Changes in social contact time following digital SARS-CoV-2 post-exposure notification: A retrospective case-control study at a US university

doi:10.21203/rs.3.rs-3760051/v1

Download PDF

Article

Changes in social contact time following digital SARS-CoV-2 post-exposure notification: A retrospective case-control study at a US university

https://doi.org/10.21203/rs.3.rs-3760051/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

In the first year of the COVID-19 pandemic, digital contact tracing smartphone applications were developed for the purpose of reducing transmission by encouraging individuals to voluntarily quarantine after exposure. Privacy protocols used in many of these apps prevented longitudinal linkage of users’ contact network data, making it impossible to observe whether users reduced social contact following digital exposure notification. We used anonymous data from one particular app deployment, which used alternative privacy measures and retained longitudinal interaction data, to evaluate changes in the users’ time in proximity to other users after digital exposure notification.

We analyzed a dataset of the anonymous app users in a US university community January − March 2021, before COVID-19 vaccines were widely available. Users provided consent for their anonymous data to be used for research. We conducted pre-post analysis of contact time with other users before and after using the app to self-report being a positive COVID-19 case, and before and after exposure notification by the app. We further conducted a retrospective case-control study by matching users who did and who did not receive exposure notification based on having similar contact time with other users prior to notification.

In pre-post analysis, we found that users who reported being a positive case to the app reduced their total contact time with others by 29% (p = 0.002) in the 7-day period immediately after time of the report vs. the 7-day period before the time of the report. However, we observed no statistically significant change in contact time before vs. after exposure notification. We also observed no statistically significant difference in contact time compared to matched controls who received no notification.

Our findings suggest that digital contact tracing app users affiliated with this US university did not reduce contact with others after receiving exposure notification. Our study provides one of the first rigorous evaluations of the effectiveness of digital exposure notifications during a time when communities were reliant on non-pharmaceutical interventions to reduce disease spread.

Health sciences/Health care/Public health

Health sciences/Health care/Health policy

Health sciences/Health care/Disease prevention

Contact tracing is a public health tool used to identify likely chains of transmission in order to slow and stop infectious disease outbreaks^1,2. Conventional contact tracing methods require trained personnel and other resources, which can be difficult to scale during times of rapid epidemic spread^3–5. Digital contact tracing tools are designed to augment conventional contact tracing by automatically detecting epidemiologically relevant contacts and distributing exposure notifications to users who are likely to have been exposed to disease. The widespread adoption of smartphones in many settings worldwide has made it possible to develop smartphone applications (apps) which serve as digital contact tracing tools^6,7. Digital contact tracing apps are designed to use smartphone proximity detection features to identify likely close contacts of individuals. Contacts then automatically receive digital exposure notifications through their own smartphones about the need to self-isolate to reduce the risk of further transmission.

Numerous digital contact tracing smartphone applications were developed during the first year of the COVID-19 pandemic^6–10. During this time, vaccines and antiviral treatments were not yet widely available, leaving health authorities with only non-pharmaceutical intervention options such as mask mandates, social distancing policies, and test-and-trace programs to mitigate the health burden of COVID-19. The goal of digital contact tracing apps in the COVID-19 context was to expand the reach of conventional contact tracing and to relieve the burden of general-population lockdowns while still identifying and isolating exposed and infected individuals^11–13. For the United States and many other global settings, this was the first time attempting to implement digital contact tracing at scale.

Like any public health effort, digital contact tracing requires evaluation of effectiveness, challenges faced, and opportunities to improve design and future implementation^9,14,15. This is especially important for digital contact tracing because it relies on public participation, motivated at least in part by belief that the program has a meaningful public health benefit^14,15. A particularly important impact metric for digital contact tracing is whether exposure notification resulted in behavior changes among notified contacts that could mitigate disease spread ^16–19. Mathematical models for predicting the potential impact of digital tracing on COVID-19 have assumed relatively high rates of behavior change among contacts, e.g., that 80–90% of individuals received exposure notifications would isolate for 14 days. This led to predictions that even low rates of digital tracing adoption could substantially reduce transmission – which led to optimistic initial estimates of the apps’ efficacy even in populations with low rates of adoption²⁰.

However, few studies have empirically evaluated behavior change after receiving a digital notification of prior exposure to COVID-19^8,21–25. The most widely-deployed digital contact tracing app in the US, the Google-Apple Exposure Notification System (GAEN), uses a decentralized approach to storing user data and does not retain longitudinally linked records of users²⁶. Thus, while GAEN may have records of issuing an exposure notification, it retains no longitudinal information that could be used to assess whether the notification resulted in protective behaviors which reduced time in close proximity to others. Other empirical studies have relied on follow-up surveys of users²⁴, which is subject to respondent bias, or on indirect population-level proxy metrics such as changes in foot traffic in an outbreak area, which is not restricted to exposed contacts and may not reflect the specific effect of digital exposure notification²⁷.

The present study used longitudinal user proximity data to assess whether digital exposure notification led recipients to reduce physical interactions on a US college campus. We leveraged an anonymous digital contact tracing app which does not use the same decentralized privacy design as GAEN, which made it possible to reconstruct the longitudinal network of contacts between de-identified users. We analyzed the interactions between app users before and after exposure notification, as compared to matched controls who had similar pre-notification social network activity, but were not recipients of an exposure notification. Our analysis provides one of the first rigorous evaluations of the effectiveness of digital exposure notification during the pre-vaccination period of the COVID-19 pandemic.

Smartphone App

Our study used data from NOVID, an app for Android and iOS developed during the COVID-19 pandemic^28,29. The app supported multiple different strategies for mitigating disease spread, one of which was exposure notification. A smartphone with the app installed broadcasts Bluetooth signals to identify potential contacts based on other nearby users. The app estimates the duration of interactions based on the time of the first and last exchanged signals, and the distance between contacts based on the strength of the Bluetooth interactions. The app further classifies contacts as either strong or weak: strong contacts occur between smartphones that are ≤ 10.1 meters apart for ≥ 15 minutes, while weak contacts consist of all other interactions.^19,30 To address limitations in the accuracy of Bluetooth-based distance measurements between smartphones^23,31, the app augments Bluetooth-based distance estimates with inaudible, ultrasonic microphone “squeaks.”²⁹ At the time of the analyzed deployment, ultrasonic “squeak” distance measurements were only operational between Android phones. The app maintains user privacy by assigning each user a random identifier at the time of app installation. No location data or personal identifiers are collected, but daily strong and weak contacts between users are collected and used to construct a contact network of cumulative contacts over the preceding two weeks.

The app allows users to report to the app when they have tested positive for COVID-19. When this happens, the app delivers notifications of prior exposure to other users who were connected up to 2 contact network “hops” away in the preceding two weeks. The exposure notifications are delivered via a smartphone push notification, i.e., a pop-up alert that typically appears on a mobile phone’s lock screen and notifications tab such that it is visible even if the application is not open.

Deployment in a university community

The present study focuses on the app’s users at the Georgia Institute of Technology (Georgia Tech), a large public research university located in Atlanta, Georgia, United States. Georgia Tech resumed in-person lectures in the Fall semester of 2020, making it one of the first major US universities to resume large gatherings after the onset of COVID-19³². As part of the university’s COVID-19 mitigation measures, students and staff were encouraged to use the app’s notification features in order to facilitate timely testing and self-isolation after a potential exposure^32,33. Although the app also introduced another different disease mitigation strategy (a personalized “COVID-19 radar” showing how many network “hops” the user was from positive cases in their contact network²⁹), that feature was not emphasized in the university’s communications, and so the present study focuses on testing digital exposure notification. To our knowledge, there was no other contact tracing app in wide use among this population; the GAEN contact tracing app was not permitted to operate in the state of Georgia at this time. Students began returning to campus in early January of 2021 before classroom instruction began on January 14. Campus health services recommended that returning students and staff test within 3–5 days of arriving on campus and to self-isolate for 10 days before commencing with in-person classes and other group activities. Students and staff who tested positive were asked to self-isolate for 10 days³⁴. Free-of-cost testing was available during this time at Georgia Tech’s campus health services, which had capacity to test all campus visitors biweekly³⁵.

Although the app became available for download in mid-2020, the present analysis focuses on data collected beginning January 1, 2021, when the app became fully functional on both Android and iOS smartphones and campus health services began verification of positive cases reported to the app. The data collection period ended on March 15, 2021 to focus the present study on a time when COVID-19 vaccines where not yet widely available and health authorities strongly recommended participation in contact tracing, in addition to other non-pharmaceutical interventions such as testing, mask-wearing, and social distancing.

Analysis of app data

Definition of app users at Georgia Tech

Because the app does not collect location information, users in the Georgia Tech community were identified based on the set of users who input Georgia Tech’s community code upon installation as well as their nearest neighbors in the contact network²⁹. This code was disseminated as part of an informational campaign on campus to encourage the use of the app. App users were considered to be active from the time the app was installed on a smartphone until the last time that device transmitted an automated daily message indicating the app remained installed. For the present analysis we assumed each user ID represented a unique individual, although we note that if a user were to uninstall and re-install the app on the same device, they would be issued a new user ID and recorded as two separate users.

Inclusion criteria for analysis

We identified three groups of active app users for analysis: “reported cases,” all users who reported a positive test result to the app; “notified users,” all users who received a digital exposure notification from because they were near to at least one reported case in the contact network; and the “control group,” all remaining users who never reported a case or received an exposure notification. For each of the reporting users and notified users, we excluded users who had fewer than three days active in the contact network in the seven days prior to the report or notification in order to focus on consistent app users with measurable baseline social interactions. We also excluded users who deleted the app within 7 days or never became active again after 7 days following the time of the report or notification.

Calculating contact time

We used Bluetooth signal data to calculate interaction duration as the time from the first signal exchange to the last signal exchange between a given pair of devices. Interactions were classified into strong and weak links using the same criteria as those used by the app itself (see Appendix 1 for further details).

Measuring changes in social interactions

For each reported case and notified user, we calculated daily contact time beginning 7 days prior to the time of the event through 7 days after the time of the event. A 7-day pre- and post-event window was chosen to ensure each day of the week was equally represented in the analysis for all users. We defined the time of the event as Day 0 and rescaled the total cumulative contact time from Day − 7 to Day 0 to be 0.5, such that a user with constant daily total contact time would have a cumulative rescaled time of 1 between Days − 7 and + 7. We performed a difference-in-differences analysis to calculate whether there was a statistically significant change to a user’s rescaled contact time after reporting a case or receiving an exposure notification from the app.

In addition, we quantified each reported case and each notified user’s behavioral change by calculating the difference in total contact time for the 7-day period after the event minus the total contact time for the 7-day period before the event. From this we obtained distributions of behavioral change for total contact time, contact time from weak interactions only, and contact time from strong interactions only, and used a paired t-test to compare the behavioral change distributions from reported cases and notified users against their respective control groups.

Controlling for secular trends in social interactions

The difference-in-differences analysis may be impacted by secular trends or fluctuations in social interaction time. To control for these, we matched each user who experienced an event – reporting a case or receiving a notification – with a control group of users³⁶ who met the following criteria: (1) had an active app installation during the same window of time as the matched user, i.e., from Day − 7 to Day + 7, (2) had comparable contact with other app users from Day − 7 to Day 0, defined as ± 25% of the reporting user’s 7-day cumulative contact time before the time of the report, (3) did not report being a case between Day − 7 and + 7, (4) were not direct contacts of a case between Day − 7 and + 7, and (5) did not receive a notification of exposure between Days − 7 and + 7.

Power analysis

We constructed a synthetic dataset by artificially reducing the total weak interaction time by for users who received an exposure notification in the 7 days following. We varied the magnitude of the reduction to test the limit of detection for the methodology applied to our study cohort.

Sensitivity analyses

We assessed the sensitivity of the results to the window of time used for analysis relative to the day of case report or exposure notification. Alternative time windows used in sensitivity analyses were ± 1-day, ± 3-days, and ± 5-days. We further assessed the sensitivity of the results to the strength of interactions considered. The analysis was repeated as described but using data for only strong interactions, and again for only weak interactions.

There were 3221 users active in the Georgia Tech contact network over the time period of analysis. The number of active users peaked at 2766 during the week beginning January 22 (Fig. 1A), the week following start of classes on campus. Prior research found that this user population constituted between 6% and 26% of the on-campus population during that time^35,37. Cumulatively, 25 positive cases were reported to the app and 127 exposure notifications were transmitted by the app during the study period. No user reported a positive test more than once. On a weekly basis, cases (Fig. 1B) and exposure notifications (Fig. 1C) both peaked during the week beginning January 22, 2021. This was the week following the start of in-person classes, and also the week with the most positive cases were tested by the campus health services³⁸.

Among active users, the app recorded 159,360 daily interactions (24,499 strong and 134,861 weak) totaling 191,257 hours of contact time (52,800 strong and 138,456 weak). Both strong and weak interactions increased steadily over the start of the semester before reaching a plateau in February, likely reflecting students returning to in-person classes. Strong links were present at steady levels throughout the study period, even during winter break, while weak links became prevalent after classes began on January 15th and occurred more often during the 5-day work week than during weekends, likely reflecting university activities such as classes and extracurricular activities (Fig. 1D). The number of strong and weak links detected in a given week was highly heterogeneous across users, with a minority of users having large numbers of contacts (Appendix 2), consistent with prior social contact research^12,29,37. Analysis of the contact network revealed the formation and persistence of a giant connected component of hundreds of users, suggesting that the study population itself had frequent enough interactions to form a connected community (Appendix 3).

A total of 16 reported cases and 116 notified users met the inclusion criteria, logging a total of 25,912 person-hours of interaction time (1753 for cases, 24,159 for notified users) in the week preceding and following the case report or exposure notification. Reported cases were matched with 1071 controls, with an average of 64.3 controls per reported case. Notified users were matched with 12,674 controls, with an average of 109.3 controls per notified user.

Among reported cases, mean cumulative contact time was markedly lower over the 7 days following reporting the positive case (Fig. 2, left panel) compared to the 7 days preceding the report. Among matched controls, no differences in interaction time were observed over the 14-day period. Compared to controls, reported cases reduced mean cumulative contact time by 29% (standard error ± 9%) in the 7-day period following the case report (p = 0.002, Appendix 4). Individual trajectories of interaction time suggested heterogeneous behavior change after reporting a case, with some users logging no further interactions within days of reporting a case, while other users continued to interact with others over this period (Appendix 5). Among notified users (Fig. 2, right panel), mean cumulative contact time increased by 4.9% (standard error ± 3.6%) compared to matched controls in the 7-day period following exposure notification, a difference not found to be statistically significant (p = 0.17, Appendix 4). Likewise, no differences were observed among their matched controls. Trajectories of daily contact time for individual notified users had some were similar to the matched controls (Appendix 5).

Reported cases were found to decrease their total contact time by 30 hours on average following self-reports of testing positive, compared to a 10-hour increase in total contact time for the control group (p = 0.001). This result is consistent for total contact time as well as for weak links only, suggesting that many who reported a positive case reduced their contact time with others in the 7 days following the time of the report (Fig. 3, left column). By contrast, notified users were found to increase total contact time by 6.2 hours on average, compared to a 0.9 hour increase for the control group, a difference that was not statistically significant. Consistent with the results of the pre-post analysis, this implies that users did not change their contact behavior following an exposure notification (Fig. 3, right column).

In counterfactual analysis, artificially reducing weak interaction times by 11% following an exposure notification would have been enough to generate a statistically significant behavioral change. In sensitivity analysis, the results were robust to changing the length of the ± 7-day window of analysis, with the overall results recapitulated with a window as short as ± 1day (Appendix 6).

This app deployment provided what is, to our knowledge, the first opportunity to measure the effect of digital post-exposure notification on social distancing in the context of a pandemic in which pharmaceutical interventions were still unavailable. We observed a significant decrease in contact time among users who self-reported having tested positive for SARS-CoV-2, but no significant change in contact time among users who were notified that they had been in direct contact with a positive case. The observation that reported cases reduced contact time served as a positive control, confirming that the app’s digital tracing system was capable of detecting expected changes in behavior through the measurement of contact time. Our analysis also made use of a negative control – contemporaneous users with similar contact behavior who neither reported a case nor were notified of exposure – helping to control for secular trends in contact time. Our findings were robust to several sensitivity analyses, including varying the number of days before and after case report or exposure notification, and constructing a synthetic dataset in which contact times were artificially reduced to show the sensitivity of the chosen metrics for changes in contact time. These additional analyses support our finding that digital exposure notification did not prompt users to reduce time in social proximity to others on this US university campus.

At the time of the study it was university policy to ask community members who tested positive to self-isolate from the campus community.³² Our results suggest at least some degree of compliance with this policy. Among those who self-reported testing positive for SARS-CoV-2, reduction in social interaction time was driven by avoidance of weak links in the social network, i.e., relatively short, physically distant, and/or infrequent interactions. In the context of a university campus, reductions in weak links may reflect less time spent in shared spaces, where one can have numerous brief social contacts. Although the university also encouraged quarantine of exposed contacts, several factors may have contributed to the lack of measurable social distancing after exposure notification. First, unlike cases who were known to the campus health authorities, contacts who received digital exposure notifications were fully anonymous. There was no mechanism for campus health or community members to identify exposed contacts and hold them accountable for behavior change. Second, users may have considered it too costly to comply with social distancing following a notification from an app¹⁴. Unlike in cases where people are known to be sick, a notification from the app may not have been strong enough justification for students to miss class or for campus workers to take time off. Third, notified users are typically asymptomatic and may have perceived their risk of COVID-19 to be low, whereas a person with confirmed infection may change behavior due to symptoms and/or perceive themselves as a more tangible threat to the community. Fourth, at the time of the study, there may not have been sufficient buy-in from users to believe and respond to notifications from the app. Prior to this study, few digital contact tracing tools had been implemented in US, and none had been evaluated to verify that they were effective in reducing infection risk to the user and to others. Finally, while the period of analysis predated COVID-19 vaccination, it began nine months into the COVID-19 pandemic. By this time, users may have become inured to COVID-related warnings – a phenomenon known as “pandemic fatigue”⁴⁰ – or felt that universal mask-wearing and other general precautions sufficed to prevent transmission.

Our findings constitute one of the first empirical assessments of the impact of digital contact tracing in the United States⁴¹. Other widely-used digital tracing systems, such as GAEN, do not support retrospective evaluation because longitudinal interaction data are not stored centrally in order to protect user privacy^26,32,33. Because this app used alternative methods to protect privacy, we were able to observe the impact of exposure notifications on social interactions by tracking each user’s contact time with others over time. Analyses such as ours could provide additional transparency to users regarding the effects of digital tracing systems^10,41, as well as offering feedback to communities regarding their performance in quarantine and self-isolation, which in turn may increase voluntary participation with social distancing recommendations for exposed contacts⁴².

Our analysis has several important limitations. First, because of the rate of app adoption, we were only able to observe a fraction of the entire Georgia Tech campus community. It is possible that app users are not representative of the community as a whole. Additionally, we are unable to measure interactions between users and non-users, resulting in an under-estimate of contact density, although app adoption was high enough that users did interact at a high frequency. Second, our analysis was only able to measure changes in contact proximity, count, and duration following a notification, and was unable to measure other risk-reducing behaviors that may have occurred such as improved mask use, more frequent testing, or use of windows or outdoor gathering spaces. Relatedly, this observational study is only able to partially measure the efficacy of the app in terms of prompting users to change their behavior in response to a notification, and is not able to directly estimate whether observed changes in contact time resulted in a meaningful reduction in risk to the user or to others, which may depend on other factors such as air circulation and filtration. Lastly, our observation of reduced social interaction time after reporting a positive case may be subject to reverse causality. Individuals who attended high-contact gatherings, such as parties, would have recorded numerous weak interactions while at high risk and may have been predisposed to test positive in subsequent days. As a result, individuals who tested positive may have appeared to have a reduced contact time following the period of high contact. While we partially controlled for this by matching cases to controls with similar contact patterns over a 7-day period, we were unable to match control for all potentially relevant factors due to data limitations. However, a consistent finding across two independent approaches, each with distinct limitations, does increase our overall confidence in the conclusion that social interaction time was unchanged by exposure notification.

While we observed no evidence of social distancing after digital exposure notification, we nevertheless believe smartphone app technology to still have potential given the rich design space that remains to be explored¹⁰. For example, the NOVID app introduced a “radar” feature, which is designed to show users the network distance to the nearest reported case²⁹. The intent behind this design is that users might be incentive-aligned to use the radar to proactively adjust their behavior to avoid their own exposure to the disease in the first place, rather than be asked to restrict their behavior after being notified that they were already exposed to the disease, in order to avoid infecting others. This app deployment at Georgia Tech did not strongly disseminate information about this radar feature or educate users about how to respond to it³³. Future studies of digital contact tracing apps could explore “radars” and other app design features as strategies to encourage social distancing behaviors, and could make use of mixed methods approaches to assess whether and how users interact with and respond to those features.

Pandemics pose an existential threat to humanity and are increasing in frequency and magnitude due to rising global travel and urbanization⁴³. After the onset of the COVID-19 pandemic, many new digital contact tracing apps were deployed, often in partnership with regional or national public health authorities, as a way of providing individuals with information about infection risk to themselves and to others. This was done in haste in response to the emergent pandemic, without time to evaluate the efficacy of the apps in reducing risk, and with the most popular system (GAEN) not equipped to readily measure the effect of exposure notification on social interactions.

The present study represents, to our knowledge, the first empirical measurement of the effect of digital contact tracing on social interactions. Our results suggest that the users who reported a positive case did on the whole manage to reduce their contacts with others, but that users who received an exposure notification did not make measurable changes in their social contact time. Future research could include soliciting input from users regarding barriers and facilitators of reducing contact time after digital exposure notification, as well as end-user input on the design and implementation of future digital contact tracing apps¹⁰. One possible contributor is the level of trust between app users and the public health institutions that recommend and implement digital contact tracing app, e.g., whether users feel that their cooperation is important and necessary^14,42. Rigorous scientific assessment of the efficacy of digital contact tracing tools can contribute to building that trust¹⁵. Ideally, future digital contact tracing tools will find a balance between protecting user privacy while still enabling the sort of impact assessment presented here^41,44,45.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Müller J, Kretzschmar M. Contact tracing – Old models and new challenges. Infectious Disease Modelling 2021; 6: 222–31.
Fetzer T, Graeber T. Measuring the scientific effectiveness of contact tracing: Evidence from a natural experiment. Proceedings of the National Academy of Sciences of the United States of America 2021. DOI:10.1073/pnas.2100814118/-/DCSupplemental.
Zastrow M. Coronavirus contact-tracing apps: can they slow the spread of COVID-19? Nature 2020; published online May 19. DOI:10.1038/D41586-020-01514-2.
Kaushal A, Altman RB. Can Contact Tracing Work At COVID Scale? Health Affairs Forefront 2020; published online July 8. DOI:10.1377/FOREFRONT.20200630.746159.
Celentano J, Sachdev D, Hirose M, Ernst A, Reid M. Perspective Piece Mobilizing a COVID-19 Contact Tracing Workforce at Warp Speed: A Framework for Successful Program Implementation. Am J Trop Med Hyg 2021; 104: 1616–9.
Anglemyer A, Theresa HM Moore, Lisa Parker, et al. Digital contact tracing technologies in epidemics: A rapid review. Saudi Medical Journal 2020; 41: 1028.
Braithwaite I, Callender T, Bullock M, Aldridge RW. Automated and partly automated contact tracing: a systematic review to inform the control of COVID-19. The Lancet Digital Health 2020; 2: e607–21.
Jenniskens K, Bootsma MCJ, Damen JAAG, et al. Effectiveness of contact tracing apps for SARS-CoV-2: A rapid systematic review. BMJ Open 2021; 11. DOI:10.1136/bmjopen-2021-050519.
Cebrian M. The past, present and future of digital contact tracing. Nature Electronics 2021; 4: 2–4.
Criddle I, Hardy A, Smith G, Ranck T, Mondal M, Page X. Designing to Fight Pandemics: A Review of Literature and Identifying Design Patterns for COVID-19 Tracing Apps. In: M Kurosu, ed. International Conference on Human-Computer Interaction. 2022: 36–49.
Ferretti L, Wymant C, Kendall M, et al. Quantifying SARS-CoV-2 transmission suggests epidemic control with digital contact tracing. Science 2020; 368. DOI:10.1126/science.abb6936.
Barrat A, Cattuto C, Kivelä M, Lehmann S, Saramäki J. Effect of manual and digital contact tracing on COVID-19 outbreaks: A study on empirical contact data. Journal of the Royal Society Interface 2021; 18. DOI:10.1098/rsif.2020.1000.
Ge Y, Chen Z, Handel A, et al. The impact of social distancing, contact tracing, and case isolation interventions to suppress the COVID-19 epidemic: A modeling study. Epidemics 2021; 36. DOI:10.1016/j.epidem.2021.100483.
Lancet Digital Health. Contact tracing: digital health on the frontline. The Lancet Digital Health 2020; 2: e561.
Colizza V, Grill E, Mikolajczyk R, et al. Time to evaluate COVID-19 contact-tracing apps. Nature Medicine 2021; 27: 360–1.
Huerta R, Tsimring LS. Contact tracing and epidemics control in social networks. Physical Review E - Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics 2002; 66: 4.
Fraser C, Riley S, Anderson RM, Ferguson NM. Factors that make an infectious disease outbreak controllable. 2004 www.pnas.orgcgidoi10.1073pnas.0307506101.
Aleta A, Martín-Corral D, Pastore y Piontti A, et al. Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID-19. Nature Human Behaviour 2020; 4: 964–71.
Almagor J, Picascia S. Exploring the effectiveness of a COVID-19 contact tracing app using an agent-based model. Scientific Reports 2020; 10. DOI:10.1038/s41598-020-79000-y.
Abueg M, Hinch R, Wu N, et al. Modeling the effect of exposure notification and non-pharmaceutical interventions on COVID-19 transmission in Washington state. npj Digital Medicine 2021; 4. DOI:10.1038/s41746-021-00422-7.
Kendall M, Milsom L, Abeler-Dörner L, et al. Epidemiological changes on the Isle of Wight after the launch of the NHS Test and Trace programme: a preliminary analysis. The Lancet Digital Health 2020; 2: e658–66.
Wymant C, Ferretti L, Tsallis D, et al. The epidemiological impact of the NHS COVID-19 app. Nature 2021; 594: 408–12.
Vogt F, Haire B, Selvey L, Katelaris AL, Kaldor J. Effectiveness evaluation of digital contact tracing for COVID-19 in New South Wales, Australia. The Lancet Public Health 2022; 7. DOI:10.1016/S2468-2667(22)00010-X.
Rodríguez P, Graña S, Alvarez-León EE, et al. A population-based controlled experiment assessing the epidemiological impact of digital contact tracing. Nature Communications 2021; 12. DOI:10.1038/s41467-020-20817-6.
Kendall M, Tsallis D, Wymant C, et al. Epidemiological impacts of the NHS COVID-19 app in England and Wales throughout its first year. Nat Commun 2023; 14: 858.
Apple IG Inc. Privacy-Preserving Contact Tracing. 2022; published online Dec 18. https://web.archive.org/web/20221217214526/https://covid19.apple.com/contacttracing (accessed Dec 21, 2022).
Yeon D, Kwak M, Chung JB. Effectiveness of wireless emergency alerts for social distancing against COVID-19 in Korea. Scientific Reports 2022; 12. DOI:10.1038/s41598-022-06575-z.
Expii Inc. NOVID. 2022; published online Dec 21. https://www.novid.org/ (accessed Dec 21, 2022).
Loh PS, Bershteyn A, Yee SK. Lessons Learned in Piloting a Digital Personalized COVID-19 “Radar” on a University Campus. Public Health Reports 2022. DOI:10.1177/00333549221112532.
Cencetti G, Santin G, Longa A, et al. Digital proximity tracing on empirical contact networks for pandemic control. Nature Communications 2021; 12. DOI:10.1038/s41467-021-21809-w.
Leith DJ, Farrell S. Measurement-based evaluation of Google/Apple Exposure Notification API for proximity detection in a light-rail tram. PLOS ONE 2020; 15: e0239943.
Georgia Tech Stamps Health Services. Operational Status Updates Archives. 2022; published online Dec 8. https://web.archive.org/web/20220929203930/https://health.gatech.edu/coronavirus/archives (accessed Dec 21, 2022).
Georgia Tech News Center. NOVID Exposure Notification App Enlists Smartphones in Coronavirus Battle. 2020; published online Aug 17. https://web.archive.org/web/20220817130859/https://news.gatech.edu/news/2020/08/17/novid-exposure-notification-app-enlists-smartphones-coronavirus-battle (accessed Dec 21, 2022).
COVID-19: When to Quarantine | CDC. 2020; published online Dec 31. https://web.archive.org/web/20201231230119/https://www.cdc.gov/coronavirus/2019-ncov/if-you-are-sick/quarantine.html (accessed Feb 16, 2023).
Gibson G, Weitz JS, Shannon MP, et al. Surveillance-to-Diagnostic Testing Program for Asymptomatic SARS-CoV-2 Infections on a Large, Urban Campus in Fall 2020. Epidemiology 2022; 33: 209–16.
Morgan AC, Way SF, Hoefer MJD, Larremore DB, Galesic M, Clauset A. The unequal impact of parenthood in academia. 2021 https://www.science.org.
Georgia Tech Institutional Research and Planning. Fact Book. 2022; published online Dec 21. https://web.archive.org/web/20221123073809/https://irp.gatech.edu/fact-book (accessed Dec 21, 2022).
Georgia Tech Stamps Health Services. Covid-19 Positive Case Detail Report Archives. 2022; published online Feb 3. https://web.archive.org/web/20220203172751/https://health.gatech.edu/coronavirus/health-alerts/archives (accessed Dec 21, 2022).
Sapiezynski P, Stopczynski A, Lassen DD, Lehmann S. Interaction data from the Copenhagen Networks Study. Sci Data 2019; 6: 315.
World Health Organization. Regional Office for Europe. Pandemic fatigue – reinvigorating the public to prevent COVID-19: policy framework for supporting pandemic prevention and management. World Health Organization. Regional Office for Europe, 2020 https://apps.who.int/iris/handle/10665/335820 (accessed April 14, 2023).
Li T, Yang J, Faklaris C, et al. Decentralized is not risk-free: Understanding public perceptions of privacy-utility trade-offs in COVID-19 contact-tracing apps. 2020.
Kaptchuk G, Goldstein DG, Hargittai E, Hofman J, Redmiles EM. How good is good enough for COVID19 apps? The influence of benefits, accuracy, and privacy on willingness to adopt. 2020; published online May 8. http://arxiv.org/abs/2005.04343.
Santiago-Alarcon D, MacGregor-Fors I. Cities and pandemics: urban areas are ground zero for the transmission of emerging human infectious diseases. Journal of Urban Ecology 2020; 6: juaa012.
Akinbi A, Forshaw M, Blinkhorn V. Contact tracing apps for the COVID-19 pandemic: a systematic literature review of challenges and future directions for neo-liberal societies. Health Information Science and Systems 2021; 9. DOI:10.1007/s13755-021-00147-7.
Hernández-Orallo E, Manzoni P, Calafate CT, Cano JC. A methodology for evaluating digital contact tracing apps based on the COVID-19 experience. Scientific Reports 2022; 12. DOI:10.1038/s41598-022-17024-2.

There is a conflict of interest Daniel T Citron declares no competing interests

Po-Shen Loh has the following interests to report:

Expii, Inc. - PSL is the founder of this company, which produced the NOVID app that supplied the anonymized data set for the study
- Effective Ventures - Provided funding to Expii, Inc. for the creation and on-going development of the NOVID app, independently of this study. (The app would have been created anyway, even if the study was not conducted.)
- Longview Philanthropy - Provided funding to Expii, Inc. for the creation and on-going development of the NOVID app, independently of this study. (The app would have been created anyway, even if the study was not conducted.)
- Tansy Charitable Foundation - Provided funding to Expii, Inc. for the creation and on-going development of the NOVID app, independently of this study. (The app would have been created anyway, even if the study was not conducted.)
- Idea Foundry - 501(c)(3) fiscal sponsor which received philanthropic donations for the development of the NOVID app, and contracted Expii, Inc. to perform the work
Payment or honoraria for lectures, presentations, etc
- Georgia State University - Gave talk about incentive-aligned app design for pandemic control
- Davidson Institute - Gave talk about incentive-aligned app design for pandemic control
- Math for America - Gave talk series about incentive-aligned app design for pandemic control
Patents planned, issued, or pending - Expii, Inc. P. Loh, H. Cai, F. Campos, J. Choi, T. Chu, D. Dijour, B. Huffman, M. Jaffer, Z. Scully, B. Wang, L. Wang, P. Wang, Crowdsourced contact tracing, U.S. Patent Application #20210400428, filed Jun. 4, 2022
Stock or stock options
- Expii, Inc. - PSL the majority owner of this company, which produced the NOVID app that supplied the anonymized data set for this study

Shannon Yee has the following interests to report:

Grants or contracts from any entity
- Bill & Melinda Gates Foundation - Reinventing the toilet
- US Dept of Energy - Thermophysical measurements
Consulting Fees
- Gates Ventures - Consulting
Payment or honoraria for lectures, presentations, etc
- Kimberly-Clark Professional, Speaker honoraria
Leadership or ficuciary role in other board, society, or other advocacy group, paid or unpaid
- Board of Directors for the Materials Research Society

Anna Bershteyn has the following interests to report:

Grants or contracts from any entity
- US National Institutes of Health
- Bill & Melinda Gates Foundation
- NYC Department of Health and Mental Hygiene
Consulting Fees
- Gates Ventures - Personal consulting

DigitalPostExposureNotificationappendices.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Changes in social contact time following digital SARS-CoV-2 post-exposure notification: A retrospective case-control study at a US university

Status:

Version 1

Abstract

Figures

Introduction

Methods

Definition of app users at Georgia Tech

Inclusion criteria for analysis

Calculating contact time

Measuring changes in social interactions

Controlling for secular trends in social interactions

Power analysis

Sensitivity analyses

Results

Discussion

Declarations

Funding

References

Additional Declarations

Supplementary Files

Status:

Version 1