis can be easily transposed to other regions, as long as their EMS provider does collect and make the data available; unfortunately, at the current stage, we have no ongoing collaboration with other EMS providers outside of AREU (provider for Regione Lombardia) , in other regions of Italy or abroad, so we could not access those data.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript ‘Mapping spatiotemporal diffusion of COVID-19 in Lombardy, Italy on the base of emergency medical services activities’ by Gianquintieri et al evaluates the spread of the COVID-19 pandemic in northern Italy during the early portion of 2020. They introduce an approach of using ambulance calls and emergency calls to investigate early spatial spread of COVID-19, since early testing and diagnosis was overwhelmed. They hypothesize that these approaches will allow us to better picture the spread and growth of the pandemic in the absence of high quality or confirmed clinical data.

Overall, this paper presents an interesting idea. They point out the well known fact that COVID-19 was likely present and spreading before clinical acknowledgment of an outbreak, and evaluated this using surrogate of ambulance/emergency calls. I found the paper interesting to read and they explained some topics quite well, such as their description of potential morphologies using their COVID score. However, the study also has some limitations. Most importantly, the Authors need to improve some of their Figures and Tables to make them more interpretable and friendly to readers. In addition, the paper could use an enhanced discussion that places their approach in context of other research done for other morbidities. I recommend the Authors address these and other limitations to improve their paper.

Major Revisions/Comments

1. One of the limitations of this study is a robust comparison between the Authors findings and the existing literature. With only 12 cited papers on a 25 page article, there is a lack of background and interpretation. While the Authors approach may be novel for COVID-19, similar methods have been applied elsewhere, including the use of internet searches to identify trends with pandemic outbreaks. There is room for the Authors to discuss this work. It would also be beneficial for them to provide more comparisons for their findings of pandemic spatio-temporal origins and official numbers stating when the pandemic started in the same regions.
2. In the Authors method, they constructed an algorithm to target municipalities that achieved their target underlying population. However, some of the territories (see Fig 4) are quite large. Since the Authors are using ER calls to evaluate underlying spread of disease, it is a valid assumption that ER calls are equally used in all of these regions? For some areas that appear more rural, are there other medical sources used to evaluate emergency conditions (such as calls to a clinic or local physician). The presence of these other sources may lead to potential bias in their outcomes. Can the Authors address this concern or discuss it as a possible limitation?
3. The Figures and Tables in this manuscript need some additional work. While I found many of them presented important information, a total of 13 Figures is quite voluminous and perhaps some might be better for the Supplement. Please see the specific comments below:

• In Figure 3 and Figure 9, the mapping choice of the Authors is misleading. They used a discrete color scale for their number of casualties/day of inflection point, yet the color gradients represent different ranges. For example, the first two color jumps for Fig 3 Panel E represent a change of 0.004 casualties/1000 people, but the last two color jumps represent 0.106 casualties/1000 people. It is strongly urged the Authors use a continuous scale and avoid this type of inconsistent visualization.
• For Figure 9, consider making the dates and colors the same for both panels. This will enhance comparability between the Calls vs. ambulance dispatches.
• In Figure 3, are these correlations based on day of casualties with day of ER calls? If so, how do the Authors consider the possibility of a lagged period? Lagged responses between those seeking emergency care and the outcome of mortality has been demonstrated for COVID-19. Additionally, as printed, the scales for the Panels B, D, and E are very small. The Authors should make these larger, especially since space is available along the left margin of their maps.
• For Figure 6, 8, 10, 11, 12, and 13, what is the Authors reasoning for reporting time as Days from Jan 1^st? It would be far more useful to have the x-axis as actual dates, which would increase interpretability from readers, especially since the Authors refer to dates when referencing official statistics.

Minor Required Revisions

• In the Methods, how were respiratory calls classified? If this was based on a specific health code equivalent, the Authors should make a brief note of it for reproducibility purposes.

Editorial/Grammatical Revisions

• The Authors or the Editor of the journal must proofread the article for grammatical checks. Articles are missing in multiple places and misused conjugation (e.g. Line 43 should read ‘testing began’). Many sentences are not in appropriate English language. Capitalizing is not consistent, such as in the Figure captions.I recognize that for the Authors, English may not be a first language, which may be the cause of this lapse.
• It is not clear what ‘OSM Standard’ means in Figure 1.
• On Line 155 & 157, did the Authors want a hyphen instead of a ‘divide by’ symbol for their age range percentage?

Minor Discretionary Revisions

• Can the ‘In Press’ citation on Line 64 be updated?
• On Line 397, can a citation be provided for that data?

Comments for author File: Comments.docx

Author Response

The manuscript ‘Mapping spatiotemporal diffusion of COVID-19 in Lombardy, Italy on the base of emergency medical services activities’ by Gianquintieri et al evaluates the spread of the COVID-19 pandemic in northern Italy during the early portion of 2020. They introduce an approach of using ambulance calls and emergency calls to investigate early spatial spread of COVID-19, since early testing and diagnosis was overwhelmed. They hypothesize that these approaches will allow us to better picture the spread and growth of the pandemic in the absence of high quality or confirmed clinical data.

 

Overall, this paper presents an interesting idea. They point out the well known fact that COVID-19 was likely present and spreading before clinical acknowledgment of an outbreak, and evaluated this using surrogate of ambulance/emergency calls. I found the paper interesting to read and they explained some topics quite well, such as their description of potential morphologies using their COVID score. However, the study also has some limitations. Most importantly, the Authors need to improve some of their Figures and Tables to make them more interpretable and friendly to readers. In addition, the paper could use an enhanced discussion that places their approach in context of other research done for other morbidities. I recommend the Authors address these and other limitations to improve their paper.

 

Major Revisions/Comments

 

One of the limitations of this study is a robust comparison between the Authors findings and the existing literature. With only 12 cited papers on a 25 page article, there is a lack of background and interpretation. While the Authors approach may be novel for COVID-19, similar methods have been applied elsewhere, including the use of internet searches to identify trends with pandemic outbreaks. There is room for the Authors to discuss this work. It would also be beneficial for them to provide more comparisons for their findings of pandemic spatio-temporal origins and official numbers stating when the pandemic started in the same regions.

 

We thank the reviewer for the suggestion, we agree that a consistent part of literature analysis was missing. New references were now integrated, also including an analysis and a comparison with our work, both in the introduction and in the discussion.

Introduction

Lines 60-121:

As of today (September 25th, 2020), COVID-19 pandemic affected 187 countries for a total of more than 32.2 million of confirmed cases [2], growing day-by-day. The worldwide scientific community is undergoing unprecedented efforts in an attempt to mitigate the effects of the pandemic, especially focusing on the application of edging technologies: several studies identified artificial intelligence (AI), machine learning (ML) and big data analytics as key assets in extracting information from available data useful to fight the pandemic [3-9]. In particular, Bragazzi et al. [4] identified four different fields of application on three time-horizons: rapid identification of outbreaks and diagnosis of cases (short-term), identification of therapeutic options (medium-term) and development of resilient smart-cities (long-term). On the same page is Ting et al. [8], where monitoring, surveillance, detection, prevention, and mitigation of indirect effects were defined as goals for the application of new technologies such as big data, AI, Internet of Things (IoT) and blockchain. These authors also focused on the long-term impact of this unprecedented situation, stating that a successful application for these technologies in public health, forced by the emergency condition, represents a unique opportunity to trigger a paradigm shift, capable of conditioning the whole future of policies in public health and medicine.

With specific reference to data collection and processing, real-time data visualization, enabled by computational techniques, is a powerful mean to spread information [4], but even more important goals can be reached in the field of prevention using epidemiological data to raise awareness and guide interventions [3, 10]. In particular, the concept of ‘surveillance’ has being re-addressed to face this new scenario thus pointing out the possibilities of new approaches such as digital and event-based surveillance [11]. The vital contribution of ‘syndromic surveillance’ in controlling and mitigating the effects of the epidemic have been recently pointed out [12] together with critical issues relevant to the difficulties in obtaining syndromic surveillance intelligence at the local level, as already evidenced with reference to the Ebola outbreak [13].

In this view, the focus on the spatial analysis, in relation to the COVID-19 pandemic, was recently explored in a systematic review [14], in which 5 related topics were identified: 1) Spatiotemporal analysis, in which a descriptive and/or predictive modeling of the evolution of the pandemic within a certain territory (usually national and provincial level) is explored using official data on positive cases, often also considering people mobility, with examples relevant to China [15] South Korea [16], USA [17], and Italy [18]; 2) Health and social geography, in which the relationship between epidemic spreading (based on confirmed cases) and healthcare resources [19], nurses [20] or surgeons [21], is explored, together with the correlation between confirmed cases and demographic and/or socio-economic characteristics [22, 23]; 3) Environmental variables, in which the correlation between confirmed cases and environmental factors, mainly climatic variables [24], such as humidity and temperature [25], is explored; 4) Data mining , in which the analysis of additional data sources, such as mobility [26, 27] and flights [28], to corroborate spatial analysis, is performed; 5) Web-based mapping, in which web services implementation for results visualization are presented.

From this review, the potentialities of Geographic Information Systems (GISs) as a set of tools for capturing, storing, checking, manipulating, analyzing and displaying spatially georeferenced data to handle the geospatial component of the pandemic analysis also at the local level was highlighted, in line with other studies [29,30].

            While the literature relevant to spatiotemporal analysis of COVID-19 is mostly based on data from official diagnoses, several studies identified such data as potentially unreliable for modelling and analysis [31, 32, 9, 11], in particular during the first phases of the epidemic when they were mostly needed [10], leading to study the possibility to exploit crowdsourced data, such as news media and social networks [10, 12, 14].

With specific reference to Italy, this scenario poses a strong limitation on the validity of any retrospective analysis concerning the spatiotemporal evolution of the epidemic based on positive cases, since a whole month of official data was probably missing during the early spreading of the disease., when possible cases of COVID-19 were probably diagnosed as other common respiratory diseases due to seasonal flu. Moreover, even after the first cases were diagnosed and the testing policy rearranged, complete and reliable data could not be guaranteed as due to quick saturation of the diagnosis capabilities, with the swab test for COVID-19 performed only in the most severe patients [33,34]. This resulted in a higher Cases Fatality Rate (CFR, number of casualties / total number of diagnosed patients, set at 11.31% in September 2020) compared to the estimated Infection Fatality Rate (IFR, number of casualties / total number of infected patients, considered to be between 0.5% and 1% [35]). Moreover, CFR values were consistently changing among Italian regions, as a consequence of the different regional testing policies and diagnostic capabilities: for example, on May 4th (date of the first reduction of lockdown measures) CFR was 18.3% in Lombardy (with 14’294 casualties and 78’105 confirmed cases), but only 8.3% in the adjacent region of Veneto (1’528 casualties and 18’373 confirmed cases).

 

Discussion

Lines 499--532:

Monitoring changes in healthcare utilization on a certain territory has been recently proposed as a key to interpret COVID-19 surveillance data, where ambulances dispatches were described as one of the possible indicators to be taken into account (yet, not developing a methodology in detail)  [12]. The need for using alternative and citizens-driven data sources has been previously identified in relation to spatial analysis of COVID-19 [14], for example assessing the informative content of social media [10,12]. In this context, the calls to 112 emergency phone number represent a pure citizen-driven parameter, while the EMS dispatches can be considered as the result of a filtering process of the calls by healthcare professionals.

 The proposed framework addresses two very relevant issues that are highlighted in literature as potential barriers:

-           Data availability: official diagnoses are limited by testing capabilities and available resources, in particular in the first phase of the epidemic [10] due to the novelty of the pathology, posing a strong limitation when used as input for spatiotemporal analyses [9,11,31,32]. By using calls to the emergency number 112 or the following ambulances dispatches, the need of having a massive testing capacity to detect an abrupt increase in possible positive COVID-19 patients is by-passed: these alternative data sources are collected daily, with no delays due to external structures and protocols (as positivity to tests), thus providing a daily updated reference for comparison. Moreover, this approach is patient-centric: it is the patient that, based on the perceived symptoms, calls the 112 number that, following a first phone interview to verify the patient’s symptoms, dispatches the ambulance. As both the emergency calls and the ambulance dispatches resulted, on a provincial level, highly correlated with the official casualties due to COVID-19, they could be considered as representative of the phenomenology of COVID-19 early diffusion on the Lombardy territory.

-           Geographic granularity: a localized early response is vital in mitigating the effects of epidemics spreading [13,14] by implementing interventions tailored on specific territorial needs and allowing a more accurate optimization of available healthcare resources. The proposed framework gives important insight considering specific geographical areas with similar numbers of resident population, that could be further refined to better highlight specific locations at risk where to suggest early lock-down preventive measures. To obtain sufficient granularity, clusters of homogeneous resident population (around 100000 citizens) where defined as a compromise between spatial resolution and the need of having enough events to trigger a visible increase of cases in a specific cluster. To do so, an original method to aggregate municipalities was developed and applied.

Lines 592-595:

This observation is in line with a key concept of modern spatial analysis, expressed by Escolano [43] as “all objects (on the earth's surface) are related to each other, but the relationships are more intense among the better connected objects regardless of their proximity”, whose applicability to the COVID-19 pandemic was already proposed [18,44].

 

Concerning the comparison between our results and the official data, it is unfortunately not possible, as official data were made available only at level of Provinces, administrative areas that are much larger than the considered population-based districts. We are aware that the lacking of a validation methodology is an important limit for this work, as discussed in the limitations section at lines 637-641:

• Validation of the results: the applied method was aimed at inferring a part of information that was not available elsewhere (start day of the exponential growth of epidemic spreading), and it was therefore impossible to validate the developed methodology against a gold standard. To minimize this limitation, both NEC and COVID-19 were proposed, thus obtaining a possible cross-comparison between their results.

 

 

In the Authors method, they constructed an algorithm to target municipalities that achieved their target underlying population. However, some of the territories (see Fig 4) are quite large. Since the Authors are using ER calls to evaluate underlying spread of disease, it is a valid assumption that ER calls are equally used in all of these regions? For some areas that appear more rural, are there other medical sources used to evaluate emergency conditions (such as calls to a clinic or local physician). The presence of these other sources may lead to potential bias in their outcomes. Can the Authors address this concern or discuss it as a possible limitation?

 

We thank the reviewer for pointing out this possible limitation, that was discussed and added at lines 647-656:

• Discrepancies in the availability and management of healthcare resources: although AREU is the manager and provider of EMS for all Lombardy, the distribution of the resources (especially ambulances) can be affected by local availability, as they strongly rely on the contribution of volunteers and private associations. Moreover, the unprecedented emergency in the early stages of the pandemic forced a rapid response and a deep re-organization of protocols at the local level, before the establishment of centralized directives. In the proposed framework, the demand for EMS was analysed in detail with high geographical granularity, but the spatial differences of service availability were not taken into account; this approximation was anyway considered acceptable, as the target of the study was the identification of a starting point, hence preceding the eventual saturation of availability.

 

The Figures and Tables in this manuscript need some additional work. While I found many of them presented important information, a total of 13 Figures is quite voluminous and perhaps some might be better for the Supplement. Please see the specific comments below:

 

In Figure 3 and Figure 9, the mapping choice of the Authors is misleading. They used a discrete color scale for their number of casualties/day of inflection point, yet the color gradients represent different ranges. For example, the first two color jumps for Fig 3 Panel E represent a change of 0.004 casualties/1000 people, but the last two color jumps represent 0.106 casualties/1000 people. It is strongly urged the Authors use a continuous scale and avoid this type of inconsistent visualization.

 

Thanks to the reviewer for the suggestion, color scales were modified in continuous scales. Please, notice that additional modifications for figures 3 and 9 (e.g. different legend) were implemented following also the suggestions by another reviewer.

 

For Figure 9, consider making the dates and colors the same for both panels. This will enhance comparability between the Calls vs. ambulance dispatches.

 

Thank you for your comment, now the color scale and the dates are the same.

 

In Figure 3, are these correlations based on day of casualties with day of ER calls? If so, how do the Authors consider the possibility of a lagged period? Lagged responses between those seeking emergency care and the outcome of mortality has been demonstrated for COVID-19. Additionally, as printed, the scales for the Panels B, D, and E are very small. The Authors should make these larger, especially since space is available along the left margin of their maps.

 

We thank the reviewer for addressing this issue. As reported in the legend, the dots represent geographical administrative areas (Provinces), where the corresponding X coordinate values represent the cumulated casualties due to COVID-19 in the period from Jan 1^st 2020 to Mar 30^th 2020 (as reported by our national statistics institute, ISTAT), and the Y coordinate values represent the cumulated number of ambulances dispatches in the period from Jan 1^st to Mar 23^rd (thus taking into account also the lagged response. As a result, the correlations are not on daily but on cumulated data over the whole analysis period, separately for different provinces. With reference to the legend, it was modified also following the suggestion from another reviewer.

 

For Figure 6, 8, 10, 11, 12, and 13, what is the Authors reasoning for reporting time as Days from Jan 1st? It would be far more useful to have the x-axis as actual dates, which would increase interpretability from readers, especially since the Authors refer to dates when referencing official statistics.

 

Thank you for the suggestion, the axes were re-labelled with dates.

 

Minor Required Revisions

 

In the Methods, how were respiratory calls classified? If this was based on a specific health code equivalent, the Authors should make a brief note of it for reproducibility purposes.

 

The respiratory calls were classified on the base of a specific field in the database that contains the result of the medical triage. The following statement was added in the text to specify this point at line 184:

            by a specific field relevant to medical triage.

 

Editorial/Grammatical Revisions

 

The Authors or the Editor of the journal must proofread the article for grammatical checks. Articles are missing in multiple places and misused conjugation (e.g. Line 43 should read ‘testing began’). Many sentences are not in appropriate English language. Capitalizing is not consistent, such as in the Figure captions.I recognize that for the Authors, English may not be a first language, which may be the cause of this lapse.

 

We apologize for the inconvenience, the grammar was revised, we hope it is now more appropriate.

 

It is not clear what ‘OSM Standard’ means in Figure 1.

 

‘OSM Standard’ is the automatic labelling for OpenStreetMap web mapping service; the official attribution was added to all reported maps.

 

On Line 155 & 157, did the Authors want a hyphen instead of a ‘divide by’ symbol for their age range percentage?

 

Thank you for noticing this typo, it was corrected accordingly.

 

Minor Discretionary Revisions

 

Can the ‘In Press’ citation on Line 64 be updated?

 

As the cited article was published, the reference was updated accordingly.

 

On Line 397, can a citation be provided for that data?

 

The official diagnoses data are made available in Italy by Protezione Civile, as presented at line 171-172.

Author Response File: Author Response.pdf

Reviewer 3 Report

Reviewer’s report
on the manuscript entitled
Mapping spatiotemporal diffusion of COVID-19 in Lombardy (Italy) on the base of Emergency Medical Services activities
Submitted to “International Journal of Geo-Information” (Manuscript ID: ijgi- 933934)

General: In this study, a signal processing method was applied to identify the beginning of anomalous trends in emergency calls and Emergency Medical Services ambulances dispatches, and reconstruct a spatiotemporal evolution of COVID-19 on the territory of Lombardy region, Italy.

I think this is an interesting and up-to-date work dealing with the anticipation of COVID-19 trends in terms of diagnoses and casualties on a provincial level, and demonstrating how emergency calls and ambulance dispatches could be used as indicators for the spatiotemporal evolution of the epidemic.

The aim of the study and details about COVID-19 in Italy and the study area have been clearly stated in the Introduction. However, a few more relevant studies should be included in this section. The methods are appropriate and clearly presented, though the softwares used should be mentioned. Regarding the results, some details to the maps could be improved to ameliorate their presentation. In Discussion, the authors could accompany their findings with appropriate references. Detailed comments and suggestions are reported below:

1. Page 2, Figure 1: Figure B is the main map demonstrating the study area and figure A is a reference map for the location of the study area (Lombardy Region) in Italy. Thus, figure B is the area of interest and should be larger than figure A. For the same reason, Milan and Codogno should be included in figure B and not in figure A. Specifically, Codogno has already been marked in figure B with red zones, so just add the name of the town, whereas Milan could be demonstrated with another mark (e.g. black dot). Furthermore, the red zones in Lombardy and Veneto should be more clearly presented in the map, as figure Β will be enlarged. Finally, based on the fact you have used as background the map tiles of OSM (figure 1, 3, 4, 7 and 9), please use the OpenStreetMap Foundation recommended attribution. You can find details for OSM copyright and license in the following url:
https://www.openstreetmap.org/copyright/en

2. Introduction: In the Abstract you mentioned that the signal processing method used in this study, has already been implemented in another analysis related to biological data.I recommend to describe it in a few lines in the Introduction section, as well as to add the reference of the relevant paper.

3. Page 3, line 100: I suggest you to report a few recent studies implementing GIS and spatial analysis in the study of COVID-19.

4. Page 5, Figure 3: In Figures 3b, 3d and 3c, the numbers in the Lombardy provinces should be better located in each province. Maybe in the center of each spatial unit, as the location over the borders of a province make it difficult to read. Τhe white mask over the text of the numbers also makes it difficult to read, as the text is too small. I suggest you to remove the mask and use white color plus bold font for the numbers in dark-colored provinces, and black color plus bold font for the numbers in light-colored provinces. Moreover, the explanation of the number for each color is not necessary in the legend, since the numbers are presented in each province. Removing these items from the legend, you gain place to move the legend to a corner and maybe enlarge the map and the font size of the numbers. Finally, the letters A, B, C, D, E are too large compared to all other texts on the images. This comment should be take into account for all the figures of the manuscript.

5. Page 6, line 188: Please report the software you used to implement the algorithm.

6. Page 7-11, Time-series analysis: A well detailed description of the analysis! I suggest to report the software(s) you used for the analysis, as well.

7. In general, report the software(s) used in the methods, sections 2.2., 2.3., 2.4. and 2.5.

8. Page 10, Figure 7: The legend is to small to read and the numbers in the clusters (I suppose they are the clusters ID numbers), are not explained. Furthermore, add A and B indications on the panels.

9. Page 13, Figure 9: I recommend you to include orange-yellow between red and green in the color palette, as the shades of green and blue are too many and make it difficult to distinguish. Adding one or two extra colors to the palette, the classes of the other colors will be reduced and the corresponding day will be easily recognized among the clusters in the map.

10. Pages 17-20, Discussion: I understand COVID-19 is an ongoing pandemic. However, studies dealing with the spatiotemporal analysis of COVID-19 have been already published. Thus, I think a more in-depth discussion of the findings compared with results of other case studies and other methods should be included in this section. Among others, the following review paper may help you to find relevant studies: Pardo et al., 2020, Spatial analysis and GIS in the study of COVID-19. A review. Science of The Total Environment. 739, 140033.

Author Response

General: In this study, a signal processing method was applied to identify the beginning of anomalous trends in emergency calls and Emergency Medical Services ambulances dispatches, and reconstruct a spatiotemporal evolution of COVID-19 on the territory of Lombardy region, Italy.

 

 

I think this is an interesting and up-to-date work dealing with the anticipation of COVID-19 trends in terms of diagnoses and casualties on a provincial level, and demonstrating how emergency calls and ambulance dispatches could be used as indicators for the spatiotemporal evolution of the epidemic.

 

 

The aim of the study and details about COVID-19 in Italy and the study area have been clearly stated in the Introduction. However, a few more relevant studies should be included in this section. The methods are appropriate and clearly presented, though the softwares used should be mentioned. Regarding the results, some details to the maps could be improved to ameliorate their presentation. In Discussion, the authors could accompany their findings with appropriate references. Detailed comments and suggestions are reported below:

 

1. Page 2, Figure 1: Figure B is the main map demonstrating the study area and figure A is a reference map for the location of the study area (Lombardy Region) in Italy. Thus, figure B is the area of interest and should be larger than figure A. For the same reason, Milan and Codogno should be included in figure B and not in figure A. Specifically, Codogno has already been marked in figure B with red zones, so just add the name of the town, whereas Milan could be demonstrated with another mark (e.g. black dot). Furthermore, the red zones in Lombardy and Veneto should be more clearly presented in the map, as figure Β will be enlarged. Finally, based on the fact you have used as background the map tiles of OSM (figure 1, 3, 4, 7 and 9), please use the OpenStreetMap Foundation recommended attribution. You can find details for OSM copyright and license in the following url:

https://www.openstreetmap.org/copyright/en

 

Thanks to the reviewer for the suggestions, the figures were modified accordingly and the OpenStreetMap attribution was added to all maps reported in the figures.

 

2. Introduction: In the Abstract you mentioned that the signal processing method used in this study, has already been implemented in another analysis related to biological data. I recommend to describe it in a few lines in the Introduction section, as well as to add the reference of the relevant paper.

 

Thanks to the reviewer for the comment. The description of the original application for that algorithm, that we used just to find the beginning of the increase in the curves, along with its literature reference, can be found in the section 2.3, dedicated to time-series analysis, at lines 308-312:

In order to identify on each sequence a fiducial point t ̅  representing the beginning of the phenomenon under analysis (i.e., the day associated to the potential start of the epidemic exponential growth), due to the morphological similarity of this curve with that of the ventricular repolarization phase (T wave) on the electrocardiogram signal, a validated methodology [38] to robustly detect the inflection point developed in that context, was applied.

As discussed at line 150 “The proposed analysis was a composite process, and many methodological decisions were based upon the previous results”, and the application of this signal processing algorithm is a perfect example, as the decision to apply it followed the explorative analysis of the time-series morphologies , as well as the test of different algorithms that resulted unsatisfying.

 

3. Page 3, line 100: I suggest you to report a few recent studies implementing GIS and spatial analysis in the study of COVID-19.

 

Thanks to the reviewer for pointing this out: we completely agree that a consistent part of literature analysis was missing. Accordingly, the text was integrated as follows at lines 84-101:

In this view, the focus on the spatial analysis, in relation to the COVID-19 pandemic, was recently explored in a systematic review [14], in which 5 related topics were identified: 1) Spatiotemporal analysis, in which a descriptive and/or predictive modeling of the evolution of the pandemic within a certain territory (usually national and provincial level) is explored using official data on positive cases, often also considering people mobility, with examples relevant to China [15] South Korea [16], USA [17], and Italy [18]; 2) Health and social geography, in which the relationship between epidemic spreading (based on confirmed cases) and healthcare resources [19], nurses [20] or surgeons [21], is explored, together with the correlation between confirmed cases and demographic and/or socio-economic characteristics [22, 23]; 3) Environmental variables, in which the correlation between confirmed cases and environmental factors, mainly climatic variables [24], such as humidity and temperature [25], is explored; 4) Data mining , in which the analysis of additional data sources, such as mobility [26, 27] and flights [28], to corroborate spatial analysis, is performed; 5) Web-based mapping, in which web services implementation for results visualization are presented.

From this review, the potentialities of Geographic Information Systems (GISs) as a set of tools for capturing, storing, checking, manipulating, analyzing and displaying spatially georeferenced data to handle the geospatial component of the pandemic analysis also at the local level was highlighted, in line with other studies [29,30].

 

4. Page 5, Figure 3: In Figures 3b, 3d and 3c, the numbers in the Lombardy provinces should be better located in each province. Maybe in the center of each spatial unit, as the location over the borders of a province make it difficult to read. Τhe white mask over the text of the numbers also makes it difficult to read, as the text is too small. I suggest you to remove the mask and use white color plus bold font for the numbers in dark-colored provinces, and black color plus bold font for the numbers in light-colored provinces. Moreover, the explanation of the number for each color is not necessary in the legend, since the numbers are presented in each province. Removing these items from the legend, you gain place to move the legend to a corner and maybe enlarge the map and the font size of the numbers. Finally, the letters A, B, C, D, E are too large compared to all other texts on the images. This comment should be take into account for all the figures of the manuscript.

 

We thank the reviewer for the graphical inputs: figure 3 was modified accordingly; please, notice that additional modifications were applied following the suggestions also from another reviewer. All the panels identifying letters were resized.

 

5. Page 6, line 188: Please report the software you used to implement the algorithm.
6. Page 7-11, Time-series analysis: A well detailed description of the analysis! I suggest to report the software(s) you used for the analysis, as well.7. In general, report the software(s) used in the methods, sections 2.2., 2.3., 2.4. and 2.5.

 

Thanks to the reviewer for the interest. The information about  the utilized software (open source, or custom) was added as follows at lines 173-178:

For all the following procedures (sections 2.2, 2.3, 2.4, 2.5), the processing and management of georeferenced data, and the generation of maps, was performed using QGIS (https://www.qgis.org/it/site/), an open source GIS developed by the OSGeo foundation (https://www.osgeo.org/), whereas all the database pre-processing, the statistical analysis and the signal processing algorithms were implemented as custom software in MATLAB (Mathworks, https://it.mathworks.com/products/matlab.html) under academic licensing.

 

8. Page 10, Figure 7: The legend is too small to read and the numbers in the clusters (I suppose they are the clusters ID numbers), are not explained. Furthermore, add A and B indications on the panels.

 

Thanks to the reviewer for noticing this! The legend font was enlarged and the explanation of the numbers represented in the clusters (which do represent the #ID) was added in the figure legend.

 

9. Page 13, Figure 9: I recommend you to include orange-yellow between red and green in the color palette, as the shades of green and blue are too many and make it difficult to distinguish. Adding one or two extra colors to the palette, the classes of the other colors will be reduced and the corresponding day will be easily recognized among the clusters in the map.

 

Thanks to the reviewer for the very useful suggestion: color palette was updated, and the resulting maps are indeed more readable.

 

10. Pages 17-20, Discussion: I understand COVID-19 is an ongoing pandemic. However, studies dealing with the spatiotemporal analysis of COVID-19 have been already published. Thus, I think a more in-depth discussion of the findings compared with results of other case studies and other methods should be included in this section. Among others, the following review paper may help you to find relevant studies: Pardo et al., 2020, Spatial analysis and GIS in the study of COVID-19. A review. Science of The Total Environment. 739, 140033.

 

We are very grateful to the reviewer for addressing this issue and providing us with a very complete reference, which was essential in deepening the analysis of current literature. Both the introduction and the discussion were extended with new references, including an analysis and a comparison with our work. Modifications to the text are reported as follows.

Introduction

Lines 60-121:

As of today (September 25th, 2020), COVID-19 pandemic affected 187 countries for a total of more than 32.2 million of confirmed cases [2], growing day-by-day. The worldwide scientific community is undergoing unprecedented efforts in an attempt to mitigate the effects of the pandemic, especially focusing on the application of edging technologies: several studies identified artificial intelligence (AI), machine learning (ML) and big data analytics as key assets in extracting information from available data useful to fight the pandemic [3-9]. In particular, Bragazzi et al. [4] identified four different fields of application on three time-horizons: rapid identification of outbreaks and diagnosis of cases (short-term), identification of therapeutic options (medium-term) and development of resilient smart-cities (long-term). On the same page is Ting et al. [8], where monitoring, surveillance, detection, prevention, and mitigation of indirect effects were defined as goals for the application of new technologies such as big data, AI, Internet of Things (IoT) and blockchain. These authors also focused on the long-term impact of this unprecedented situation, stating that a successful application for these technologies in public health, forced by the emergency condition, represents a unique opportunity to trigger a paradigm shift, capable of conditioning the whole future of policies in public health and medicine.

With specific reference to data collection and processing, real-time data visualization, enabled by computational techniques, is a powerful mean to spread information [4], but even more important goals can be reached in the field of prevention using epidemiological data to raise awareness and guide interventions [3, 10]. In particular, the concept of ‘surveillance’ has being re-addressed to face this new scenario thus pointing out the possibilities of new approaches such as digital and event-based surveillance [11]. The vital contribution of ‘syndromic surveillance’ in controlling and mitigating the effects of the epidemic have been recently pointed out [12] together with critical issues relevant to the difficulties in obtaining syndromic surveillance intelligence at the local level, as already evidenced with reference to the Ebola outbreak [13].

In this view, the focus on the spatial analysis, in relation to the COVID-19 pandemic, was recently explored in a systematic review [14], in which 5 related topics were identified: 1) Spatiotemporal analysis, in which a descriptive and/or predictive modeling of the evolution of the pandemic within a certain territory (usually national and provincial level) is explored using official data on positive cases, often also considering people mobility, with examples relevant to China [15] South Korea [16], USA [17], and Italy [18]; 2) Health and social geography, in which the relationship between epidemic spreading (based on confirmed cases) and healthcare resources [19], nurses [20] or surgeons [21], is explored, together with the correlation between confirmed cases and demographic and/or socio-economic characteristics [22, 23]; 3) Environmental variables, in which the correlation between confirmed cases and environmental factors, mainly climatic variables [24], such as humidity and temperature [25], is explored; 4) Data mining , in which the analysis of additional data sources, such as mobility [26, 27] and flights [28], to corroborate spatial analysis, is performed; 5) Web-based mapping, in which web services implementation for results visualization are presented.

From this review, the potentialities of Geographic Information Systems (GISs) as a set of tools for capturing, storing, checking, manipulating, analyzing and displaying spatially georeferenced data to handle the geospatial component of the pandemic analysis also at the local level was highlighted, in line with other studies [29,30].

            While the literature relevant to spatiotemporal analysis of COVID-19 is mostly based on data from official diagnoses, several studies identified such data as potentially unreliable for modelling and analysis [31, 32, 9, 11], in particular during the first phases of the epidemic when they were mostly needed [10], leading to study the possibility to exploit crowdsourced data, such as news media and social networks [10, 12, 14].

With specific reference to Italy, this scenario poses a strong limitation on the validity of any retrospective analysis concerning the spatiotemporal evolution of the epidemic based on positive cases, since a whole month of official data was probably missing during the early spreading of the disease., when possible cases of COVID-19 were probably diagnosed as other common respiratory diseases due to seasonal flu. Moreover, even after the first cases were diagnosed and the testing policy rearranged, complete and reliable data could not be guaranteed as due to quick saturation of the diagnosis capabilities, with the swab test for COVID-19 performed only in the most severe patients [33,34]. This resulted in a higher Cases Fatality Rate (CFR, number of casualties / total number of diagnosed patients, set at 11.31% in September 2020) compared to the estimated Infection Fatality Rate (IFR, number of casualties / total number of infected patients, considered to be between 0.5% and 1% [35]). Moreover, CFR values were consistently changing among Italian regions, as a consequence of the different regional testing policies and diagnostic capabilities: for example, on May 4th (date of the first reduction of lockdown measures) CFR was 18.3% in Lombardy (with 14’294 casualties and 78’105 confirmed cases), but only 8.3% in the adjacent region of Veneto (1’528 casualties and 18’373 confirmed cases).

 

Discussion

Lines 499--532:

Monitoring changes in healthcare utilization on a certain territory has been recently proposed as a key to interpret COVID-19 surveillance data, where ambulances dispatches were described as one of the possible indicators to be taken into account (yet, not developing a methodology in detail)  [12]. The need for using alternative and citizens-driven data sources has been previously identified in relation to spatial analysis of COVID-19 [14], for example assessing the informative content of social media [10,12]. In this context, the calls to 112 emergency phone number represent a pure citizen-driven parameter, while the EMS dispatches can be considered as the result of a filtering process of the calls by healthcare professionals.

 The proposed framework addresses two very relevant issues that are highlighted in literature as potential barriers:

-           Data availability: official diagnoses are limited by testing capabilities and available resources, in particular in the first phase of the epidemic [10] due to the novelty of the pathology, posing a strong limitation when used as input for spatiotemporal analyses [9,11,31,32]. By using calls to the emergency number 112 or the following ambulances dispatches, the need of having a massive testing capacity to detect an abrupt increase in possible positive COVID-19 patients is by-passed: these alternative data sources are collected daily, with no delays due to external structures and protocols (as positivity to tests), thus providing a daily updated reference for comparison. Moreover, this approach is patient-centric: it is the patient that, based on the perceived symptoms, calls the 112 number that, following a first phone interview to verify the patient’s symptoms, dispatches the ambulance. As both the emergency calls and the ambulance dispatches resulted, on a provincial level, highly correlated with the official casualties due to COVID-19, they could be considered as representative of the phenomenology of COVID-19 early diffusion on the Lombardy territory.

-           Geographic granularity: a localized early response is vital in mitigating the effects of epidemics spreading [13,14] by implementing interventions tailored on specific territorial needs and allowing a more accurate optimization of available healthcare resources. The proposed framework gives important insight considering specific geographical areas with similar numbers of resident population, that could be further refined to better highlight specific locations at risk where to suggest early lock-down preventive measures. To obtain sufficient granularity, clusters of homogeneous resident population (around 100000 citizens) where defined as a compromise between spatial resolution and the need of having enough events to trigger a visible increase of cases in a specific cluster. To do so, an original method to aggregate municipalities was developed and applied.

Lines 592-595:

This observation is in line with a key concept of modern spatial analysis, expressed by Escolano [43] as “all objects (on the earth's surface) are related to each other, but the relationships are more intense among the better connected objects regardless of their proximity”, whose applicability to the COVID-19 pandemic was already proposed [18,44].

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thank you for making the suggested manuscript changes.

Reviewer 3 Report

I want to thank you for the time and effort you put in revising the manuscript, as well as for providing detailed responses to my comments and suggestions. I consider the revised manuscript as suitable for publication.