A Petition to Create a Privacy-Respecting Pandemic Inference App

1. Abstract:

The major issue with tracking the COVID19 outbreak is that individuals who test positive have typically been transmitting to other individuals for 5+ days prior, and those new carriers produce a tree of new unknown infections. However, using a Bluetooth 4+ protocol, our phones could automatically keep a list of the people with which we came into close proximity. Then, by repeatedly comparing this local list with a server containing the encrypted names of people who have tested positive or are likely carriers of coronavirus, our chance of being exposed could be inferred.

This idea has been simulated in a paper by researchers at Oxford, who strongly think it would reduce the number of cases.

If you sign this petition, please share it with at least 25 of your colleagues. This all needs to happen in a matter of days if not hours.

2. Proposed Method of Operation:

We begin by assuming that a large majority of the population will download this application onto their smartphone (iOS or Android) and allow it to run in the background. Because this application is not intended to collect protected health information, we will not ask for your name or age. Rather, you will be given only a sliding scale of self-reported flu-like symptoms, as well as the option to report the results of an official COVID19 test. The app will create a random unique identifier for your smartphone. Over the course of up to two weeks, your phone will retain information regarding Bluetooth 4+ (RSSI) connections with other smartphones, as well as your location, which is immediately encrypted. From the past 14 days, your phone will contain a running list of a) smartphone identifiers that came within 2 meters of yours, b) the distance of closest approach, c) duration of this potential exposure, d) scrambled location, and e) timestamps. Once someone receives a positive diagnosis, their list relevant to the window of transmission (typically the past 5 days prior to appearance of symptoms, or date of exposure if known) will be pushed to a public server. Then, all smartphones will make individual inferences by looking for their unique identifier on the public server. Individuals who have a large chance of being exposed, especially those who also report flu symptoms, will also have their lists pushed to the public server, along with their estimated probability of being a carrier. This in effect automates work of epidemiologist interviews.

3. Intended Benefits:

This app would allow individuals to make informed decisions regarding whether they should self-quarantine, especially when they corroborate their exposure status with potentially weak flu symptoms. It would provide an inference-based stopgap for the current shortage of widely available testing in the United States.

Up-to-date recommendations will be provided to correspond to your chance of exposure, as well as your likelihood of being infected. This may also help to prioritize testing in the near future.

4. Current Major Issues:

• This system would require wide-scale adoption over the next few days to catch COVID19 while cases are still relatively sporadic.
• Apple has banned any applications that are not endorsed by the WHO on the AppStore.
• There is minor software protocol issue regarding the ability of iPhones to both search and broadcast for Bluetooth devices in the background. We suspect that we could find a work-around hack for this problem, especially with Apple’s cooperation.
• We would like to contact and argue for the opportunity to develop this app directly to state public health departments, the CDC, and the WHO, but they are inundated with work to address COVID19.

With enough signatures, I plan to attempt to contact the CDC or WHO directly, in addition to reporting this as an 'urgent bug' to Apple.

5. Privacy Concerns:

• Most individuals’ information will be stored locally on their phone. Once community transmission becomes widespread in any community (>10% prevalence), all local information will be deleted, and it will be reported that individuals have a 100% chance of prior exposure in that area.
• Locations will be first transformed by a high-distortion topographic map non-linear confidentiality algorithm and then hashed.
• Information will be automatically deleted from your device after the longest possible incubation period (14 days).
• You will be able to opt-out of having your information list pushed to the server or saved at all. In the latter case, this app will simply store your cumulative probability of being exposed to a COVID19 carrier, a prospective counter.
• This app will not report directly whether the individuals within 2 meters of you are likely COVID19 carriers, as this would be a violation of patient privacy. Users will be encouraged to self-quarantine and self-report this information.

6. Inference Method:

This portion of the app will be continually updated, and re-download from the public server. Bayesian statistical methods will inform how the exposure duration, distance of closest approach, and location (if you were exposed indirectly, such as the same room within a day) affect the probability of successful transmission. Inference will be performed bi-directionally along the chain of transmission.

7. About Me:

My name is John Matthew Nicklas. I am currently a senior in the undergraduate portion of Brown’s 8-year combined ScB-MD program. I study Applied Mathematics and Biology and am adept at coding, but I have contacted many of my friends who are Computer Science concentrators and they are willing to help. We intend to make the simplest possible product, with no fancy graphics or advertisements or anything of that sort. We do not intend to collaborate with any tech companies, except to the extent required to get the app to work smoothly. Rather, we hope to work at the command of a major healthcare provider. The most I dream of getting out of this is to write a short paper and help to slow the rate of transmission.