Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Mar 26.
Published before final editing as: Exp Clin Psychopharmacol. 2026 Feb 9:10.1037/pha0000827. doi: 10.1037/pha0000827

The Utility of Bluetooth and Smartphone Technology to Detect Peer Contact

Nancy P Barnett 1, Matthew K Meisel 1, Alexander W Sokolovsky 1, Shannon R Forkus 2, Kristina M Jackson 3
PMCID: PMC13014808  NIHMSID: NIHMS2143124  PMID: 41661567

Abstract

Adolescents and young adults often engage in risk behaviors in close proximity to peers. Detecting peer presence could identify high-risk contexts, but typically relies on self-report, which is prone to bias. This study evaluated the feasibility, acceptability, functionality, and validity of a smartphone-based passive detection system using Bluetooth beacons to identify real-world peer proximity. Twenty-one young adult participants (38% women) and their peers (N=55; 40% women) completed a 3-week protocol, during which peers carried a small Bluetooth beacon that was detected by participant smartphones. Participants indicated the presence of peers on beacon signal-contingent ecological momentary assessment (EMA) reports and three daily random reports. Feasibility of participant recruitment was low, primarily due to Android OS updates requiring app revisions that interrupted recruitment. However, feasibility of peer enrollment was high, occurring rapidly but at a lower-than-expected number. Response latencies to signal-contingent and random reports were similar, indicating good feasibility of the EMA report procedures. Acceptability, reflected in high retention for participants and peers, participant self-report ratings, and good EMA report response rates (76–79%), was high. Functionality was moderate; problems with the app were reported by almost half of participants, and functionality ratings were lower than for acceptability. For validity, the beacon detection technology identified 61% of participant-reported encounters (true positives), with 5.6% false positives. False negatives (39%) were likely due to peer noncompliance or misreporting. Results support the initial utility of Bluetooth-based passive detection for identifying peer presence in real time, offering potential for use in just-in-time interventions targeting health-risk behaviors.

Keywords: Bluetooth-based proximity sensing, Ecological Momentary Assessment, Peer detection, Feasibility and Acceptability, Validity

High-risk behavior among adolescents and young adults commonly occurs in the presence of peers (Beck et al., 2013; Hoeben et al., 2016; Meisel et al., 2018). Self-report of risk behavior is the gold standard in alcohol research, including in event sampling, but is limited by multiple biases, including inaccuracies in retrospective report (Stone et al., 2023) and for alcohol specifically, drinking-induced impairment and social desirability (Del Boca & Darkes, 2003). Ecological momentary assessment (EMA) and other fine-grained data collection methods obtain real-time information about individuals and their environment via self-initiated or prompted reports (Shiffman et al., 2008). Although EMA minimizes recall bias, biases due to social desirability (Stone et al., 2023) and missingness due to nonresponse still exist (Kovalchik et al., 2018; Sokolovsky et al., 2014). An additional limitation of EMA when applying it to the study of social context is that it requires that individuals be both aware of and able to report on peer presence or influence.

Passive detection of peers can overcome limitations of self-report and provide additional valuable information about the social context of risk behaviors (Girolami et al., 2020; Harari et al., 2017). Seminal work has used smartphone sensors to detect alcohol use (Santani et al., 2018) and subsequent work used the same dataset to show how data from the smartphone sensors could help characterize the social contexts of alcohol use (Meegahapola et al., 2021). One smartphone-based technology with potential applications for understanding social contexts is Bluetooth, a ubiquitous connectivity protocol embedded in mobile phones and other wearable devices. Coupled with software that identifies nearby Bluetooth beacons (e.g., Apple Air Tags), smartphone applications can assess the duration and frequency of interpersonal interactions (Eagle & Pentland, 2006; Girolami et al., 2020). Researchers have integrated Bluetooth and other proximity sensing methods to understand social interactions in school environments (Hernández-Heredia et al., 2024), and during the COVID-19 pandemic to study disease spread (Hatke et al., 2020; Maghdid & Ghafoor, 2020), facilitate contact tracing (Admiraal et al., 2022), and provide notifications to exposed individuals (Elder et al., 2024). Most of this prior work focused on the technical development of the various approaches, and so provides important foundational work, but included little or no evaluation of feasibility or acceptability among human participants (but see Maharjan et al., 2020). Moreover, evaluations have tended to be experimental with tightly controlled interactions or conducted in restricted settings (e.g., Baronti et al., 2020; Girolami et al., 2020), and generally have not intentionally investigated this technology among close friends.

There is clear value to developing passive, in-real-time detection of risk behaviors that commonly occur in the presence of others. Any circumstance in which social mixing is relevant could benefit from this methodology, including, for example, the measurement of infectious disease transmission. The objectives of this paper were to evaluate (a) the feasibility, (b) acceptability, (c) functionality, and (d) validity of using passive detection methods with young adults and their peers to identify when two individuals are physically proximal in real-world situations.

Methods

Design

A small Bluetooth-based beacon was coupled with a smartphone application; the application detected and recorded the beacon’s unique signal. Index participants and participant-identified peers participated for three weeks, during which the peers were expected to always carry the beacon, and the participants answered reports about who of their peers were nearby. Detected beacon encounters triggered EMA signal-contingent reports in which the participant indicated or denied peer presence. Random reports, identical to signal-contingent reports, also were triggered three times a day. All procedures were approved by the Brown University Institutional Review Board.

Participants

Index participants.

Participants were young adults (N=21; Mage=23.6; 38% women, 38% White). Inclusion criteria were being 18–24 years old, able to read English, having an Android phone with data plan used daily1, and willingness to ask peers to participate. Exclusion criteria were being in treatment or seeking treatment for alcohol or substance use or having plans to be out of their usual routine during the study.

Peer participants.

Peers of index participants (N=55; 40% women; race not measured) were identified by the participant at baseline as someone the participant expected to have in-person contact with in the coming weeks.

Procedures

University notices and emails to student listservs advertised the research. After an online screener, eligible respondents attended an in-person session with research staff at which the baseline assessment was administered, procedures described, and the EMA application installed on the participant’s phone. Participants were instructed that they were expected to complete reports on the EMA app every day for 21 days. The Social Network Interview (SNI; adapted from Barnett et al., 2019 and Longabaugh & Zywiak, 2002) was administered to identify close peers as potential peer participants. A post-participation survey was conducted at the end of the EMA phase. Participants received $50 for the baseline, $5 per day for completing at least two EMA reports, $20 each week for completing 80% or more reports, and $40 for the post-participation survey. Compensation was provided as gift cards to an online vendor and the most a participant could receive was $255.

Peer participant enrollment and beacon provision.

Participants invited their peers to participate by sharing a weblink from research staff which provided information about project procedures, the opportunity to consent to participate, and collected contact information. Peer participation entailed agreeing to carry a beacon for three weeks and answering a web survey each week. Beacons were delivered to the peers along with a small adhesive patch and a key ring to facilitate carrying the beacon, and a self-addressed stamped envelope to facilitate beacon return. Peers received $10 for each weekly survey and $10 for returning the beacon. The most a peer participant could receive was $40.

Smartphone application.

The smartphone application was developed by MEI Research (PiLR Health) and had two primary functions: a background process that scanned for the identifiable Bluetooth beacons and recorded their presence, and an EMA function that triggered EMA random and (beacon) signal-contingent reports. In addition, we created a customized “friends list” for each participant in which the names of up to six peers identified in the SNI interview (including all who carried a beacon) was presented to participants in all EMA reports.

Signal-contingent reports were triggered and one initial notification was sent when the presence of the peer beacon met criteria for being in close proximity, which we defined as 15 minutes (called an “actual encounter”), consistent with other work (Hatke et al., 2020). No reminder notifications were sent for these reports. It was possible for signal-contingent reports associated with more than one peer to overlap with each other and to overlap with random reports. No reports were suppressed, even if they overlapped with other reports. The three random reports per day were randomly triggered at noon to 6 PM, 6 PM to 9 PM, and 9 PM to midnight. Reminder notifications for random reports were sent 15 and 30 minutes after the first. Both signal-contingent and random reports expired after 1 hour.

Measures

Baseline.

Demographics information was collected and the SNI was administered which collected peer information (names, gender, likelihood of future interaction).

Beacon detection of peer presence.

The beacon (Kontakt.io asset tag) is 4.4 cm square and 1 cm thick (Figure 1) and produces Bluetooth Low Energy signals that can be detected by a receiver (in this case, our EMA app). The hardware vendor provides the ability to modify beacon settings, which allowed us to specify the advertising interval (how often the beacon sends out its signal) and transmission strength to calibrate our desired detection distance. Transmission power was set to 3 (on a range from 1-low to 7-high). Each beacon has a unique ID that is transmitted with the signal and is recorded by the participant’s app, and only the participant’s app records the presence of (and triggers a signal-contingent report for) their peers’ beacons, allowing us to associate the beacon encounter with the participant and peer. For this research, detection of peer presence was reflected in transient and actual encounters, which occurred whenever a beacon was detected, with signal detection criteria (i.e., Received Signal Strength Indicator [RSSI]; the measurement of the power of a wireless signal) corresponding to a range of approximately 15 feet for study beacons at the specified transmission power. Transient encounters were encounters that lasted less than 15 minutes; once a given beacon was detected continuously for 15 minutes (with no continuous period ≥ 2 minutes without a signal), the transient encounter was converted to an actual encounter and a signal-contingent report notification was triggered. After 60 minutes of beacon signal non-detection, an actual encounter was assumed to have ended and the time stamp at the point of initial non-detection (called the end encounter) was recorded. After the end of an encounter was recorded, a new encounter for that peer could start if the beacon were detected. Additional details about the beacon detection algorithm have been previously published (Barnett et al., 2024).

Figure 1.

Figure 1.

Open in a new tab

Kontakt.io Asset Tags

EMA reports.

The random and signal-contingent notifications and reports were identical to minimize participants showing differential attention. On all reports, participants were presented with their list of (six) peers and asked, “In the past hour, who have you been around for any length of time?” Any peers who were endorsed were included in the answer options to the next question: “In the past hour, who have you been within 15 feet of for at least 15 minutes?” Each peer on the list had a specific ID associated with their participant and their beacon if relevant. For all reports, timestamps for notifications and report submission were recorded. From these timestamps we calculated notification latency for the signal-contingent reports, i.e., the time between the actual encounter start and the notification, expected to be 15 minutes. We also calculated response latency for both report types, defined as the minutes from notification delivery to report submission.

Post-participation assessment.

Six items from the modified System Usability Scale (SUS) (Bangor et al., 2008) measured acceptability, for example, “I found the app unnecessarily complex” and five items that were created to reflect project procedures (e.g., “I didn’t like that I had to change settings on my phone to use the app”) scored on a five-point scale from 1 “strongly disagree” to 5 “strongly agree”. Four questions were created to evaluate functionality (e.g., “The app drained my battery”) with the same response options. Items were averaged separately for acceptability and functionality.

Peer weekly surveys.

At the end of each week, peers indicated on a web survey the days or parts of days in the previous week when they were not carrying the beacon. These days were used to evaluate peer compliance. Days when peers reported not carrying their beacons were excluded for validity analysis.

Definition of Project Objectives

Feasibility for participants was reflected in the time it took to enroll the participant sample and in report response latency, i.e., the time between report notification and EMA report submission for both random and signal-contingent reports. Feasibility of peer participation was evaluated as the time needed to enroll the first peer, the proportion of participants who enrolled 3 peers (target 90%), and the beacon return rate (target 80%).

Acceptability was defined as the proportion of enrolled participants who completed the protocol (target of 90%), EMA report completion rates (target of 75%), and high ratings on the SUS acceptability items. Acceptability of peer participation was defined as the proportion of peers who completed the study protocol and high compliance with study procedures, defined as proportion of days the beacon was carried and 80% or better completion of weekly surveys.

Functionality of notifications was defined as the proportion of notifications that were delivered in 15 minutes as expected (notification latency; the time between an actual encounter start and the sending of the signal-contingent notification). Functionality of app stability was evaluated using the proportion of participants who reported functionality problems and their evaluation of functionality on post-participation SUS survey items.

Validity was defined as the concordance between participant reports of peer presence (considered ground truth) and beacon detection. We classified data as true positive (the proportion of participant reports of peer presence when a beacon was detected), false positive (the proportion of participant reports of denied peer presence when a beacon was detected), false negative (the proportion of participant reports of peer presence when a beacon was not detected), and true negative (the proportion of participant reports of denied peer presence when a beacon was not detected).

Data Management and Analysis Plan

Data preprocessing.

We extracted metadata from all transient and actual encounters to facilitate linking beacon detection of peer presence to self-report of peer presence. Metadata included: 1) session id - a unique encounter identifier that was propagated to the signal-contingent self-report tied to a given actual encounter to facilitate direct linking to the beacon information; 2) encounter start and end timestamps - used to link encounters to self-reported peer presence in either random reports or signal-contingent reports where the peer’s presence was reported but was not the trigger for the report (i.e., observations where session id could not be used); and 3) orphan status - an indicator of whether either the start or end timestamp for an encounter was missing (i.e., one of the timestamps was orphaned). Having a missing start timestamp represented a technical error but only occurred on n=2 occasions. We identified 50 encounters without an end timestamp of which n=27 were a ‘forced end’ when there were technical issues that prevented the detection of an end encounter so the encounter was forced by the developers to end, n=17 when participation ended prior to the end of an ongoing encounter (i.e., the last day of participation), and n=6 of unknown origin (possibly related to phone shutdown in the course of the encounter). For actual encounters with no end timestamp, we set the end timestamp to 75 minutes after the start timestamp, conservatively 10 minutes shorter than the median beacon-detected encounter length observed in the study. Thus, for each transient or actual encounter, we recorded the index participant ID, session ID, peer ID, start and end timestamp, orphan status, and a recoded end timestamp where relevant.

Participant report data for validity analysis came from random reports (N=1,051) and signal-contingent reports (N=368). Data was initially structured such that each row represented valid data on the presence or absence of a peer participant for each index participant from any EMA report (i.e., [1,051 + 368] = 1,419 * 6 possible peers = 8,514), then reduced to include only reports of peers who had beacons (excluding days prior to peers receiving beacons and days when peers reported not carrying them). Since more than one peer could be present at the same time, and since report notifications could overlap, we used information from all reports, regardless of the source. We created two subsets of these data: (1) observations where peers were reported present on participant self-report in order to compute true positives and false negatives (n=946); and (2) observations where peers were reported by participants as not around or around but not within 15 feet, in order to compute false positives and true negatives (n=2,408). Thus, a total of n=3,354 self-reported observations of peer presence or absence were used in our final analyses.

Linking report and encounter data.

Signal-contingent reports of peer presence where the peer was the one who triggered the report were matched to beacon encounter data based on session ID. Other EMA reports of peer presence were either (a) reported in a random report or (b) reported in a signal-contingent report but when the peer was not the one who triggered the report notification. These were matched to beacon encounter events (transient or actual) by participant ID, peer ID, and time submitted (where the time of submission of the report had to fall between the start and end encounter of the beacon encounter).2

Analysis.

Frequencies, means, and standard deviations, of feasibility, acceptability, functionality, and validity measures were computed. T-tests compared response rates and acceptability/functionality scores. Generalized estimating equations were conducted to examine the correspondence between beacon and EMA reports on the notification latency and response latency (in minutes). The models employed a normal distribution with an identity link function and robust standard errors to account for repeated measures within subjects.

Transparency and Openness

We report all data exclusions, manipulations, and measures in the study and the sample size utilized in analyses. We explain in the methods how analytic decisions were made. Measures and syntax are available from the corresponding author. The data are not publicly available. The protocol details were published (Barnett et al., 2024) but analyses were not preregistered.

Results

Descriptive Information

Participants had an average of M=2.62 (SD=1.12; median=3; range=1–5) peers carrying beacons during the study period, corresponding to M=49.19 (SD=21.32; median=51; range=10–87) peer beacon-days per participant. The average number of actual encounters per participant during the study period was 22.3 (SD=21.2, median=19, range=1–71). The modal number of actual encounters per day per participant (i.e., the most common number of actual encounters and event-contingent surveys triggered) was 0. On days when there were one or more encounters, the mode was 1, the median was 2. The highest number of actual encounters per day averaged 3.05 (median=3; range=1–7). In other words, the most common number of surveys triggered by the beacon was 0 per day, on days when they were triggered the number was quite low, and even the day with the highest number per participant was manageable.

Peers carried beacons for an average of M=18.98 days (SD=3.41; median=20; range=10–22). Among the 55 peers who carried beacons, 44 (80%) experienced at least one detected encounter (transient or actual) during the course of the study. Among peers who triggered an actual encounter, the average number of encounters per peer was 16.52 (SD=19.05; median=7; range=1–64).

Feasibility

The feasibility of recruiting our targeted sample was somewhat low; it took one year to fully enroll our sample due to unexpected disruptions in app functioning. These disruptions, which were in part a function of updates to the Android OS, required developer diagnosis and fixes and considerable subsequent testing by our research team, during which time we suspended participant enrollment. To facilitate project completion, we relaxed our inclusion criteria related to alcohol use. Participant response latency (time from notification to EMA response submission) was virtually identical for signal-contingent reports (M=14.1 mins; median=7 mins) and random reports (M=14.3 mins; median=7 mins). Feasibility of recruiting at least one peer was high: on average it took 1.8 days to enroll the first peer, but only 57% of our sample enrolled 3 peers, missing our target of 90%. The return rate of the beacons from peers was 82% and met our expected rate of 80%.

Acceptability

Of the 21 enrolled participants, 19 (90%) completed the three-week protocol. The response rate for signal-contingent reports was 76% (SD=0.20; median=.82; range=.25–1) and for random reports was 79% (SD=0.13; median=0.81; range=0.47–0.97), a nonsignificant difference, t(20)=.74, p=.47. Participants reported high acceptability of the app, with an average rating of M=4.13 (SD=.43) on a five-point scale, where higher scores indicated greater acceptability (Table 1). No peers withdrew from the study, peers responded to 93% of weekly surveys (target 80%), and they reported carrying the beacon on 96% of days.

Table 1.

Acceptability Items

Item M (SD)
I imagine that most people would learn to use this app quickly 4.71 (.61)
I thought the app was easy to use 4.50 (.52)
I would use this app again in a different research study 4.21 (.89)
The app worked as expected 4.21 (.43)
I felt very confident using the app 4.07 (.92)
I didn’t like that I had to change settings on my phone to use the app 2.50 (1.29)
There were too many survey notifications 2.43 (.76)
I found the app very awkward to use 2.21 (.82)
I found the app unnecessarily complex 1.71 (.61)
I needed to learn a lot of things before I could get going with the app 1.64 (.93)
The app affected other apps on my phone 1.43 (.76)
Average score for acceptability items † 4.16 (.43)
Open in a new tab

Note. Answer options: 1=strongly disagree, 2=somewhat disagree, 3=neither agree nor disagree, 4=somewhat agree, 5=strongly agree.

†

Item scoring was adjusted (since some items were reverse scored) to calculate the average; a high average score reflects high agreement with acceptability.

Functionality

Some issues with app functionality were presented above in the feasibility results. There were 453 actual beacon encounters, 3 of which had no notification timestamp. Of the remaining 450, 418 (93%) had notifications triggered at exactly 15 minutes, and 95% in 15–20 minutes. There was a significant difference in notification latency between signal-contingent reports that were answered vs. unanswered (p = .004). Among participants, 48% reported one or more problems with the app. Functionality items are in Table 2 and reflect moderate scores. A paired t-test comparing acceptability and functionality average scores showed significantly lower scores for functionality, p<.001.

Table 2.

Functionality Items

Item M (SD)
The app drained my battery 3.07 (1.54)
I encountered problems using the app 3.07 (1.07)
The app crashed a lot 4.07 (1.07)
I had to reinstall the app 4.36 (1.23)
Average score for functionality items † 3.64 (.78)
Open in a new tab

Note. Answer options: 1=strongly disagree, 2=somewhat disagree, 3=neither agree nor disagree, 4=somewhat agree, 5=strongly agree.

†

Higher scores reflect lower functionality. The reverse-scored average was 2.35 (SD=.78) indicating disagreement with functionality.

Validity of Beacon Detection

True positives.

There were 946 reports of peer presence on either a random or signal-contingent report. Of these, 577 matched with either a transient or actual beacon encounter, for a 61.0% overall detection rate (see Table 3). Excluding two participants with only one encounter each, the mean detection rate per participant of reported peer encounters was M=.56 (SD=.18; range .21-.84) with M=49.7 (SD=41.6) peer encounters. Of the 946 reports, 535 encounters were from random reports (43.9% detection), and 411 encounters were from signal-contingent reports (83.2% detection; Table 4). As explained above, a peer could be indicated as present on a signal-contingent report that was triggered by a different peer’s beacon. Of the reports of peer presence on signal-contingent reports, 84% were matched to the beacon that triggered the report, and 16% were matched because the peer was reported on a signal-contingent report that was not from their own beacon.3 We can interpret this as reflecting that approximately one in six beacon encounters detected multiple peers.

Table 3.

Validity of Beacon Detection

Participant Self-report
Peer present Peer not present
Beacon Detection
Positive (beacon detected) 577 (61.0%) (True Positive) 134 (5.6%) (False Positive)
Negative (no beacon detected) 369 (39.0%) (False Negative) 2,274 (94.4%) (True Negative)
Totals 946 2,408
Open in a new tab

Notes. Data were restricted to days when peers reported carrying the beacon (i.e., days when they should be able to be detected). Peer presence was indicated by participants who selected the peer on the item “In the past hour, who have you been within 15 feet of for at least 15 minutes?” in either a random or signal-contingent report. Beacon detection utilized both transient and actual encounters.

Table 4.

Source of Participant Report of Peer Presence and Beacon Detection of that Presence.

Beacon Detected
Random Report 235/535 (43.9%)
Beacon Report 342/411 (83.2%)
Total 577/946 (61.0%)
Open in a new tab

False positives.

Of the 2,408 observations when peers were reported not to be present, 134 (5.6%) had an associated beacon detection (Table 3). One explanation for false positives is a delay or error in notification resulting in the beacon triggering a report but the peer being absent when the participant answered the report. We analyzed notification latency and found that it was longer when the participant (subsequently) denied vs. reported the peer was present (p=.036; Table 5). A second possible explanation for false positives is that participants took longer to answer when peers were absent (e.g., the peer left before the report was answered), but we did not find this to be the case; there was no difference in the time to answer when the peer presence was denied vs. reported (p=.187; Table 5), so this explanation was not supported.

Table 5.

Notification Latency and Answer Latency in Minutes for Actual Beacon Encounters According to Peer Presence Status

Latency Type Peer presence reported (n = 317) Peer presence denied (n = 52) Statistic B (SE) [95% CI]
Notification latency (mins) M (SD) 20.0 (14.9) 69.7 (154.7) 54.68 (26.12) [3.48, 105.87], p =.036
Answer latency (mins) M (SD) 14.9 (20.36) 10.5 (15.68) −1.97 (1.49) [−4.89, .96], p =.187
Open in a new tab

Notes. Includes only actual encounters. Notification latency was the difference between the start of the actual encounter and the time the signal-contingent report was sent. Answer latency was the difference between the time the signal-contingent report was sent and the time the report was submitted by the participant. Analyses were generalized estimating equation models (normal distribution with an identity link function and an unstructured covariance matrix) and robust standard errors.

False negatives.

The false-negative rate was 39%. One possible explanation for the participant reporting peer presence when there was no beacon encounter detected is peer noncompliance. We asked peers to respond to surveys once a week indicating which days they were not carrying the beacon for any interval. As reported in acceptability above, survey response rates were good, and in our analysis we filtered out days on which peers reported not carrying their beacon for the validity analysis. However, peer recollection may be poor, there could be a bias in self-reporting compliance, or the survey itself may have been confusing in identifying the days we were asking them to report on.

Discussion

The aim of this investigation was to evaluate the feasibility, acceptability, functionality, and validity of using Bluetooth technology to detect social contact between young adults and their peers. Overall, we found the technology and methods to be feasible and acceptable. Functionality issues hindered data collection significantly due to necessary technological adjustments and compromised the feasibility of recruiting participants quickly. Validity, reflected in the true-positive rate of 61%, was lower than exploratory or feasibility study expectations, but nevertheless shows promise as rates were high for some participants and detection failures due to peer compliance are modifiable. Given this, while this technology and methods have strong potential to passively detect contact, technological considerations must be addressed for such a system to be more successfully used to accurately detect social contact between individuals.

Strengths of our methods include that we collected information about all peers nearby on every report, which allowed us to pair reports of peer presence with beacon encounter data even if the signal-contingent report for a peer was not answered. This could happen when two signal-contingent reports were available but only one answered, or if a signal-contingent report was not answered but a close-in-time random report was answered. Our system was designed to differentiate between a transient and actual beacon encounter for purposes of the signal-contingent notification, but transient encounters (<15 minutes) were captured and included in our analysis, reflecting that different length encounters can be detected and evaluated. We also created a personalized friend list for every participant, facilitating report responding and data linking. Ensuring the signal-contingent and random reports were identical minimized the likelihood of reactivity in responding, which was supported by response rates and response latency not differing between report types.

Feasibility

Feasibility was evaluated based on time to enroll participants and peers, the percentage of participants who enrolled three peers, beacon return rates, and latency of EMA responses. Recruiting participants was challenging due primarily to requiring Android phones and subsequent delays following Android OS updates. It took only two days to recruit at least one peer participant, although only a little over half the sample (57%) enrolled three peers. Over four-fifths of peer participants returned their beacon showing good feasibility of this procedure. On average it took approximately 15 minutes for index participants to respond to random and signal-contingent reports, consistent with other EMA research (Boukhechba et al., 2018). Taken together, the protocol was moderately feasible.

Acceptability

We found high acceptability among participants, with 90% of participants completing the full protocol, 76–79% completion rates of the EMA reports and generally high ratings on our acceptability measure. For peers, there were high response rates on the weekly surveys, peers reported carrying the beacons with them on 96% of days, and none withdrew. This high acceptability is consistent with studies using passive methods for detecting alcohol use (Rosenberg et al., 2023), Bluetooth technology to detect social contact (i.e., contact tracing; Li et al., 2023; Shelby et al., 2021), and combinations of social network and daily assessments (Reblin et al., 2022).

Functionality

Roughly half of the participants reported issues with the app, and ratings of functionality items were moderate and lower than acceptability items. Notifications were largely correctly delivered to participants; approximately 95% of notifications were triggered within 5 minutes of an actual encounter, although longer latency resulted in a lower answering rate. There were multiple interruptions while building and implementing the technology (similar technical issues have been noted by others; Girolami et al., 2020), resulting in unanticipated delays.

Validity (accuracy)

Validity was defined as the concordance between beacon encounters and participant reports of peer presence. We classified data as true positive, false positive, false negative, and true negative using participant report from signal-contingent and random reports as ground truth. There was considerable variability across participants in the number of encounters they reported having with their peers and the proportion of those encounters that were detected by the beacon (ranging from 21% to 84%). This suggests that individuals vary in the extent to which participants spend time around their peers, the extent to which they notice peers being nearby, and/or in the compliance of peers carrying the beacon.

A true-positive case was indexed by the proportion of participant reports of peer presence that was associated with a beacon encounter; this rate was 61%. While the detection rate is lower than desirable for sensitivity evaluation, it is consistent with sensitivity rates of wearables for detection of more complex health behaviors, including physical activity (Doherty et al., 2024) and COVID-19 detection (Cheong et al., 2022) and is reasonable given the developmental stage of this research. Moreover, capturing almost two out of three contacts is likely better than relying on individuals to self-report risky situations (with its inherent bias). Further, considering only times when the person reported on peer presence when explicitly prompted by a signal-continent report (because an actual encounter was detected), the confirmation rate was 83%, reflecting very high confirmation of beacon-detected encounters. In sum, using beacons or similar technology has promise for detecting proximal interactions that could be used to accurately prompt just-in-time interventions (e.g., Nahum-Shani et al., 2018).

False positives, the proportion of participant reports of denied peer presence when a beacon was detected were very low in this study (<6%). We anticipated that a delay in signal-contingent report notification would be a main contributor to false positives (Barnett et al., 2024), as the longer the notification delay, the higher the likelihood the peer would no longer be in the participant’s presence. In support of this conclusion, the notification latency was over three times longer when peer presence was denied vs reported, a significant difference. This technology issue reflects the importance of developing a system that prioritizes timely notifications. We also anticipated that a delay between when the participant’s phone was notified and when they submitted the report would account for false positives (i.e., there would be a higher likelihood the peer would no longer be there). This was not the case, as the latency to report submission was no different when peer presence was reported vs. denied. Together these findings suggest that notification latency (i.e., technology) but not answer latency (i.e., participant behavior) was a factor responsible for the (few) false positives that were observed. The latter finding of no difference in answering latency importantly indicates that participants did not delay answering reports when they were with peers, reflecting positively on the validity of our methods.

Other human factors that could account for false positives that could not be evaluated include inaccurate participant report whereby a peer is physically present, but the participant is not aware, and poor peer compliance such that the beacon is present, but the peer is not. In the former case, the participant may be in the same dwelling as the peer, but the peer is in a different room. In the latter case, the participant may be at a location (e.g., their residence) where the beacon is present only because the peer (e.g., a roommate) inadvertently left it there.

False negatives (39%) were cases when participants reported peer presence, but a beacon was not detected. The most likely reasons for these cases are human factors specific to the peer and participant. Peers reported carrying the beacons with them on nearly all (96%) days, but self-presentation or social desirability biases or poor recall may have inflated this number, and our approach to collecting past-week reports from peers may have contributed to poor reporting detail on their compliance. Our sense that compliance with beacon carrying was a primary issue is supported by findings from Admiraal and colleagues (2022) who conducted a 7-day trial using a Bluetooth-enabled card for contact tracing and found that only 65% of self-reported contact events were detected by cards (not unlike our 61%) which they attributed to decreasing compliance during the course of the (short relative to ours) protocol. Modifying procedures to ensure peer compliance or, even better, utilizing a system that uses phone-to-phone transmission rather than beacon-to-phone transmission should lower false negatives by lowering the likelihood of noncompliance given that most individuals consistently keep their phone nearby.

A second (but we think less likely) explanation for the false negatives is that the participant misreported - that is, the participant indicated that the peer was present when they were not or inadvertently indicated the wrong response on the report. Finally, it is possible that failure of technology accounted for false negatives including that the beacon settings were not sufficient for the app to detect the beacon presence or the physical setting interfered with detection. On the former point, the beacon RSSI settings were extensively tested and consistently demonstrated that they were sufficient for detection by phone Bluetooth. On the latter point, it is possible that physical conditions could interfere with beacon detection, including for example, if the participant was aware that a peer was physically nearby (e.g., in a different room) but the signal was not sufficiently strong because of a physical barrier (e.g., a wall).

Commensurate with our low false-positive rate, our true-negative rate was high (94%). That is, there was a high proportion of cases where participants denied peer presence and the beacon was not detected. Indeed, the majority of all the reports used for the validity evaluation (2,272/3,354; 68%) fell in this cell. This indicates that using passive detection of specific peers would reduce unnecessary alerts in research designed to understand the influence of peer contact, and specifically just-in-time interventions that involve a friend (e.g., detecting contact with a drinking buddy).

Limitations

Eligibility criteria were adjusted so we were unable to evaluate the eligibility rate of respondents. Due to human subjects’ protections requirements, participants were responsible for contacting and inviting peers, so we do not know how many peers were invited and were unable to document the peer enrollment rate. Although we provided supplies to facilitate carrying, it appears some peers did not always have the beacon with them. Phone-to-phone detection, wherein the Bluetooth transmitter in one phone is identified and detected by another is the obvious solution to this problem; this approach would involve both participants and peers installing the application, with the additional benefit that both phones would record the others’ presence (in our study only participant phones recorded peer beacon presence, i.e., beacon direction was unidirectional). The scope of this project and its limited funds also precluded us from building the app for iOS which tends to have more restrictions on app developers; other recent research has also only used Android phones (Girolami et al., 2020). There was a wide range in the number of peer encounters, with some participants having none. We are unable to determine if this was due to app function issues or to peer noncompliance. While we believe our procedures could be utilized with other populations, it is possible that findings would not generalize beyond young adults.

Future Directions

This proof-of-concept investigation indicates that the fundamental elements of detecting influential peer proximity can work under optimal circumstances. Knowledge derived from contact tracing and exposure notification systems developed during the COVID-19 pandemic (Chambers et al., 2024; Chen et al., 2021; Leith & Farrell, 2020) in which Bluetooth Low Energy was the most commonly used technology (Quach et al., 2025), can inform future work, which should include improving the functionality of the technology for use with different smartphone operating systems and their embedded Bluetooth technology (Meijerink et al., 2021). Utilizing passive detection of known peers for the purpose of understanding behavioral influence in situations in in which the social context is relevant could include other smartphone sensors, including geolocation/geofencing to detect physical risk locations (Bae et al., 2017; Meegahapola et al., 2021), and wearables to detect target behaviors (e.g., alcohol use; Brobbin et al., 2022), but must also grapple with issues related to data protection and privacy concerns (Bradford et al., 2020). Finally, combining detection of socially risky situations with a notification system in which users are made aware and provided guidance or intervention (Nahum-Shani et al., 2018) are valuable next steps for behavioral health research.

Public Significance Statement:

The development of passive sensing technology has implications for real-world interventions that could interrupt behavioral risk associated with interpersonal contact as well as other circumstances in which social mixing occurs such as infectious disease transmission.

Disclosures and Acknowledgements

This research was supported in part by grant numbers R21AA027329 (MPI Barnett and Jackson), K01AA025994 (PI Meisel), and K08DA048137 (PI Sokolovsky). NIH had no role in the study design, collection, analysis, or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication.

All authors have contributed to the manuscript and have read and approved this submission. The findings in this manuscript were presented in part at the 2025 meeting of the Society for Ambulatory Assessment in Leuven, Belgium, but have not been previously published.

Authors have no conflicts of interest.

Footnotes

1

We initially required non-solitary alcohol consumption on 2+ days per week, and at least one heavy drinking occasion (4+ for women, 5+ for men) per week in the past month. We experienced delays with the technology development, so in the interest of evaluating our primary goals, after our first 5 participants, we removed the alcohol use inclusion criteria. We do not think this change affects the goals of the research presented here, which is solely to evaluate the technology.

2

This procedure allowed us to make use of participant reports of any peer presence on any report and to then look for beacon detection at that approximate time. This accommodates the fact that more than one peer with a beacon can be present at the same time and reported by the participant, even if all signal-contingent reports are not answered.

3

More than one beacon-triggered report could be launched and available at a time (if more than one peer overlapped in their encounter with the participant). If a peer’s presence triggered a report that was not answered, but their presence was indicated by the participant on a different peer’s beacon-triggered report, we could still “match” the report of the peer’s presence with their beacon encounter using our time window definitions. This way, the report of a peer’s presence could be matched with the peer’s beacon encounter even if the beacon-triggered report associated with their encounter was not answered.

References

  1. Admiraal R, Millen J, Patel A, & Chambers T (2022). A case study of Bluetooth technology as a supplemental tool in contact tracing. Journal of Healthcare Informatics Research, 6, 208–227. 10.1007/s41666-021-00112-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bae S, Ferreira D, Suffoletto B, Puyana JC, Kurtz R, Chung T, & Dey AK (2017). Detecting drinking episodes in young adults using smartphone-based sensors. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(2), 1–36. 10.1145/3090051 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bangor A, Kortum P, & Miller JA (2008). The System Usability Scale (SUS): An empirical evaluation. International Journal of Human-Computer Interaction, 24, 574–594. 10.1080/10447310802205776 [DOI] [Google Scholar]
  4. Barnett NP, Clark MA, Kenney SR, DiGuiseppi G, Meisel MK, Balestrieri S, Ott MQ, & Light J (2019). Enrollment and assessment of a first-year college class social network for a controlled trial of the indirect effect of a brief motivational intervention. Contemporary Clinical Trials, 76, 16–23. 10.1016/j.cct.2018.10.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Barnett NP, Sokolovsky AW, Meisel MK, Forkus SR, & Jackson KM (2024). A Bluetooth-based smartphone app for detecting peer proximity: Protocol for evaluating functionality and validity. JMIR Research Protocols, 13, e50241. 10.2196/50241 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Baronti P, Barsocchi P, Chessa S, Crivello A, Girolami M, Mavilia F, & Palumbo F (2020). Remote detection of social interactions in indoor environments through bluetooth low energy beacons. Journal of Ambient Intelligence and Smart Environments, 12(3), 203–217. 10.3233/ais-200560 [DOI] [Google Scholar]
  7. Beck KH, Caldeira KM, Vincent KB, & Arria AM (2013). Social contexts of drinking and subsequent alcohol use disorder among college students. American Journal of Drug and Alcohol Abuse, 39(1), 38–43. 10.3109/00952990.2012.694519 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Boukhechba M, Cai L, Chow PI, Fua K, Gerber MS, Teachman BA, & Barnes LE (2018). Contextual analysis to understand compliance with Smartphone-based ecological momentary assessment Proceedings of the 12th EAI International Conference on Pervasive Computing Technologies for Healthcare, New York, NY, USA. 10.1145/3240925.3240967 [DOI] [Google Scholar]
  9. Bradford L, Aboy M, & Liddell K (2020). COVID-19 contact tracing apps: a stress test for privacy, the GDPR, and data protection regimes. Journal of Law and the Biosciences, 7. 10.1093/jlb/lsaa034 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Brobbin E, Deluca P, Hemrage S, & Drummond C (2022). Accuracy of wearable transdermal alcohol sensors: Systematic review. Journal of Medical Internet Research, 24(4), e35178. 10.2196/35178 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chambers T, Anglemyer A, Chen A, Atkinson J, Elers P, & Baker MG (2024). An evaluation of the population uptake and contact tracer utilisation of the Covid-19 Bluetooth Exposure Notification Framework in New Zealand. Australian and New Zealand Journal of Public Health, 48(6), 100197. 10.1016/j.anzjph.2024.100197 [DOI] [PubMed] [Google Scholar]
  12. Chen T, Baseman J, Lober W, Revere D, Klemfuss N, Hills R, & Karras B (2021). WA Notify: The planning and implementation of a Bluetooth exposure notification tool for COVID-19 pandemic response in Washington state. Online Journal of Public Health Informatics 13(1). 10.5210/ojphi.v13i1.11694 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Cheong SHR, Ng YJX, Lau Y, & Lau S (2022). Wearable technology for early detection of COVID-19: A systematic scoping review. Preventive Medicine, 162, 107170–107170. 10.1016/j.ypmed.2022.107170 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Del Boca FK, & Darkes J (2003). The validity of self-reports of alcohol consumption: State of the science and challenges for research. Addiction, 98(Suppl.2), 1–12. 10.1046/j.1359-6357.2003.00586.x [DOI] [PubMed] [Google Scholar]
  15. Doherty C, Baldwin M, Keogh A, Caulfield B, & Argent R (2024). Keeping pace with wearables: A living umbrella review of systematic reviews evaluating the accuracy of consumer wearable technologies in health measurement. Sports Medicine, 54(11), 2907–2926. 10.1007/s40279-024-02077-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Eagle N, & Pentland A (2006). Reality mining: Sensing complex social systems. Personal and Ubiquitous Computing, 10(4), 255–268. 10.1007/s00779-005-0046-3 [DOI] [Google Scholar]
  17. Elder A, Arrouzet C, Miljacic L, Karras B, Higgins A, West L, Lorigan D, Revere D, Polissar N, Segal C, Lober W, & Baseman J (2024). Evaluation of the effectiveness of Washington State’s digital COVID-19 exposure notification system over one pandemic year. Frontiers in Public Health, 12. 10.3389/fpubh.2024.1408178 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Girolami M, Mavilia F, & Delmastro F (2020). Sensing social interactions through BLE beacons and commercial mobile devices. Pervasive and Mobile Computing, 67, 101198. 10.1016/j.pmcj.2020.101198 [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Harari GM, Müller SR, Aung MSH, & Rentfrow PJ (2017). Smartphone sensing methods for studying behavior in everyday life. Current Opinion in Behavioral Sciences, 18, 83–90. 10.1016/j.cobeha.2017.07.018 [DOI] [Google Scholar]
  20. Hatke GF, Montanari M, Appadwedula S, Wentz M, Meklenburg J, Ivers L, Watson J, & Fiore P (2020). Using Bluetooth Low Energy (BLE) signal strength estimation to facilitate contact tracing for COVID-19. arXiv preprint arXiv:2006.15711. 10.48550/arXiv.2006.15711 [DOI] [Google Scholar]
  21. Hernández-Heredia TK, Reyes-Manzano C, Flores-Hernández D, Ramos-Fernández G, & Guzmán-Vargas L (2024). Proximity sensor for measuring social interaction in a school environment. Sensors, 24(15), 4822. 10.3390/s24154822 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Hoeben EM, Meldrum RC, Walker DA, & Young JTN (2016). The role of peer delinquency and unstructured socializing in explaining delinquency and substance use: A state-of-the-art review. Journal of Criminal Justice, 47, 108–122. 10.1016/j.jcrimjus.2016.08.001 [DOI] [Google Scholar]
  23. Kovalchik SA, Martino SC, Collins RL, Shadel WG, D’Amico EJ, & Becker K (2018). Scaled inverse probability weighting: A method to assess potential bias due to event nonreporting in ecological momentary assessment studies. Journal of Educational and Behavioral Statistics, 43(3), 354–381. 10.3102/1076998617738241 [DOI] [Google Scholar]
  24. Leith D, & Farrell S (2020). Measurement-based evaluation of Google/Apple Exposure Notification API for proximity detection in a light-rail tram. PLOS One, 15. 10.1371/journal.pone.0239943 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Li D, Shelby T, Brault M, Manohar R, Vermund S, Hagaman A, Forastiere L, Caruthers T, Egger E, Wang Y, Manohar N, Manohar P, Davis JL, & Zhou X (2023). Implementation of a hardware-assisted Bluetooth-based COVID-19 tracking device in a high school: Mixed methods study. JMIR Formative Research, 7, e39765–e39765. 10.2196/39765 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Longabaugh R, & Zywiak WH (2002). Project COMBINE: A manual for the administration of the Important People Instrument. Brown University. [Google Scholar]
  27. Maghdid HS, & Ghafoor KZ (2020). A smartphone enabled approach to manage COVID-19 lockdown and economic crisis. SN Computer Science, 1(5), 271. 10.1007/s42979-020-00290-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Maharjan S, Poudyal A, Van Heerden A, Byanjankar P, Thapa A, Islam C, Kohrt B, & Hagaman A (2020). Passive sensing on mobile devices to improve mental health services with adolescent and young mothers in low-resource settings: the role of families in feasibility and acceptability. BMC Medical Informatics and Decision Making, 21. 10.1186/s12911-021-01473-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Meegahapola L, Labhart F, Phan TT, & Gatica-Perez D (2021). Examining the social context of alcohol drinking in young adults with smartphone sensing. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 5(3), 1–26. 10.1145/3478126 [DOI] [Google Scholar]
  30. Meijerink H, Mauroy C, Johansen M, Braaten SM, Lunde CUS, Arnesen T, Feruglio S, Nygård K, & Madslien E (2021). The first GAEN-based COVID-19 contact tracing app in Norway identifies 80% of close contacts in “real life” scenarios. Frontiers in Digital Health, 3, 731098. 10.3389/fdgth.2021.731098 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Meisel MK, Kenney SR, & Barnett NP (2018). Characteristics of prospectively identified negative alcohol-related events among college students. Addictive Behaviors, 78, 138–144. 10.1016/j.addbeh.2017.11.025 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Nahum-Shani I, Smith SN, Spring BJ, Collins LM, Witkiewitz K, Tewari A, & Murphy SA (2018). Just-in-Time Adaptive Interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine, 52(6), 446–462. 10.1007/s12160-016-9830-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. PiLR Health. Ecological Momentary Assessment Made Simple. Retrieved 2024-03-16 from https://pilrhealth.com/ [Google Scholar]
  34. Quach H-L, Walsh EI, Hoang TNA, Terrett RNL, & Vogt F (2025). Effectiveness of digital contact tracing interventions for COVID-19: A systematic scoping review. Public Health, 242, 146–156. 10.1016/j.puhe.2025.02.019 [DOI] [PubMed] [Google Scholar]
  35. Reblin M, McCormick R, Mansfield KJ, Wawrzynski SE, Ketcher D, Tennant KE, Guo J-W, Jones EC, & Cloyes KG (2022). Feasibility, usability, and acceptability of personalized web-based assessment of social network and daily social support interactions over time. Journal of Cancer Survivorship: Research and Practice, 16(4), 904–912. 10.1007/s11764-021-01083-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Rosenberg M, Kianersi S, Luetke M, Jozkowski K, Guerra-Reyes L, Shih PC, Finn P, & Ludema C (2023). Wearable alcohol monitors for alcohol use data collection among college students: Feasibility and acceptability. Alcohol, 111, 75–83. 10.1016/j.alcohol.2023.05.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Santani D, Do T-M-T, Labhart F, Landolt S, Kuntsche E, & Gatica-Perez D (2018). DrinkSense: Characterizing youth drinking behavior using smartphones. IEEE Transactions on Mobile Computing, 17(10), 2279–2292. 10.1109/tmc.2018.2797901 [DOI] [Google Scholar]
  38. Shelby T, Caruthers T, Kanner OY, Schneider R, Lipnickas D, Grau LE, Manohar R, & Niccolai L (2021). Pilot evaluations of two Bluetooth contact tracing approaches on a university campus: Mixed methods study. JMIR Formative Research, 5(10), e31086. 10.2196/31086 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Shiffman S, Stone AA, & Hufford MR (2008). Ecological momentary assessment. Annual Review of Clinical Psychology, 4(1), 1–32. 10.1146/annurev.clinpsy.3.022806.091415 [DOI] [PubMed] [Google Scholar]
  40. Sokolovsky AW, Mermelstein RJ, & Hedeker D (2014). Factors predicting compliance to ecological momentary assessment among adolescent smokers. Nicotine & Tobacco Research, 16(3), 351–358. 10.1093/ntr/ntt154 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Stone AA, Schneider S, & Smyth JM (2023). Evaluation of pressing issues in ecological momentary assessment. Annual Review of Clinical Psychology, 19(1), 107–131. 10.1146/annurev-clinpsy-080921-083128 [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES