ALL RIGHT RESERVED © 2020 Amey Desai
-
Aid containment through proactive identification of critical transmission zones and modalities as well as streamlining of population flow
-
Enable seamless communication between 3 major stakeholders (Government, Healthcare Facilities and individuals.
The coronavirus has two important properties that have led to the current pandemic: it’s high infectivity and significant death rate{[1][2]}.
Measures for reducing the impact of infection broadly hinge on one aspect: Reducing the exponent. This can be achieved through aggressive contact tracing and isolation of positive cases{[4][5]}. While testing every person is impractical, informed decisions regarding the deployment of resources for social distancing and testing can substantially reduce spread and improve outcomes{[3]}. Visiting certain places such as hospitals or essentials’ stores is unavoidable. These are the primary avenues where infection can potentially spread. Social distancing needs to be followed diligently in such places. We use a two-pronged approach to facilitate adherence. Providing crowd density feedback for public places to allow for planning trips at less congested times or alternative routes, etc Providing an interface to book appointments at hospitals to avoid overcrowding We propose a digital platform to facilitate self-organization of population flow. The platform consists of a web portal and a mobile app. The implementation is based on a few key characteristics.
1. Crowd-sourced data The scale of the problem is beyond the ability of any single agency. Data collection requires that individuals opt-in to voluntarily share their data.
2. Privacy-first approach One of the most important factors affecting voluntary sign-up is privacy. Safety and privacy of the individual with respect to each stakeholder (other individuals, health authorities), is a basic requirement.
3. Proactive deployment of resources The general approach to contact tracing is retrospective. When implemented digitally through approaches such as Singapore’s TraceTogether, this effectively helps find and isolate potential cases. However, this approach only tackles direct human-to-human proximity based spread. Other means of spread, such as through contaminated surfaces in public places are not tracked by this method. Our platform also prioritizes GPS-based movement data to spot points of convergence in location trails of individuals, helping unearth such hard-to-find sources.
4. Closed Loop System Government officials need a sense of the situation on the ground to operate effectively. In particular, they need to be made aware of the inventory and patient load at hospitals.
The prevalence of smartphones and the internet makes mobile applications ideal candidates for this method of data collection. Our data-driven platform uses crowdsourced anonymized location data and Bluetooth proximity data from a user’s cellphone, along with demographic data and lifestyle data entered by the user. This helps us in determining other people that have come in close proximity in the past 2-3 weeks (through the proximity of their cellphones) as well as places in the neighbourhood where people generally tend to visit in large numbers (supermarkets, etc). In the event that a person tests positive, resources for testing, fumigation etc. can be directed towards these local hotspots of convergence in addition to notifying those who have come in close contact.
The demographic and lifestyle/health data collected will include prescribed drugs, preexisting conditions, other factors that can help medical professionals and epidemiologists gain more insight into clinical management and potentially aid development of vaccines.
Government officials require information about the situation on the ground at hospitals to effectively deploy resources as and when required. The web portal Infomed.live facilitates this need. Hospitals can update their inventory, their patient load and capacity, personnel availability, etc. regularly so that health officials can keep in touch with the situation on the ground.
Most countries that have made significant headway against the epidemic have temporarily suspended privacy concerns. Singapore’s TraceTogether app traces contact between people through the proximity of their phones. The app generates random ‘Contact Event Numbers (CEN)’ which are exchanged when two devices come in close proximity with each other. These CENs are linked to user entries in a central database that are accessible to the authorities{[8]}. While this grants anonymity to individuals from each other, it does not grant them anonymity from the authorities. The system also relies on the honesty of users{[8]}.
South Korean authorities have implemented one of the most successful screening systems in the form of drive-thru testing. While successful, it completely does away with privacy by using phone numbers to report results{[6]}. There have been cases of community ostracization and witch-hunts, which highlight the need for privacy in this matter{[6][7]}.
Our system largely relies on self-reporting to prevent privacy intrusions. We believe that making individuals aware of the impact that the epidemic can have on not just themselves, but their families and loved ones, sufficiently incentivizes voluntary participation. Comparing various methods of collecting data, and their levels of privacy (Table 1{[8]}), we have come to the conclusion that using an encrypted, multi proxy-server based approach works best to satisfy privacy requirements and constraints for scalability.
To make the system more resilient to spam, we implement a verification system that can be completely managed through either the mobile app or through the web portal. On self-reporting (after going through the symptom tracker), the user receives a ‘Patient number’ generated by the app. This number is used to schedule an appointment through the app. This ensures there isn’t overcrowding at hospitals/test centers. The results of the test can be securely notified to the user by hospitals through the app or the web portal. This system prevents false positives.
Preventing overcrowding at hospitals is quite important. As learnt from New York, hospitals themselves can become hotspots of infection. Given the resource constrained nature of the healthcare system in times of a pandemic, we need to ensure only as many patients visit as it is possible to manage.
Bluetooth based Proximity Detection Bluetooth based contact-tracing works by detecting proximity between cellphones through data broadcast using the Bluetooth communication protocol. Since the Bluetooth radio protocol is meant for short-range data transfer, it aligns perfectly with such usage.
When two phones are in close proximity, they exchange “Contact Event Numbers”, which are randomly generated numbers (tokens) unique to every such occurrence. Every phone stores all such tokens. In the event that a user reports positive, their tokens are uploaded to a public database. Every user’s phone periodically checks against these public tokens and notifies the user if any token in their list matches those of positive cases. In this case, the user is advised to test themselves. None of these tokens have any personally identifiable information. Nor are the health authorities aware of the owners of these tokens, hence testing and self-reporting are the responsibility of the user. This method is very accurate and scalable for person-to-person transmission detection{[8]}.
We use the decentralized crowd detection algorithm to detect vulnerable areas
GPS data also enables determination of hyperlocal points of convergence where resources can be directed for testing and containment efforts. A proximity alert is sent to users to avoid travelling to areas where the risk of infection is high.
Protocol
DCD divides the map into a grid with unit 10x10m. This is close to the limit of accuracy for mobile phone GPS systems.
Client Side (User’s phone) The mobile phone compares its location to the grid. For every grid unit it crosses into, it sets a stopwatch. If it remains in the grid for more than 300s, it notes the grid and sends the grid’s central coordinates to the server. Given that typical walking speeds are in the range of 1.5-3km/hr (0.42-0.83m/s), crossing a 10m path would generally take 12-25 s. We use the threshold of 180s to detect significant ‘contact’ with an area. ‘Contact’ refers to extended presence in an area, indicative of an increased probability of contact with other people/surfaces in the area. If the phone stays in the same place for a long time, DCD checks every two hours if the area is crowded, and only sends a packet every two hours rather than every 3 minutes.
Packets are routed through Tor, so the sender remains anonymous.
General location trails are stored securely on the phone. These are necessary for self-report triggered flagging. In addition, locations of visited clusters for the day will also be stored on the phone.
Server Side The server maintains a list of grid coordinates and a counter for each coordinate. Given the high number of such grids due to the resolution(105/sq.km), only those points that are reported are stored. Data received in the preceding 2 hours is used for analysis, and total data from the past 48 hours is stored on the server.
When a packet is received the grid point counter is incremented by 1.
Map Visualization
Client Side The mobile phone downloads the local density map from the server at regular intervals. Downloads are routed through Tor. Based on local cluster information, services like route optimization, travel recommendations, etc can function.
Server Side A map will be rendered on the server, graphically representing density. A grid will only display when it’s counter exceeds a threshold value of 20. The color of the grid will be used to represent relative density. Given social distancing norms, each person needs to maintain a distance of at least 1 m from others, thus occupying 3.14 sq.m on the ground. Thus about 31 people can safely occupy a 100 sq.m grid.
Flagging hotspots upon self-reporting When a user reports themself as positive, their visited clusters for the past week will be uploaded to the server. These will be flagged and notified to users according to the design policy
Preserving privacy when the number of users is low Initially, a lower number of users can enable linkage attacks. To prevent this, a grid point on the map is only rendered when the number of users crosses a threshold
Flowing traffic There may be scenarios where a moving crowd may be in present yet not register since individuals are moving faster than the threshold time(eg. Train stations). To overcome this potential blindspot, we liaise with bluetooth. We check the number of phones in the immediate vicinity (1 m). If the number exceeds a threshold, we check for time spent in the grid for a reduced threshold(1 minute as opposed to 3)
In addition to location and proximity data, we aim to include a short questionnaire in the process of building the user profile. This will comprise demographic{[9]} and health data (Age group, Gender, Pre-existing health conditions etc.) and lifestyle data (Shopping habits, Eating habits, etc). Users with high risk of and from the infection will be advised to take precautionary measures. Such data for positive cases will also help improve clinical care protocols by revealing co-morbidities, risk-factors and adverse drug interactions.
The app has a verified self-reporting feature, where the user can check their risk of having the infection through a symptom tracker and schedule an appointment if necessary. Scheduling an appoinment securely and privately arranges for test results to be communicated back to the user. It seamlessly integrates with hospital protocols since results can be updated through either the app or the web portal.
Infomed allows officials to stay updated about the status of hospitals. Hospitals can be viewed individually or grouped by geographical area or other criteria. The total inventory, patient load/capacity, personnel availability can be seen. Hospitals can update this information regularly with ease, as well as coordinate patient test scheduling and result communication through the infomed or equally through the mobile app.
-
Similarities between COVID-19 and Influenza. WHO-int. Mar 2020 [Online] Available: https://www.who.int/news-room/q-a-detail/q-a-similarities-and-differences-covid-19-and-influenza
-
Coronavirus and Acute Respiratory Diseases (COVID-19, SARS-Cov and MERS). March 2020 [Online]. Available:https://www.msdmanuals.com/professional/infectious-diseases/respiratory-viruses/coronaviruses-and-acute-respiratory-syndromes-covid-19,-mers,-and-sars
-
Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand; Ferguson, Neil M and Laydon, Daniel and Nedjati-Gilani, Gemma and Imai, Natsuko and Ainslie, Kylie and Baguelin, Marc and Bhatia, Sangeeta and Boonyasiri, Adhiratha and Cucunub, Zulma and Cuomo-Dannenberg, Gina and others; Imperial College, London. DOI: https://doi. org/10.25561/77482
-
Estimated effectiveness of symptom and risk screening to prevent the spread of COVID-19; Gostic, Katelyn and Gomez, Ana CR and Mummah, Riley O and Kucharski, Adam J and Lloyd-Smith, James O; Elife; eLife Sciences Publications, Ltd; 2020
-
Quantifying dynamics of SARS-CoV-2 transmission suggests that epidemic control and avoidance is feasible through instantaneous digital contact tracing. Available:https://github.com/BDI-pathogens/covid-19_instant_tracing/blob/master/Manuscript%20-%20Modelling%20instantaneous%20digital%20contact%20tracing.pdf
-
As Coronavirus Surveillance Escalates, Personal Privacy Plummets. Available:https://www.nytimes.com/2020/03/23/technology/coronavirus-surveillance-tracking-privacy.html
-
Coronavirus privacy: Are South Korea's alerts too revealing? Available: https://www.bbc.com/news/world-asia-51733145
-
Contact Tracing Mobile Apps for COVID-19: Privacy Considerations and Related Trade-offs; Hyunghoon Cho and Daphne Ippolito and Yun William Yu; 2003.11511; 2020
-
Evaluating and Testing Persons for Coronavirus Disease 2019 (COVID-19) Mar 2020 [Online]. Available: https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-criteria.html
ALL RIGHT RESERVED © 2020 Amey Desai