As part of the CoVis project (Project CoVis) at DESY, Hamburg, we developed an app for real-time COVID-19 risk assessment for individuals. We made the app available for free in the USA, Germany and Austria and will expand the coverage to other countries as and when we have time and resources. While the app does not use SafeGraph data directly, we used the data to build the theoretical distributions of how crowded a place of a certain type can get. We use machine learning to forecast the prevalence and Bayesian inference to compute the risk-score. Questions/comments/suggestions welcome. We will put out a preprint soon. The app can be found here:
Android: https://play.google.com/store/apps/details?id=com.covisinc.covis
iOS: CoVis on the App Store
This topic was automatically generated from Slack. You can find the original thread here.
This is extremely cool! Any chance you are looking for new people to join your team at the CoVis project?
For this: “We use machine learning to forecast the prevalence”, would you feel comfortable discussing a little bit more in depth how this was done? Might benefit others who are working on similar projects.
I was also curious- what is your goal in creating the app? Do you have target number of users or sessions?
Finally, I have some feedback from an app user perspective:
-Might be good to get some more detail on the page where it says “Calculating” risk score (i.e. “Calculating typically takes 24 hours”, or "Calculating will show a value after you’ve visited 2+ locations)
-Last, it might be interesting to layer on information about how behavior affects risk (in addition to location). I’m not sure if you’ve seen this, but it’s a super interesting tool that does this: https://www.microcovid.org
Hi ! Thanks for the feedback. We are considering ways of adding layers of information and can certainly incorporate a lot of details. The app is barebones now and we wanted to put it out instead of waiting till its perfect.
For the “Calculating…” part, yes some more visibility should be put in. maybe a banner on top might help.
I will be very happy to discuss all that goes into the algorithms that CoVis uses, the machine learning and the risk score computations.
We are always looking for help with CoVis especially data ninjas, creative people and coders (both for the app and the backend). We will welcome anyone with open arms.
I know about the website. The problem is a website is a book compared to an app being a few sheets of paper. While the goal of the app is to guide user behaviour toward what is safe, we really need to put our heads down and understand the most concise way of doing this effectively. All ideas are welcome.
“I was also curious- what is your goal in creating the app? Do you have target number of users or sessions?” — we are aiming for users that will use this as a companion. So more sessions. Of course more users are always welcome but this is not really a once in a week use app. The other good thing is that all the methods we developed can be expanded to cover other communicable diseases, allergies etc.
Thomas Young (Econometric Studios/Interesting take, . I noticed the app doesn’t initially pick up my location, instead having as the default of Massachusetts. How come? I think it’d be interesting to have some comparative stats, like “You are XX% riskier than your neighbor …”.
Hi ! During onboarding one has to specify ones residence. Since the residence county can be different from where one is during onboarding, we do not automatically identify the location there. However, its a good idea to suggest the location there. I will pass this on to the team. Once you get to the score dashboard, the app takes about 15 mins to get a “significant location”. This is because we want to ignore in and out visits and only count the visits that are prolonged. Of course, there is a lot of debate around this and we are exploring many other possibilities (like user input etc.) to enhance this algorithm.
We also thought about the “You are XX% riskier than your neighbor …“. but for that we have to know the average in the areas for which we have to collect data in an area. Firstly, that would require enough users (which is difficult initially) and we do not want to collect data (we are very strict on data privacy). One can always cite a baseline. However, since this is real-time risk assessment, the baseline depends on a long list of factors and we could not find an easy way of making this clear to all users, even those who are not mathematically inclined or have very little knowledge of public health. In fact, that is one of the biggest hurdles we faces, making the app accessible and welcoming for all users.
Hey - congratulations on this release! This is fantastic, and thanks for sharing to the community.
I’m curious about one of the questions that you’re using to assess risk. I noticed the app asks Do you work in any of these fields? and have various choices (education, delivery, healthcare, etc.) I noticed that the app doesn’t ask the same question about significant other, family members, children, or other people that might be living with you.
Does the literature you’re basing this off of show a higher risk for those that live with someone who works in a public-facing industry? For example, my spouse works in a hospital, I would have thought I might be at a higher risk since we live together. Curious to hear your thoughts!
I am not sure if there is a study on partner’s occupation. We also know that having kids go to school also elevates risk. However, we cannot work without numbers since the entire calculation is based in odds ratios and probabilities and we try not to put in our own back-of-the-envelope estimates.
The other consideration is the actual number of questions and how it ties in with user experience. Us scientists are used to answering a bunch of questions. However, for most users of the app, 10 or 11 questions are already too many. If we asked about children we would also have to be aware of whether the schools are open or not for which we would need to know local policies etc.
As you see, more data is always useful, but at some point, if you think about 3142 counties in the US 401 districts in Germany etc. building the data stream becomes complicated if we consider local regulations. We would love to do it if we have resources.
As for user inputs, we are exploring other means of gathering data from users outside the onboarding like pop-up or something so that they answer them once they are comfortable with the app.
Ryan Kruse : Super cool! I have a follow-up question to your response to Niki. Might you consider asking the standard 10 questions, then (if you have other pieces of information you’d like to collect) giving the option to answer an additional X questions? Or for users who’ve been using the app consistently for a day or two, give them the option to answer an additional question every now and then? Could balance the effort required from the user while still allowing you to get more specific.
Additionally, I have one minor piece of feedback. I have the font size larger than normal on my phone, and the state/county selection screen was somewhat cut off and would not let me scroll (screenshot attached). I was able to select my county, but I had to click it just right. If necessary, I would have temporarily shrunk the font size in my settings, but I could see that being prohibitive to some people’s use of the app.
Hi ! Thanks for taking it for a spin. That screen certainly needs to go into a ScrollView. We tested it out with giant fonts but maybe we also had giant screens. So We never caught that one. Thanks for letting us know. If you use large fonts, some elements will look a bit strange. We tested with various font sizes as much as possible but we need to iron out some issues.
As for the questions, yes, we would like to layer in more questions which might be asked, lets say, a couple of days after the user has started using the app. It makes the app flow a bit more complex to design out and we are working with the UX/UI team members to see what can be done.
Hi everyone! CoVis has made quite a bit of progress since I introduced it here. We have been now nominated for the M&K award in Germany and would like to ask you for your support to win the competition. CoVis is an app for assessing individual risk from COVID-19 and is available in the USA, Germany, Austria and Belgium (coming this week). You can cast your vote for us here:https://www.pro-4-pro.com/de/specials/mka.html **** Please Vote**** Please mark: “Eine neue App zur Corona-Risikobewertung – Deutsches Elektronen-Synchrotron DESY”The description for CoVis can be found here
We have put in extensive documentation in the app to explain how the score is computed and what it means.
The location type algorithm has been further tuned for higher accuracy. (We finally conquered Google lingo)
We have incorporated recent development in vaccination studies. We are planning to incorporate reduction in vaccine efficacy once the numbers around that gets a bit clearer
We have updated the odds ratios for underlying conditions with more recent and much larger studies (Like the one by CDC on more that half a million participants)
We have expanded coverage to Belgium and are working on the other countries we mentioned before.
