CoVis Risk Score computation

May 28

What goes into calculating the CoVis Risk Score?

The CoVis Risk Score that you see on the dial of the score dashboard tell you how much risk you have faced today. The score depends on the follow:

o The locations that you visit every day. Not all locations are equally risky and CoVis takes this into account. Crowded places are more risky that open spaces. This is also taken into account in the risk score calculation

o Your day-to-day behavior. If you move around a lot then your chances of coming across more people who can possibly be infected increases. The CoVis app takes into account when you move from location to locations and accounts for the risk accordingly

o Your health background and demographics. Scientific research shows that people with certain underlying conditions and older people are more at risk of getting infected. In addition frontline workers are more likely to meet people who are infected. The CoVis Risk score takes all of this into account.

o The latest scientific literature on COVID-19 and its risk. CoVis is based on science. We at CoVis keep track of the on-going development in the scientific literature and update the score calculations accordingly. How do we do it? We update the parameters in our servers and these are sent to your phone so that your phone can calculate your risk score with updated scientific knowledge.

o The latest COVID-19 prevalence data at the county/district level. Every day we update the CoVis databases with the current number of cases prevalent all over the world. For some countries we have data at the county or district level (the highest granularity available) and we are constantly adding countries to our databases. These numbers are sent to your phone every night so that you can get the updated data.

Your risk score is calculated on a scale of 1 – 1000. The higher the risk of your activities and the higher the prevalence of COVID-19 in your area, the higher your risk score will be. Your score will be very low if you stay at home all day. Keep reading to get a better insight into what sources we use to calculate your score and how we make predictions.

Forecasting of case rates at the county/district level

We use data sourced from Johns Hopkins University for the USA and Robert Koch Institute for Germany for COVID-19 prevalence. For the other countries we use the respective national sources. The case rate prediction is done with a time series analysis using ARIMA. This is the standard machine learning algorithm used for making any time-series predictions. We have checked our algorithm with train and test data and the accuracy is about 10% or better. We work with 7 days averages as is standard practice and we cannot predict a sudden spike in the case rates or irregular reporting. For validations of ARIMA in the context of COVID-19 please refer to:

o Andres Hernandez-Matamoros, Hamido Fujita, Toshitaka Hayashi and Hector Perez-Meana, “Forecasting of COVID19 per regions using ARIMA models and polynomial functions,” Applied Soft Computing, Volume 96, 2020, DOI: 10.1016/j.asoc.2020.106610.

o Yanchun Pan, Mingxia Zhang, Zhimin Chen, Ming Zhou and Zuoyao Zhang, “An ARIMA based model for forecasting the patient number of epidemic disease,” 2016 13th International Conference on Service Systems and Service Management (ICSSSM), 2016, pp. 1-4, DOI: 10.1109/ICSSSM.2016.7538560.

We cross-check the predictions of the ARIMA with an AI algorithm called LSTM. While the AI algorithms are sometimes better than ARIMA in predicting time series data, they are often much slower and do not provide a significant gain in the accuracy of the score. We continue testing with better forms of LSTM, like the GRU, to see if they can perform better. This is work in progress

Location based risk

Every time a user goes to a location that is deemed significant (the user has been there for more than 15 minutes) the location is registered and the risk score for that location is calculated on the basis of the location type and time of the day. The inherent risk of the location is based on recommendations from the Texas Medical Association and the prevalence of COVID-19 in that county/district (computed from the forecast described above).

We have made an estimation of how crowded a place can be at a certain time of the day based on mobility data. We also use the data on the prevalence of COVID-19 in that region. For example a movie theatre in a low prevalence region will have a lower risk score than a movie theatre in a high prevalence region. The risk score of each location visited adds up to the total score for the day and is displayed on the main dial in the score dashboard.

Odds ratios from the medical literature

Several factors like age, gender, profession, underlying health conditions etc. affect the risk a person faces from COVID-19. We use the odds ratios associated with health conditions, age, gender etc. that we have gathered from peer-reviewed literature published in journals like the Science family, the Nature family, BMJ, JAMA, Lancet, NEJM, CDC MMWR etc. We also use recommendations from CDC, ECDC and RKI to assess what factors increase the risk to COVID-19. A shortened list of peer-reviewed journals and reports that we source the data from are:

Gender:

o Zhaohai Zheng et al., “Risk factors of critical & mortal COVID-19 cases: A systematic literature review and meta-analysis” Journal of Infection, Volume 81, Issue 2, e16 - e2. DOI: 10.1016/j.jinf.2020.04.021

Profession:

o Long H Nguyen, Christine M. Albert, et al. “Risk of COVID-19 among front-line health-care workers and the general community: a prospective cohort study,” The Lancet Public Health, Volume 5, Issue 9, e475 - e483. DOI: 10.1016/S2468-2667(20)30164-X

o https://www.weforum.org/agenda/2020/04/occupations-highest-covid19-risk/

o https://www.onetcenter.org/overview.html

Smoking:

o N. S. Hopkinson, N. Rossi, J. El-Sayed_ Moustafa, et al, “Current smoking and COVID-19 risk: results from a population symptom app in over 2.4 million people,” Thorax Published Online First: 05 January 2021. DOI: 10.1136/thoraxjnl-2020-216422

o Killerby ME, Link-Gelles R, Haight SC, et al. “Characteristics Associated with Hospitalization Among Patients with COVID-19 — Metropolitan Atlanta, Georgia, March–April 2020. MMWR Morb Mortal Wkly Rep 2020;69:790–794. DOI: 10.15585/mmwr.mm6925e1

o https://www.who.int/news-room/commentaries/detail/smoking-and-covid-19

Pregnancy:

o https://www.cdc.gov/coronavirus/2019-ncov/need-extra-precautions/pregnancy-breastfeeding.html

Comorbidities (metanalyses):

o N. Rosenthal, Z. Cao, J. Gundrum, J. Sianis, S. Safo. „Risk Factors Associated With In-Hospital Mortality in a US National Sample of Patients With COVID-19.” JAMA Netw Open. 2020;3(12):e2029058. doi:10.1001/jamanetworkopen.2020.29058

o https://www.cdc.gov/coronavirus/2019-ncov/hcp/clinical-care/underlyingconditions.html

o https://www.cdc.gov/coronavirus/2019-ncov/science/science-briefs/underlying-evidence-table.html

Cancer:

o https://ascopost.com/news/july-2020/study-finds-higher-risk-for-covid-19-infection-among-minorities-and-patients-with-cancer/

o D. A. Drew et al, “Rapid implementation of mobile technology for real-time epidemiology of COVID-19,” Science 19 Jun 2020: Vol. 368, Issue 6497, pp. 1362-1367 DOI: 10.1126/science.abc0473

o M. Dai, et al, “Patients with Cancer Appear More Vulnerable to SARS-CoV-2: A Multicenter Study during the COVID-19 Outbreak,” Cancer Discov June 1 2020 (10) (6) 783-791; DOI: 10.1158/2159-8290.CD-20-0422

Cardiovascular Diseases:

o J. Yang et al, “Prevalence of comorbidities and its effects in patients infected with SARS-CoV-2: a systematic review and meta-analysis,” International Journal of Infectious Diseases, vol. 94, May 2020, Pages 91-95. DOI: 10.1016/j.ijid.2020.03.017