Content uploaded by Jay Luthar
Author content
All content in this area was uploaded by Jay Luthar on Nov 14, 2020
Content may be subject to copyright.
COVID-driven Risk Profile
Jay Luthar?
Cambridge Health Alliance
jluthar@challiance.org
Vivek Sharma?
MIT Media Lab/Harvard Medical School
vvsharma@mit.edu
Siddhant Gokhale
Harvard Kennedy School
siddhant_gokhale@hks.harvard.edu
Ramesh Raskar
MIT Media Lab/Pathcheck Foundation
raskar@mit.edu
1 Introduction
Novel strategies for combating COVID-19 are sorely needed. As COVID-19 completely disrupted
day-to-day economic and social activity, policies and individuals are relaxing basic public health
measures such as social distancing, and cases as well as hospitalizations are rising across the globe
as a result. Testing is a foundational strategy for combating pandemics, but it has limitations. Tests
viewed as a static binary answer are most useful in the clinic and hospital setting to diagnose a
symptomatic patient. But even in this setting a positive or negative test is truly only giving a statistical
chance of being positive or negative in respect to COVID-19. As a public health measure, this
uncertainty is not factored into the testing paradigm, and ignores multiple other variables that are
relevant to determining COVID-19 risk status. The technical sensitivity of tests varies between test
manufacturers, and time course in the potential illness. Some studies have suggested the false negative
rate for the gold standard PCR tests are up to 100% on the day of exposure, and 38% on day 5 of
symptoms (Kucirka et al., 2020). The paradigm of a positive or negative test result having true utility
in asymptomatic carriers is challenged by this data, and leaves a huge hole in the prevention of the
spread of this disease as asymptomatic carriers carry and shed virus at rates very similar to the sick
individual (Lee et al., 2020). Estimates that at least one-third of asymptomatic carriers would need to
be identified, isolated, and contact traced to keep infection rates below 1% (Moghadas et al., 2020).
Strict isolation and lockdowns work to prevent the spread of the disease, but are economically and
socially enviable. Restoration of these social and economic activities in a safe and responsible way is
crucial for the economy to recover from a deep recession, prevent physical and mental suffering due
to isolation.
Solution:
To solve these challenges, we propose a statistically validated risk model on a novel
digital ecosystem, Nexus, that utilizes machine learning, contact tracing, public health databases, and
individual symptoms and behaviors to calculate a risk level. This risk level can prompt potentially
asymptomatic carriers to get testing to confirm or rectify possible risk, thus greatly reducing the
spread of COVID-19, while allowing normal economic, social, and educational activity to continue
where safe.
An interoperable system that provides risk profiles would enable retailers, hospitality services, and
small businesses to allow those with sufficiently less risk to use their services and purchase their
products, resuming commerce. It would also enable individuals with low risk to congregate more
safety and interact in open public places, in cinema halls, and for sports events, knowing that they are
likely not perpetuating the disease or being infected by it, especially if combined with point of entry
rapid testing. Finally, aggregation of risk profiles of individuals by geographical area would allow us
to make heat maps of potentially risky areas with high chances of infection.
?Jay Luthar and Vivek Sharma contributed equally to this work and listed in alphabetical order.
2 Technology Review and Progress-To-Date
Currently, contact tracing platforms (such as the one developed by the Pathcheck Foundation
1
and
EN Express (Raskar et al., 2020a)) can help alert those individuals who may have been in close
proximity with an infected person, while preserving the individual person’s identity, and ensuring
their GPS and Bluetooth (Raskar & Sathya, 2020) data on proximity remains private. However, we
still do not collect sufficient information, about how frequently each person tests for COVID, about
the frequency of the testing, and the demographics to be able to create a risk profile of individuals.
There are few players in this space at the moment and most are at a very nascent stage of development
(for example, Virtual In/Out
2
). While some solutions are emerging, they either lack interoperability
or sufficient privacy or integration of testing information.
3 Potential Solution(s)
A platform (such as a smartphone app) that:
•
Combines data on testing across multiple test centers (inducing location, frequency, type
of test taken by an individual as well as very basic demographics, such as age) with data
about proximity among individuals (from GPS and Bluetooth), symptoms, and geographical
exposure.
• Uses machine learning algorithms to construct a risk index
•
Collaborates with businesses that require users of the platform to scan a QR code which
reads their risk profile, and based on criteria, allows a person to enter a restaurant or public
gathering.
• Ensures a high level of privacy for individual data
4 Potential Stakeholders
This requires collaboration from consumers, small businesses, and testing centers. Consumers would
be interested in this platform because it would allow them access to hospitality services (flights,
tourism, restaurants) with the reassurance that (1) they likely are not spreading the disease, and (2)
they are likely not being infected by the disease from others in the venue of the gathering. Businesses
would be interested because they can attract more consumers, especially those conscious consumers
that have been avoiding hospitality services due to health risks. Test centers would be interested in
collaborating because their participation will increase testing (since consumers would now want to
test frequently to be able to maintain their risk profiles). Governments (local and state) would also be
interested as such a solution would help restart the economy
5 Potential Use Cases
• Universities/Schools looking to get back to fully in-person classes
• Offices/Commercial buildings looking to reopen/business conferences/world events
• Entertainment (movie theaters, stadiums
• Hospitality (restaurants, airlines, hotels, cruises, event managers, beauty parlors, etc.)
• Retail (supermarkets)
• Transport (Public Transport+Uber/Lyft)
• Airbnb
• Dating apps
• Hospitals
1https://www.pathcheck.org/
2https://www.virtualinout.com/blog/2020/05/05/QR_Code_Contact_Tracing
2
6 Our Model
Our model utilizes 4 categories to assign a risk level to individuals, businesses, and geographic areas.
6.1 Symptoms and Testing
The symptoms experienced by someone who is infected with and contagious can range from nothing
(asymptomatic) to generalized symptoms that may be indistinguishable from any other respiratory
viral infection such as cough, fever, myalgias, fatigue. In its most severe form, hospitalized patients
can have life threatening pulmonary, neurological, vascular, cardiac, or renal damage. The public
health challenge is identifying asymptomatic and presymptomatic individuals. One unique symptom
that seems to have high specificity for COVID-19 (95% with 90-98% CI), but poor sensitivity is
abrupt loss of taste and smell in someone who otherwise has no ear nose and throat issues at baseline.
That is to say it can be highly suggestive of Covid infection, but lack of this symptom cannot rule out
the disease (Bénézit et al., 2020).
Practically speaking, any patient with any of the following symptoms would be deemed as maximum
risk, until two negative tests at time validated time points were obtained:
• Fever
• Chills
• Cough
• Dyspnea
• Anosmia
• Aguesia
• Myalgias
• Sore Throat
• Runny Nose
• Acute GI symptoms
• New Unusual Headache
• Chest pain
A negative test would provide a likelihood of negative status, dependent on type of test, time since
symptom onset, and number of tests taken, and the nature and resolution of the symptoms themselves.
Our app would load testing results from a number of local and national testing facilities. Using
blockchain to create an immutable ledger of status, an individual’s QR code, live generated by the
app, will keep an up to date status of their testing, and risk profile.
6.2 Contact tracing and Public Health Data
Individuals would be assigned a baseline rate of being infected based on updated public health
databases.
At level 1 of the algorithm, every individual that a user of Nexus comes in contact with is assigned a
risk score solely based on public health data. Based on proximity (Raskar & Sathya, 2020; Raskar
et al., 2020a; Shankar et al., 2020), time, and mask usage, air temperature, wind, and number of
contacts, there is a statistical likelihood of transmission to the nexus user from this pool of contacts
that will be calculated and factored into the risk profile.
Several scientists have created models of aerosol transmission based on what is known of general
aerosol science applied to COVID
3
as well as based on case studies such as the Skagit Valley Chorale
superspreading event (Miller et al., 2020).
At Level 2, each individual has their own risk profile that is a culmination of symptoms, testing,
contact tracing (Shankar et al., 2020; Raskar et al., 2020b), and public health data (Kogan et al., 2020).
3https://docs.google.com/spreadsheets/d/16K1OQkLD4BjgBdO8ePj6ytf-RpPMlJ6aXFg3PrIQBbQ/
3
An individual may have a different risk score if they are in contact with one high risk individual, vs
several very low risk individuals. At this point, the machine learning algorithm will constantly be
updating and learning more accurate risk models, and testing the model against actual testing and
symptom data.
Anonymous notifications of contacts of an infected individual would give instructions to quarantine
and seek testing via Pathcheck.
References
Bénézit, François, Le Turnier, Paul, Declerck, Charles, Paillé, Cécile, Revest, Matthieu, Dubée,
Vincent, Tattevin, Pierre, Arvieux, Cédric, Baldeyrou, Marion, Chapplain, Jean-Marc, et al. 2020.
Utility of hyposmia and hypogeusia for the diagnosis of COVID-19. The Lancet Infectious
Diseases.
Kogan, Nicole E, Clemente, Leonardo, Liautaud, Parker, Kaashoek, Justin, Link, Nicholas B, Nguyen,
Andre T, Lu, Fred S, Huybers, Peter, Resch, Bernd, Havas, Clemens, et al. 2020. An early warning
approach to monitor COVID-19 activity with multiple digital traces in near real-time. arXiv
preprint arXiv:2007.00756.
Kucirka, Lauren M, Lauer, Stephen A, Laeyendecker, Oliver, Boon, Denali, & Lessler, Justin.
2020. Variation in false-negative rate of reverse transcriptase polymerase chain reaction–based
SARS-CoV-2 tests by time since exposure. Annals of Internal Medicine.
Lee, Seungjae, Kim, Tark, Lee, Eunjung, Lee, Cheolgu, Kim, Hojung, Rhee, Heejeong, Park, Se Yoon,
Son, Hyo-Ju, Yu, Shinae, Park, Jung Wan, et al. 2020. Clinical course and molecular viral shedding
among asymptomatic and symptomatic patients with SARS-CoV-2 infection in a community
treatment center in the Republic of Korea. JAMA internal medicine.
Miller, Shelly L, Nazaroff, William W, Jimenez, Jose L, Boerstra, Atze, Buonanno, Giorgio, Dancer,
Stephanie J, Kurnitski, Jarek, Marr, Linsey C, Morawska, Lidia, & Noakes, Catherine. 2020.
Transmission of SARS-CoV-2 by inhalation of respiratory aerosol in the Skagit Valley Chorale
superspreading event. Indoor air.
Moghadas, Seyed M, Fitzpatrick, Meagan C, Sah, Pratha, Pandey, Abhishek, Shoukat, Affan, Singer,
Burton H, & Galvani, Alison P. 2020. The implications of silent transmission for the control of
COVID-19 outbreaks. Proceedings of the National Academy of Sciences,
117
(30), 17513–17515.
Raskar, Ramesh, & Sathya, Sai Sri. 2020. Bluetooth based Proximity, Multi-hop Analysis and
Bi-directional Trust: Epidemics and More. arXiv preprint arXiv:2009.06468.
Raskar, Ramesh, Singh, Abhishek, Zimmerman, Sam, & Kanaparti, S. 2020a. Adding Location
and Global context to the Google/Apple Exposure Notification Bluetooth API. arXiv preprint
arXiv:2007.02317.
Raskar, Ramesh, Dhillon, Ranu, Kapa, Suraj, Pahwa, Deepti, Falgas, Renaud, Sinha, Lagnojita,
Prasad, Aarathi, Singh, Abhishek, Nuzzo, Andrea, Iyer, Rohan, et al. 2020b. Comparing manual
contact tracing and digital contact advice. arXiv preprint arXiv:2008.07325.
Shankar, Sheshank, Chopra, Ayush, Kanaparti, Rishank, Kang, Myungsun, Singh, Abhishek, &
Raskar, Ramesh. 2020. Proximity Sensing for Contact Tracing. arXiv preprint arXiv:2009.04991.
4
