CalCORVID: a dynamic RShiny dashboard approach to visualize spatiotemporal clusters for public health surveillance

TL;DR

CalCORVID is an open-source dashboard that helps public health officials visualize and share spatiotemporal disease cluster data for real-time surveillance.

Contribution

CalCORVID introduces a customizable RShiny dashboard to enhance SaTScan outputs for public health surveillance.

Findings

CalCORVID integrates SaTScan outputs with interactive maps and historical data for better cluster visualization.

The dashboard includes the Social Vulnerability Index to provide context for identified clusters.

Epidemiologists can adapt the tool for their jurisdictions to improve real-time disease monitoring.

Abstract

Infectious disease surveillance is an essential component of public health for preventing and mitigating outbreaks. Systematically applying statistical methods for anomaly detection to surveillance data can expedite outbreak response through early warning. A commonly used approach is the usage of spatiotemporal scan statistics as implemented in SaTScan, a software that analyzes spatiotemporal data to identify clusters of events over space and time that deviate from expected values. Some health departments identify outbreaks and prioritize resources using SaTScan for early cluster detection for diseases such as salmonellosis, legionellosis, and COVID-19. However, as a standalone software, SaTScan v10.2.1 does not provide functionality to easily disseminate visual cluster results over time in a way that is tailored to epidemiologists’ needs for real-time disease surveillance.

We developed an open source dashboard that provides a customizable framework for displaying results and facilitating the use of SaTScan for public health surveillance. The California Clustering for Operational Real-time Visualization of Infectious Diseases (CalCORVID) dashboard is built using RShiny, is specifically designed for SaTScan outputs, and can be easily adapted to display any jurisdiction’s results. This dashboard features a map and corresponding results table, the option to view historical results, integration of the Social Vulnerability Index (SVI) to contextualize clusters, and interactive elements to enhance usability for epidemiologists.

We present CalCORVID as a complementary tool to native outputs of SaTScan v10.2.1, allowing users to visualize, customize, and distribute their results for specific public health use cases. Epidemiologists currently using SaTScan can adapt the provided code repository and dashboard template to display their own jurisdictions’ results, facilitating dissemination of cluster results for real-time, ongoing disease surveillance.

The online version contains supplementary material available at 10.1186/s12889-026-26201-1.

Figures

Sequence diagram depicting the operational flow of the CalCORVID dashboard from data input to dashboard display. The user initiates the application by launching app.R via runApp(), which immediately sources global.R. The global.R lifeline performs data preparation tasks, including cleaning the latest SaTScan cluster results, merging with historical data, calculating Social Vulnerability Index values, and generating cluster geographies. If coordinate and county boundary files are missing, global.R also generates these resources. Once processing is complete, the cleaned dataset is returned to ap

Flow diagram of CDPH workflow culminating in the RShiny dashboard display of results. We pull our NAAT electronic laboratory reports (ELRs) from our data warehouse, preprocess the testing data into input files for SaTScan, run the SaTScan software, apply a bipartite network analysis to the SaTScan cluster results, and output the final results into the CalCORVID-based dashboard

Full text

Background

Infectious disease surveillance has many goals that include monitoring disease trends, establishing the current burden of disease, and identifying outbreaks [1]. Responding to disease outbreaks is a critical responsibility of health departments to prevent further transmission and to find potential outbreak sources [2]. To identify outbreaks, health departments typically rely on either facilities reporting outbreaks or epidemiologists monitoring the disease surveillance data to empirically determine whether there are anomalous patterns that merit investigation [2]. Both approaches may result in delayed outbreak identification, allowing further transmission to occur before detection and response. Therefore, methods that can reduce the time to outbreak detection are of great interest to public health practitioners. Spatiotemporal scan statistics, as implemented by SaTScan™ software, are one of the methods used in applied public health settings to improve time to outbreak detection [3–5]. In SaTScan, users provide an input dataset aggregated to a geographic unit (e.g., census tracts) along with the corresponding dates and associated number of cases or events. In applied public health settings, this input dataset often contains electronic laboratory reports because these are typically the first data source that a health department receives. The scan statistic conceptualizes each potential cluster as a 3-dimensional cylinder with the base and height corresponding to the spatial and temporal units, respectively. The scan statistic creates all possible combinations of these cylinders over a specified spatial area and temporal window, resulting in a large number of cylinders to scan over. Then, the scan statistic identifies anomalous clusters as cylinders with greater or lower observed events than expected when set to scan for high or low rates, respectively. For further technical details, please refer to Kulldorff et al. [6].

While numerous spatial clustering methods are available, the spatiotemporal scan statistic remains the preferred choice among public health departments due to its proven track record, extensive documentation, and compatibility with routinely available surveillance data. Compared to other hot spot analysis techniques such as the Getis-Ord Gi* statistic, Local Moran’s I, and kernel density estimation, SaTScan offers a model-based framework that adjusts for population at risk and provides formal statistical inference, making it well-suited for epidemiologic applications [7–9]. GeoMEDD, or Geographic Monitoring for Early Disease Detection, is a more recent method that emphasizes operational utility by identifying clusters based on repeated case occurrences within user-defined space and time windows [10]. While GeoMEDD and GIS-based hot spot tools are valuable for rapid detection and visualization, they often lack population adjustment and formal hypothesis testing. SaTScan’s spatiotemporal scan statistic offers a balance of statistical rigor, interpretability, and scalability that aligns well with the needs and constraints of public health practice, particularly when working with routinely collected case and population surveillance data.

SaTScan has been adopted by several public health departments for routine surveillance. The New York City Department of Health and Mental Hygiene automates these spatiotemporal analyses for over 35 reportable diseases and has utilized results for early outbreak detection and resource allocation [3, 11, 12]. Similarly, the New Jersey Department of Health implements SaTScan to aid in legionellosis surveillance and has used results for early outbreak detection and identification of additional cases [13]. Other health departments have also successfully utilized the scan statistic to identify outbreaks [13–17].

Although there is existing documentation and literature to aid in the development of a SaTScan model [6, 18], the dissemination of results for ongoing surveillance is less straightforward. Based on the selected output options, the SaTScan software produces several files for users to view results. The main output of SaTScan is a text file containing anomalous clusters listed in order of decreasing test statistic and associated cluster information. For users who want to manually visualize results, SaTScan provides spatial files in Keyhole Markup Language (KML) and shapefile formats that overlay the clusters on a map when imported into geographic software. Opening the KML file displays the clusters in Google Earth software while the shapefiles must be exported to geographic software [6]. Both approaches require some manual configuration for the end user; the KML file must be in a shared file location that all users can access and have Google Earth software installed, and the shapefiles require an understanding of how to manipulate spatial data or the user of licensed spatial software. These requirements may exceed technical and resource capacity for some health departments and, in the case of ongoing monitoring, introduces repetitive, manual, non-reproducible steps into the surveillance process. The built-in visualization solution for SaTScan users is a Hypertext Markup Language (HTML) file that overlays the model results on a map with options to customize the view. Users can choose to display clusters by recurrence interval, change the map zoom level, and remove or add grid lines indicating the latitude and longitude coordinates. These native SaTScan outputs can be shared either through an automated email that is sent once the model analysis completes or by users accessing the files in a shared file location. However, maintaining a list of users for the automated email list or configuring file sharing permissions can be cumbersome or impractical, and users may also want to tailor results for different audiences (e.g., sharing line list information for internal audiences only). An open source, R-based solution to display and share results is a tractable alternative approach with the growing usage of R and RShiny in public health.

In California, several of our 61 local health jurisdictions have already implemented SaTScan in some form as part of their disease surveillance efforts. We developed California Clustering for Operational Real-time Visualization of Infectious Diseases (CalCORVID) in conversation with epidemiologists across California to address their need for an accessible solution to display and share their cluster results in real time. CalCORVID is an open source, interactive dashboard built in RShiny that facilitates visualization, customizability, and distribution of SaTScan results. While there are other SaTScan-related dashboards, these dashboards are wrappers for the rsatscan package that enable running SaTScan within the R software [19, 20]. Our open source SaTScan dashboard is distinguished by its focus on use in applied public health settings and includes features such as the ability to customize the results display, view clusters over time, interact directly with cluster results, and incorporate socioeconomic spatial variables to contextualize findings. The primary user interactions with these features are illustrated in Supplementary Fig. 1, which presents the use case diagram for the dashboard.

Implementation

Key features of the CalCORVID dashboard include:

• Displaying cluster results without requiring any manipulation of spatial objects, which improves dashboard processing and usability.
• Results tab that features a map displaying clusters and corresponding cluster information.
• Date drop-down element to view historical data and compare results visually over time.
• Ability to visualize clusters overlaid with Social Vulnerability Index (SVI) to help contextualize results.
• Reactive elements to interact with SaTScan cluster results.
• Technical notes tab to describe model details for end user.
• Option to automate ongoing surveillance tasks using RShiny framework.

Implementation details

CalCORVID is written in the R programming language and is released under the MIT Open Source License (https://opensource.org/license/mit/). The current distribution includes CalCORVID dashboard source code, tutorials on how to use the tool, and sample datasets. The documentation also describes the overall structure of the dashboard and basic RShiny functionalities for individuals who may be interested in customizing CalCORVID for their own use.

The primary output of this repository is an open source dashboard code for users to adapt for their spatiotemporal cluster outputs from SaTScan. This dashboard can be used by epidemiologists conducting routine surveillance or sharing with other stakeholders who may be interested in visualizing disease clusters. The documentation describes the input files needed to display clusters correctly on the dashboard and the parameters to change for successful adaptation to individual use cases. A visual overview of the system architecture is provided in the component diagram (Supplementary Fig. 2), which illustrates how components interact to support the end-to-end process from SaTScan output to public health action.

CalCORVID is distributed with example data from the California Open Data Portal to demonstrate data preprocessing and dashboard display. An additional simulated dataset is provided and detailed in the README file to help users become more familiar with the structure of the dashboard before incorporating their own data.

User interface and distribution

We developed the CalCORVID dashboard using R (version 4.0.4.) [21] and Shiny (version 1.7.1) [22]. The required software and R packages necessary to run this dashboard are listed in Supplementary Table 1. By using the Shiny package, the dashboard can utilize reactive functions which allows the user to interact with widgets in the dashboard to filter and display further information.

The CalCORVID dashboard is based on two files that must be in the same directory: app.R and global.R. The app.R file contains code for the user interface (UI) and the server functions. Elements displayed on the dashboard’s visual interface are stored in the UI object, and must first be rendered by the server functions. The global.R file must be sourced prior to running app.R, which loads libraries, preprocesses data, loads data, and allows the user to specify function options. Although the global.R file is not required by Shiny, loading and preprocessing data in this file improves the performance of app.R.

In addition to these three files, the repository contains data, R and www folders, which are essential to run this dashboard. The data folder contains all the relevant raw data files needed for the dashboard, with subfolders for different inputs such as the coordinates file (“coords”), county boundary shapefiles (“county_boundary”), combined “gis” and “col” files (“giscol”), SaTScan cluster outputs (“satscan_output”), and Social Vulnerability Index (SVI) files (“svi”). The R folder holds the R scripts containing functions that are used to generate the data in the global.R file. Lastly, the www folder is RShiny-specific and is the directory that locally stores elements that are rendered in the web browser (e.g., images for the home page).

Data generating process and input file

Generating the dashboard input dataset is a separate process from what this open source repository provides. As part of their own workflow, users must obtain comma-separated value (CSV) files of cluster results after running their data in the SaTScan software, which can be specified from the SaTScan Outputs tab and checking the boxes corresponding to “Cluster Information” and “Location Information.” Checking these boxes will output files with “col” and “gis” file extensions, respectively. The “col” file contains the cluster centers and the columns the dashboard requires as shown in Table 1. The “gis” file contains the associated location IDs with each cluster center, which are used to calculate location-level spatial covariates.

For demonstration purposes, we include a sample results file generated from California COVID-19 vaccination data. We used publicly available, weekly California vaccination data aggregated at the ZIP code level from the California Open Data Portal, sourced June 3, 2024 [23]. We ran a prospective space-time analysis with the discrete Poisson probability model to detect low clusters of vaccination events for an arbitrary time frame from September 21, 2021, to February 8, 2022, to provide sample SaTScan results for display in the dashboard [24]. Supplementary Table 2 details the specific model parameters used to generate the sample results.

Table 1. Columns required for raw input data for both .col and.gis files that are generated by the SaTScan softwareSaTScan VariableDescriptionOriginating FileCLUSTERAssigned cluster number. This number is assigned by descending log likelihood ratios (i.e., the first cluster will have the highest log likelihood value.).col,.gisLOC_IDCluster center (.col) or location ID (.gis) in the given geographic unit. For example, if analysis is at the census tract level, these will correspond to the census tract at the center of the cluster and census tracts associated with each cluster, respectively..col,.gisLATITUDECluster center latitude coordinate..colLONGITUDECluster center longitude coordinate..colRADIUSCluster center radius in kilometers..colSTART_DATECluster start date..colEND_DATECluster end date. This date will always correspond to the most recent date your dataset is run on. For example, if the last date of your analysis is 1/30/24, the clusters will all have the end date of 1/30/24..colOBSERVEDObserved number of cases in cluster between START_DATE, END_DATE, and associated LOC_IDs..colEXPECTEDExpected number of cases in cluster between START_DATE, END_DATE, and associated LOC_IDs..colRECURR_INTRecurrence interval, or how often the observed cluster would be observed by chance, assuming analyses are repeated on a regular basis with a periodicity equal to the specified time interval length (6). For example, a recurrence interval of 100 days indicates that a cluster like the observed cluster would be observed only every 100 days (occurrence is unusual.).col

Pre-processing files

After the user obtains the required output files, some data preprocessing is necessary before running the dashboard. Figure 1 contains a sequence diagram that illustrates the operational flow from input to dashboard display. Table 2 details the different functions provided with the open source repository that facilitate this process, and sample code is provided in the global.R file. These functions are sourced and automatically run by global.R when app.R is run, but users can opt to run these functions individually to debug their code.

Briefly, global.R first checks whether their latest output files are in the correct formats for the dashboard using the clean_data() function. If they are, the historical file can be created with the combine_datasets() function which combines the cluster center and location ID files over the specified historical period. After combining the datasets for the desired time period, the average Social Vulnerability Index (SVI) percentiles across all detected geographies for each cluster are calculated using the generate_svi_vars() function. SVI is a socioeconomic index created by the CDC based on 16 US census variables and is commonly used to compare the social vulnerability between different geographic units [25]. There are four themes with percentile rankings: socioeconomic status, household characteristics, racial and minority status, and housing type and transportation, and an overall ranking for vulnerability [26]. Since the percentile ranking for a geographic unit is calculated relative to other geographic units in the dataset, percentiles may be slightly different depending on the unit of analysis chosen; therefore, the average percentile for a cluster containing ZIP code units will differ from the average percentile for a cluster containing census tract units, even if the ZIP codes contain all census tracts. After the SVI metrics are calculated for each cluster, we use generate_geo_layer() to obtain the spatial geometries for each of the detected clusters for display on the leaflet map. The last step is the clean_combined_datasets() function that reformats the combined file containing average SVI percentiles and spatial geometries for display on the CalCORVID dashboard. This generates a CSV file in the/data/folder on which the dashboard is based.

There are also two additional functions within global.R that are run the first time a new analysis is displayed on the dashboard. The generate_county_shapes() function creates a geojson spatial file that contains the county boundaries for the state or states encompassing the data. The resulting object is used as a layer with Leaflet, an open source JavaScript library used for interactive mapping, and will be detailed in a later section [27]. The R leaflet package adapts Leaflet functionality into R to generate the main basemap display on the CalCORVID dashboard [28]. After the base Leaflet map is created, we overlay spatial files such as county boundaries to provide additional spatial information. The second function that runs is either the generate_state_coords() or generate_cluster_coords() function, depending on the zoom level specified in the parameter settings. The generate_state_coords() function orients the Leaflet map to zoom to the state level of the data and the generate_cluster_coords() function orients the map to the center of the detected clusters.

Fig. 1. Sequence diagram depicting the operational flow of the CalCORVID dashboard from data input to dashboard display. The user initiates the application by launching app.R via runApp(), which immediately sources global.R. The global.R lifeline performs data preparation tasks, including cleaning the latest SaTScan cluster results, merging with historical data, calculating Social Vulnerability Index values, and generating cluster geographies. If coordinate and county boundary files are missing, global.R also generates these resources. Once processing is complete, the cleaned dataset is returned to app.R, which activates server and UI components to render the dashboard interface. The diagram illustrates synchronous messages between components and conditional logic based on file availability

Table 2. Functions included in the dashboard in the R folder, listed in the necessary order for preprocessingScript/FunctionPurposeOutputcheck_data.R/clean_data()Checks whether the latest output data from SaTScan is in the correct format to use for the dashboard.A printed statement confirming whether the output is in the correct format and all required columns are present. Otherwise, the dashboard will error out.check_data.R/combine_datasets()Combines cluster center (.col) files with location IDs files (.gis) using END_DATE and CLUSTER as identifiers to create a singular file over a specified time period.A CSV file in the/data/giscol_files folder containing the combined datasets over the specified time period.generate_map.R/generate_svi_vars()Calculates average SVI percentiles for the geographic unit of analysis for each cluster. This requires obtaining SVI for each geographic unit and averaging over all units in a cluster.A data frame object of the combined dataset and average SVI for each cluster.generate_map.R/generate_geo_layer()Downloads the spatial geographies (census tract, ZIP code, county) associated with each detected cluster using the tigris package.A data frame object containing the geographies associated with each detected cluster.check_data.R/clean_combined_datasets()Reformats the data frame object created with generate_svi_vars() to only include relevant variables, create a new variable converting radius to meters for leaflet map, and round numeric variables.A CSV file in the main/data folder that the dashboard displays.generate_map.R/generate_county_shapes()Obtains a shapefile containing county boundaries for the specified state/s.A geojson file containing county boundaries for the specified state/s in the/data/county_boundary folder.generate_map.R/generate_state_coords()Obtains geographic coordinates to orient the leaflet map to the specified state.A CSV file containing geographic coordinates for the centroid of the specified state in the/data/coords folder.generate_map.R/generate_cluster_coords()Obtains geographic coordinates to orient the leaflet map to the center of the detected clusters.A CSV file containing geographic coordinates for the centroid of the detected clusters in the/data/coords folder.

Dashboard features

Sample dataset and date dropdown menu

The dataset generated by the clean_combined_datasets() function is the underlying dataset for the CalCORVID dashboard. Our sample final dataset is named CAvax_combgiscol_fnl.csv and is located in the/data/folder. The main reactive functions filtering the map and table are the date dropdown and recurrence interval slider elements, which allow users to filter display results by date or recurrence interval, respectively. The available date range is determined in the combine_datasets() preprocessing step described above.

Leaflet map displaying visual clusters

The map is the main visual element of the results page. First, we render the map with our selected base map using renderLeaflet() and set the default view to the state of analysis. Leaflet uses the WGS 84/Pseudo-Mercator geographic coordinate system (EPSG:3857) projection by default for map display. It expects all input data to be specified in geographic coordinates using WGS 84 (EPSG:4326). Leaflet them automatically reprojects this data to EPSG:3857 for rendering. Therefore, we specify all our input files using WGS 84 (EPSG:4326) to ensure correct projection and alignment with map tiles. The geographic coordinates used to set the view are generated either by the generate_state_coords() or generate_cluster_coords() functions provided in the R folder depending on the parameter specified in the global.R file. After rendering the map, we use leafletProxy() to add features to the map. Using leafletProxy() improves the processing time of the dashboard because it does not need to render the base map every time the dashboard is reloaded. We use leaflet::addPolygons() to display the location geographies associated with each detected cluster, randomly assigning colors to more easily distinguish clusters. Clicking on these clusters results in a pop-up containing information about the average SVI for each cluster. For circular scans we provide an optional overlay of the circles using leaflet::addCircles(), which draws the circles using the geographic coordinates of the cluster center and radius in meters. This approach allows for the display of additional information without compromising dashboard loading times by only requiring a CSV file with the coordinates and radii instead of loading multiple shapefiles. We also use the leaflet::addLayersControl() function to create a toggleable layer for displaying county boundaries, which are generated using the generate_county_shapes() function provided in the R folder.

Data table displaying cluster information

Building the table requires converting the dataset to a reactive element using the reactive() function so reactive elements such as the date dropdown will work. Then, we convert the reactive dataset to a datatable widget using the datatable() function from the DT package, select the datatable display options, select variables of interest to display, and rename variables to be user-friendly. The display options for the datatable object can be found in the documentation for the DT package [29]. Finally, we render the table using the renderDT() function, which allows the table to be displayed in the user interface using the DTOutput() function.

In addition to displaying the cluster data, clicking a row in the table drops a pin on the corresponding cluster on the map to make it easier to identify. This reactive element uses the observeEvent() function to identify the row that is clicked and highlight the cluster it belongs to. Clicking on the row again clears the pin.

Results/discussion

Dashboard

The final CalCORVID dashboard consists of a homepage, a main results page, and a technical notes page. The homepage serves as the landing page for users when they first load the dashboard with the option for users to navigate onward to the main results page or technical notes page. The homepage is customizable and can be configured with different icons and personalized with the jurisdiction’s contact information at the bottom if shared with stakeholders. The main results page contains the date dropdown, map, and table elements described in the Implementation section. The date dropdown allows the user to select results from a predefined list of dates generated during the data preprocessing step using the combine_datasets() function. This functionality helps users visually track whether clusters become more or less statistically significant over time and whether they merit investigation. The main results page contains several other features that facilitate usability and provide further epidemiologic context (Fig. 2A-C). Clicking on a row in the cluster results table highlights the corresponding cluster in the map as shown (Fig. 2A). This interactivity ties the two visualizations together and aids in identification of map clusters, which can be difficult when multiple clusters are displayed. When clicking on a cluster on the map, the tooltip displays additional information about spatial covariates incorporated into the map (Fig. 2B). In this use case, the tooltip contains the average overall SVI and average SVI for each of the four SVI themes over all the geographic units included in the cluster. Finally, users have the option to select and view the county boundary layer generated by the generate_county_shapes() function which is used to geographically orient the user (Fig. 2C).Fig. 2A Main results page with the analysis name and state of analysis defined at the top of the page. A date drop down above the map allows users to switch between different dates for map and table results and a slider is provided to filter by recurrence interval. The map is set to the “state” zoom level. Clicking on a table row will highlight the corresponding cluster on the map to identify clusters more easily, with each cluster distinguished by a randomly selected color. B Checking the “Cluster” layer overlays the circular cluster scan on the cluster boundaries. Clicking on a cluster on the map provides the average Social Vulnerability Index (SVI) metrics for each of the four themes the overall SVI, and the number of geographic units in the cluster missing SVI data. C Checking the “County” layer on the map displays the county boundary layers to help spatially orient the user

Considerations for dashboard implementation

We do not provide specific guidance on cluster interpretation due to the range of model parameterizations, disease etiologies, and goals of each implementing jurisdiction. Instead, we recommend developers use the Technical Notes tab to clearly document their specific model configurations and to ensure users are familiar with the underlying methods prior to dashboard deployment and dissemination. Understanding the needs of end users can also guide decisions about which dashboard elements to include. For example, jurisdictions may choose to display variables beyond those provided by default in CalCORVID or incorporate additional features such as line lists of individual case characteristics. Existing documentation may support jurisdictions in tailoring their implementations [6, 18]. Additionally, findings from previous cluster validation studies can inform decisions about which data layers and outputs are most appropriate to include for their end users 11, 13, 30, 31]. Ultimately, CalCORVID serves as a starting point for enhancing ease-of-use of SaTScan output, offering a flexible framework that jurisdictions can adapt to meet their unique analytic goals and evolving user needs over time.

California Department of Public Health workflow and implementation

We describe our current implementation to provide an example of how other jurisdictions may want to integrate the dashboard into their own processes. The California Department of Public Health (CDPH) is interested in methods for identifying spatiotemporal clusters of disease to understand trends across the state. We run an automated prospective space-time analysis with a discrete Poisson probability model using nucleic acid amplification test (NAAT) results as our input data [11]. This allows us to monitor COVID-19 test positivity trends state-wide and regionally. Our data workflow is broken down into five distinct steps as shown in Fig. 3. We pull NAAT testing data from our data warehouse weekly and preprocess data in R to aggregate individuals to the census tract level and format the input files for SaTScan. Once the input files are created, we call the SaTScan software within R using the rsatscan package and run our analyses. After obtaining our results, we conduct a bipartite network analysis on the clusters to analyze whether they are part of the same extended cluster. Although we do not detail this process in the paper, this step is intended to distinguish unique spatial patterns over multiple overlapping clusters. Finally, the processed cluster results are output to the RShiny dashboard for local health departments in California to access. We automate our pipeline using an online server and schedule each job to run sequentially. The dashboard automatically updates with new data.

Local health departments and epidemiologists use this dashboard as a complementary tool for their routine surveillance activities to better understand COVID-19 trends in their region and at the state-wide level. The dashboard is shared with users in local health departments in a secure internal server, where health departments can explore the cluster results at their discretion and apply them according to their capacity, resources, and priorities. In practice, the dashboard facilitates the visualization and dissemination of cluster results that may warrant closer review and provides a starting point for preliminary investigations into potentially concerning clusters. The dashboard features, such as displaying historical cluster results and providing socioeconomic context, support the early assessment phase of whether a cluster merits further investigation. Even in the absence of clusters within their own jurisdiction, users can gain situational awareness of activity in surrounding areas or statewide, which can help inform local preparedness and coordination efforts.

Fig. 3. Flow diagram of CDPH workflow culminating in the RShiny dashboard display of results. We pull our NAAT electronic laboratory reports (ELRs) from our data warehouse, preprocess the testing data into input files for SaTScan, run the SaTScan software, apply a bipartite network analysis to the SaTScan cluster results, and output the final results into the CalCORVID-based dashboard

Comparison with other dashboards

Although other SaTScan-related dashboards exist, CalCORVID is the only one that explicitly addresses the challenges of visualizing spatiotemporal cluster results in an applied public health context. The most similar dashboard to CalCORVID is SpatialEpiApp, a RShiny-based dashboard that integrates SaTScan for cluster detection and disease risk estimation using integrated nested Laplace approximation (INLA) [20]. The input page requires two files: a spatial file containing the underlying base map and a CSV file for the SaTScan input case file. The user can then choose from spatial or spatiotemporal analyses to run. The results page contains several outputs, including an interactive map, model results, cluster results, and a downloadable report. One of the main differences between SpatialEpiApp and CalCORVID is that SpatialEpiApp requires users to run their data through the dashboard’s SaTScan mechanism to display the visualizations, while CalCORVID does not. Another SaTScan-related dashboard is EpiExploreR, which is a dashboard that provides tools to analyze spatiotemporal data, explore input data, and visualize results [19]. EpiExploreR combines multiple R packages and is distinguished by its ability to analyze epidemiological data using multiple methods, including SaTScan. Similar to SpatialEpiApp, EpiExploreR requires the user to run their data using the dashboard’s wrapper for SaTScan before results are displayed. Although both interfaces simplify running SaTScan for the user, the limited SaTScan parameter options may restrict users from more complex or nuanced analyses. For example, the SaTScan software allows users to select prospective or retrospective analyses, choose from eight different probability models, and scan for clusters with high and/or low rates while SpatialEpiApp only allows for prospective analyses, two probability models, and scans for high rates. SaTScan software also contains advanced options to refine cluster results, such as changing the spatial scan radius, temporal scan window, minimum number of cases to consider a cluster, and the ability to scan for clusters along a specified network. These advanced options are essential to correctly configure the spatiotemporal scan statistic for the disease of interest and jurisdiction [18]. CalCORVID simply requires the SaTScan output files and can be adapted for a variety of different models or use cases. The CalCORVID dashboard also streamlines the results page so epidemiologists can quickly browse through results as part of their surveillance routine.

Future directions

The first iteration of CalCORVID is a RShiny dashboard that is easily adaptable for different SaTScan models and provides basic functions that facilitate epidemiologic surveillance activities. Future extensions of the dashboard include support for a quantitative method to track clusters over time, inclusion of additional socioeconomic variables beyond the Social Vulnerability Index, support for display of linelists associated with each cluster, and extension to other spatiotemporal cluster detection methods. We plan to maintain CalCORVID as an open source repository and welcome contributions and suggestions to improve the functionality of the dashboard.

Conclusion

Spatiotemporal scan statistics are a useful complementary tool to traditional infectious disease surveillance activities. A dashboard is an effective method to visualize, monitor, and disseminate cluster results to a wide audience, and the open source code provided in the CalCORVID repository provides a starting point for epidemiologists interested in using the scan statistic as part of their workflow.

Availability and requirements

Project name: CalCORVID: A Dynamic RShiny Dashboard Approach to Visualize Spatiotemporal Clusters for Public Health Surveillance.

Project home page: https://github.com/cdphmodeling/CalCORVID

Operating system(s): Windows, MacOS, Linux

Programming language:R 4.0 or higher

Other requirements: See Supplementary Table 1

License: MIT

Any restrictions to use by non-academics: N/A

Supplementary Material

Supplementary Material 1.