Abstract
Introduction:
Wearable biosensors have the potential to monitor physiological change associated with opioid overdose among people who use drugs.
Methods:
We enrolled 16 individuals who reported ≥4 daily opioid use events within the previous 30 day. Each was assigned a wearable biosensor that measured respiratory rate (RR) and actigraphy every 15 seconds for 5 days and also completed a daily interview assessing drug use. We describe the volume of RR data collected, how it varied by participant characteristics and drug use over time using repeated measures one-way ANOVA, episodes of acute respiratory depression (≤5 breaths/minute), and self-reported overdose experiences.
Results:
We captured 1626.4 hours of RR data, an average of 21.7 daily hours/participant over follow-up. Individuals with longer injection careers and those engaging in polydrug use captured significantly fewer total hours of respiratory data over follow-up compared to those with shorter injections careers (94.7 vs. 119.9 hours, p = 0.04) and injecting fentanyl exclusively (98.7 vs. 119.5 hours, p = 0.008), respectively. There were 385 drug use events reported over follow-up. There were no episodes of acute respiratory depression which corresponded with participant reports of overdose experiences.
Discussion:
Our preliminary findings suggest that using a wearable biosensor to monitor physiological changes associated with opioid use was feasible. However, more sensitive biosensors that facilitate triangulation of multiple physiological data points and larger studies of longer duration are needed.
Keywords: people who use drugs, wearable sensor, biosensor
1. INTRODUCTION
The opioid-related overdose epidemic is one of the most significant public health crises confronting the United States. Opioids accounted for 75% of the 90,000 overdose deaths from November 2019 – October 2020 (Ahmad et al., 2021; Center for Disease Control and Prevention, 2020). Interventions targeting this crisis have included overdose education and naloxone distribution programs (Walley et al., 2013), fentanyl test strips (Peiper et al., 2019), warm handoff programs that utilize peers to link recent overdose victims to drug treatment (Ashford et al., 2019), a smartphone-based application that link lay first responders to an overdose event (Schwartz et al., 2020), and remote supervised consumption spaces that connect people who use drugs (PWUD) to peer monitors by phone (Jean and Bonn, 2020; Perri et al., 2021). A common limitation among these interventions is that most required active engagement by the person who uses drugs (e.g., drug checking or requesting remote supervision) or from a nearby peer/responder (e.g., naloxone administration). Given more than half of fatal overdoses occur when individuals use alone (Sherman et al., 2007; Siegler et al., 2014), novel interventions designed specifically for these events are needed.
Respiratory depression, slow heart rate, and physical inactivity are hallmark physiological changes associated with opioid overdose (Boyer, 2012; Chen and Ashburn, 2015). In clinical settings, wearable biosensors such as pulse oximeters or respiratory monitors are the gold standard for monitoring the physiologic impact of opioids. However, there is a dearth of information on whether wearable devices with remote sensing technology can reliably monitor these important opioid-related outcomes as individuals go about their daily lives. In this proof-of-concept study, we assessed whether a commercially available wearable biosensor could feasibly capture physiological changes associated with opioid overdose in a real-world setting.
2. METHODS
2.1. Setting and Participants
Data were collected from December 2019 – March 2020 in Philadelphia. Participants were recruited from UnityPhilly (Schwartz et al., 2020), a longitudinal trial of a smartphone-based opioid overdose intervention set in the neighborhood with the highest number of non-fatal and fatal overdoses and naloxone administrations in the city (Philadelphia Department of Public Health, 2020). Eligibility criteria for UnityPhilly were age ≥18 years, spending majority of your day in ZIP codes with high overdose rates, being able to read/write in English, owning a smartphone with an active data plan, and allowing the study team to access geolocation data (GPS coordinates). We used a two-step process to recruit participants from UnityPhilly into the present study: research staff contacted UnityPhilly participants by phone or in-person who reported injection drug use within 30 days at their baseline (n=59); staff sequentially enrolled participants who met the following inclusion criteria: (1) snorting or injecting opioids at least four times daily for the past seven days, (2) willingness to wear a biosensor for 5 days, and (3) willingness to return to the study office each day of follow-up. Sixteen participants provided written informed consent and were enrolled in this pilot study.
2.2. Biosensor Provision and Collection of Physiological Outcomes
The research team reviewed 6 potential devices (Fuller et al., 2020) which included the Wavelet Health/Biostrap, Fitbit Charge 3, AutoSense, Oxitone, a smartphone sonar system (Nandakumar et al., 2019), and the Spire® Health Tag (Spire® Health, https://www.spirehealth.com/). We selected the Spire® Health Tag based on various factors: the one-year battery life, discrete size (5.6 cm x 1.3 cm x 0.5 cm), and unobtrusive placement (attached to underwear waistband for self-identified men or wing of bra for self-identified women). Participants were asked to wear the device consistently over the 5-day period, only removing it for the purpose of bathing. While the Spire® is active and touching the wearer’s skin, it collects respiratory rate (RR) and physical movement, which is binned every 15 seconds with a 60 second sliding window, using a force sensor and 3-axis accelerometer, respectively, as well as heart rate every 4 minutes using a photoplethysmography sensor. After collection, the Spire® transmits data via Bluetooth to a secure cloud-based server and activity is recoded to active (>1 step within a 60 second bin) versus sedentary (0 steps within a 60 second bin). However, there were concerns that participants may not regularly have secure internet access and were asked to return to the study office daily for device placement verification as well as data download and transfer to a secure cloud-based storage.
2.3. Self-reported drug use and intoxication
Participants completed a daily timeline follow back (TLFB) interview, a reliable method for capturing data on substance use in vulnerable populations (Hjorthøj et al., 2012). TLFB items had a 24-hour recall period and measured: number of drug events, time of each event, drug(s) consumed at each event (e.g., fentanyl, heroin, powder cocaine, crack cocaine, methamphetamine, and benzodiazepines), route of administration for each drug consumed (e.g., injected, smoked, sniffed/snorted, and swallowed), dose (e.g., number of bags or pills), and resultant level of intoxication (e.g., too little, just right, and too much). For opioid-related “too much” events, we measured whether participants received any of the following interventions (yes/no): physical stimulation, naloxone, or had EMS called.
2.4. Measures and Statistical Analyses
2.4.1. Participant Characteristics
Over the study period, one participant was lost-to-follow-up (after third visit). Using R studio version 4.0.2 (R Core Team, 2020), we compared their pattern of drug use to those who completed the study. Their drug use pattern (i.e., primarily snorted heroin) differed from the remaining sample (i.e., primarily fentanyl injection); thus, we decided to remove them from the present analyses. We then summarized key participant characteristics (i.e., age, gender, race/ethnicity, homelessness in past 30 days, years using injection drugs, lifetime overdoses, and perceived risk of overdose within the next year) drawn from the UnityPhilly baseline survey.
2.4.2. Biosensor Data
To assess feasibility, we downloaded data in a csv format file (one file per participant) that contained information about the RR measured every 15 seconds. To assess feasibility, we summed the 15-second bins and calculated the total volume of RR data captured at the group-level over follow-up (1800 possible hours from 15 participants x 5 days x 24 hours), as well as daily average hours and 5-day total at the participant-level. We defined respiratory depression as ≤5 breaths/minute for 60 seconds as recorded by the Spire® (Nandakumar et al., 2019). Respiratory depression was computed over a sliding (moving) window 60 seconds, around the time of each reported drug use event (see Supplemental Materials for additional information on data management and visualization of RR and physical movement). We used repeated measures one-way ANOVA to assess for differences in the volume of RR data collected by participant characteristics and drug use pattern to explore emerging trends that may help design future studies.
2.4.3. Drug Use and Self-Reported Overdose
We calculated the total number of drug use events by route of administration and percentage of polydrug use (i.e., multiple drugs by various routes; yes/no) at the participant-level over follow-up among the final sample (n=15). We summarized self-reported intoxication across drug use events and receipt of intervention (physical stimulation naloxone, or had EMS called) during events participants described as “too much”. For each “too much” event, we analyzed whether participants were active or sedentary or if RR fell below the predefined threshold for acute respiratory depression, based upon self-reported time of consumption from the TLFB.
2.5. Compensation and Ethics Approval
Participants received $10 for the initial enrollment and training visit, $20 for each completed follow-up visit, and another $20 for completing qualitative interviews during their 5th visit, a maximum total of $130 USD for the entire study. Study protocol was approved by the Drexel University Institutional Review Board.
3. RESULTS
3.1. Participant Characteristics
Most participants identified as non-Hispanic White (93%) and male (53%) with mean age of 42 (standard deviation [SD]=7.1; Table 1). The average length of drug injection career was 19 years (SD=7.5). The majority (87%) had experienced an opioid-related overdose in their lifetime and 27% believed they were somewhat or very likely to overdose again within the next year.
Table 1.
Volume of respiratory rate data collected by Spire® Health Tag, a commercially available wearable biosensor, by sociodemographic, drug use, and overdose characteristics (n=15)
| Number of RR Hours Captured by Biosensor Daily Mean (SD) | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| n | Day 1 | Day 2 | Day 3 | Day 4 | Day 5 | 5-Day Total (SD)a | F valueb | P | |
| Total sample | 15 | 22.2 (5.7) | 23.9 (0.1) | 19.5 (9.1) | 20.6 (8.4) | 22.2 (6.2) | 108.4 (15.6) | ||
| Age, years | 0.65 | 0.54 | |||||||
| 30 – 36 | 5 | 19.3 (9.8) | 23.9 (0.1) | 24.0 (0.1) | 23.8 (0.5) | 24.0 (0.1) | 115.0 (9.0) | ||
| 37 – 45 | 5 | 23.9 (0.3) | 23.9 (0.1) | 15.7 (11.5) | 19.2 (10.7) | 23.6 (0.7) | 106.2 (12.7) | ||
| 46 – 54 | 5 | 23.4 (1.2) | 24.0 (0.1) | 18.9 (10.6) | 18.8 (10.5) | 19.1 (10.7) | 104.1 (22.6) | ||
| Sex | 1.56 | 0.23 | |||||||
| Male | 8 | 20.8 (7.7) | 23.9 (0.1) | 20.8 (8.3) | 17.7 (10.9) | 20.7 (8.4) | 103.8 (18.6) | ||
| Female | 7 | 23.8 (0.5) | 24.0 (0.1) | 18.1 (10.3) | 23.9 (0.4) | 24.0 (0.1) | 113.7 (10.1) | ||
| Experiencing homelessness, 30 days | 0.07 | 0.77 | |||||||
| No | 10 | 23.7 (0.8) | 24.0 (0.1) | 17.3 (10.6) | 21.4 (7.5) | 21.4 (7.5) | 107.6 (17.6) | ||
| Yes | 5 | 19.2 (9.8) | 23.9 (0.1) | 24.0 (0.1) | 18.9 (10.6) | 23.9 (0.1) | 110.0 (12.3) | ||
| Length of injection career, years | 4.18 | 0.04 | |||||||
| 9 – 14 | 5 | 23.9 (0.2) | 24.0 (0.1) | 24.0 (0.1) | 24.0 (0.1) | 24.0 (0) | 119.9 (0.4) | ||
| 15 – 24 | 6 | 20.0 (8.9) | 23.9 (0.1) | 17.1 (10.9) | 23.5 (0.8) | 23.7 (0.6) | 108.1 (11.7) | ||
| 25 – 31 | 4 | 23.3 (1.3) | 23.9 (0.1) | 17.5 (11.7) | 12.0 (13.9) | 17.9 (11.9) | 94.7 (21.0) | ||
| Number opioid overdose(s), lifetime | 0.89 | 0.38 | |||||||
| 0 | 2 | 23.8 (0.3) | 23.8 (0.2) | 23.8 (0.1) | 23.9 (0.2) | 23.9 (0.2) | 119.2 (0.1) | ||
| 1 – 2 | 6 | 20.1 (9.0) | 24.0 (0.1) | 20.0 (9.8) | 23.5 (0.8) | 23.9 (0.2) | 111.4 (11.5) | ||
| 3 – 4 | 3 | 22.9 (1.4) | 24.0 (0.1) | 15.5 (13.3) | 15.9 (13.8) | 15.5 (13.4) | 93.8 (26.1) | ||
| 5 – 50 | 4 | 24.0 (0) | 24.0 (0.1) | 19.6 (8.8) | 18.0 (12.0) | 23.9 (0.1) | 109.5 (12.4) | ||
| Perceived risk of overdose within next year | 0.06 | 0.80 | |||||||
| Very/somewhat unlikely | 10 | 23.6 (0.9) | 24.0 (0) | 19.7 (8.8) | 18.9 (10.0) | 21.5 (7.6) | 107.7 (17.3) | ||
| Very/somewhat likely | 5 | 19.3 (9.8) | 23.8 (0.1) | 19.2 (10.6) | 23.9 (0.1) | 23.6 (0.7) | 109.8 (13.2) | ||
| Total daily injections at follow-up | 1.01 | 0.39 | |||||||
| 6 – 14 | 5 | 23.5 (1.2) | 23.9 (0.1) | 20.1 (7.7) | 14.4 (13.1) | 19.1 (10.7) | 101.0 (21.3) | ||
| 15 – 24 | 5 | 19.6 (9.9) | 23.9 (0.1) | 24.0 (0) | 23.6 (0.9) | 23.9 (0.2) | 115.0 (9.7) | ||
| 25 – 39 | 5 | 23.5 (0.6) | 24.0 (0.1) | 14.4 (13.1) | 23.7 (0.4) | 23.6 (0.7) | 109.2 (13.3) | ||
| Polydrug use over follow-upc | 11.7 | 0.008 | |||||||
| No | 7 | 23.9 (0.2) | 24.0 (0.1) | 24.0 (0.1) | 23.7 (0.8) | 23.9 (0.2) | 119.5 (0.9) | ||
| Yes | 8 | 20.6 (7.7) | 23.9 (0.1) | 15.6 (11.3) | 17.8 (11.0) | 20.7 (8.4) | 98.7 (16.0) | ||
SD = standard deviation; TLFB = timeline follow back.
For each participant, we calculated the sum of daily number of hours captured by biosensor and subsequently estimated the mean of the summed hours for all participants
Based on repeated measure one-way ANOVA
Polydrug use (Yes = reporting fentanyl, stimulants, benzodiazepines by various routes of administration over the course of a day vs. No = injection fentanyl only)
3.2. Biosensor Data
Out of 1800 possible hours, a total of 1626.4 hours of RR data were captured over follow-up which is equivalent to an average of 108.4 total hours (SD=15.6) or 21.7 daily hours (SD=3.1) per participant (Table 1). The volume of RR data collected was consistent over time, though participants with shorter injection careers had, on average, collected more total hours of data than those with longer injection careers (p = 0.04) as did individuals exclusively injecting fentanyl compared to those reporting polydrug use (p = 0.008). We did not detect any episodes of acute respiratory depression and thus, were unable to triangulate these data with movement.
3.3. Drug Use and Self-Reported Overdose
Via TFLB, we identified 385 drug administrations during the study period (Table 2), including 349 (91%) events involving fentanyl. This translates to an average of 4.7 (SD=2.8) daily fentanyl injection per participant. Seven participants (46%) reported only fentanyl injection during the study period while the rest (54%) reported at least some sequential or simultaneous polydrug use (Table 1). There were 27/385 administrations (6.8%) after which participants characterized the intoxication level as “too much”. Among those, 23/27 were related to fentanyl injection. There were no self-reports of experiencing an opioid overdose, receiving naloxone, nor having EMS called. While among 5/23 events, participants reported needing physical stimulation to wake up, lack of escalation to the preceding measures suggest the level of sedation did not cause an overdose.
Table 2.
Frequency of drug use and route of administration by participants, events, and average events reported per day collected from the timeline follow back interviews
| No. of Participants N (%) | No. of Events N (%)a | Average Daily Events (SD)b | |
|---|---|---|---|
| Injected fentanyl only | 15 (100) | 277 (71.9) | 3.7 (2.3) |
| Injected fentanyl and smoked crack cocaine, sequentially | 5 (33.3) | 55 (14.3) | 2.2 (2.4) |
| Injected fentanyl and powder cocaine, simultaneously | 2 (13.3) | 30 (7.9) | 3.0 (0.7) |
| Smoked crack cocaine only | 2 (13.3) | 17 (4.4) | 2.1 (1.9) |
| Injected powder cocaine only | 1 (6.7) | 1 (0.3) | -- |
| Injected methamphetamine only | 1 (6.7) | 4 (1.0) | 1.3 (0.6) |
| Injected fentanyl and heroin | 1 (6.7) | 1 (0.3) | -- |
| Smoked crack cocaine and used benzodiazepine, sequentially | 1 (6.7) | 1 (0.3) | -- |
Calculated based on the total number of drug use events over follow-up, excluding methadone (N = 385)
Number of events / (number of participants x 5 days of follow-up)
4. DISCUSSION
This proof-of-concept study is among the first to collect data on wearable biosensors that passively monitor physiological changes associated with opioid use in a real-world setting. Over the course of five days, the Spire® collected RR data for more than 90% of the total hours the device was assigned, suggesting that this may be a feasible manner of data collection among PWUD. This differs from prior work using the Spire® Stone (an older version of the device) among LinkedIn employees which indicated lower levels of engagement (Smith et al., 2019). Increased willingness to wear the Health Tag in our sample may be related to the device’s less obstructive and discrete design (Kanter et al., 2021), sampling participants from UnityPhilly, and the high level of community awareness of overdose risk in the Kensington neighborhood, which has been Philadelphia’s epicenter for the opioid-related overdose crises. Furthermore, we found that individuals with longer injection careers and those engaging in polydrug use collected fewer hours of RR data. Moving forward, individuals in these groups may benefit from additional training on the device to improve data collection.
While our findings provide preliminary data on the feasibility of using the Spire®, there were limitations to the study design that could provide guidance for future research. Despite setting our study within real-world conditions, we were unable to validate our results with a gold-standard measure (e.g., video-recorded direct observation to evaluate physical activity) (Keadle et al., 2019). For instance, 10% of RR data was not captured by the device; however, we were unable to identify the source of missing data given that we did not directly observe when or how the device was worn. User error (e.g., improper placement, removing the device to shower and forgetting to reapply) or device error (e.g., data not recorded) could be responsible for missing data. Future studies should consider periods where multiple devices are utilized to better characterize missingness.
Relatedly, we were limited in our ability to determine opioid-related overdoses. After reviewing changes in RR over time, we did not identify any events meeting our threshold for an acute respiratory depression nor were there any self-reports of an opioid overdose (i.e., no reports of receiving naloxone). It is possible that this may be related to a pattern of lower-frequency, low-dose injection that is associated with a protective effect (Colledge et al., 2020) since our sample reported injecting an average of 1.4 bag of fentanyl per event (data not shown), despite an average of roughly 5 fentanyl injections per day. In qualitative exit interviews, participants indicated low volume use was a conscious act of harm reduction and reported using just enough to “get well” or mitigate withdrawal (unpublished results). For these reasons, along with the short duration of the study, we did not capture any events meeting our definition of overdose and were unable to cross-reference RR with actigraphy data. Furthermore, we could not corroborate these results with heart rate given differences in the frequency in which it was captured over time. Future work might consider testing the wearable device to detect opioid-related overdoses in settings utilized by a large number of PWUD such as an overdose prevention facility.
5. CONCLUSION
Despite these limitations, this study provides important information about the potential role of wearable biosensors as a novel intervention to combat the overdose crisis. While most interventions require an active bystander, our data show that participants are willing to wear a device engaging in passive surveillance of physiological response to their opioid use which is promising. Critical next steps are to design and test a device more appropriate for detecting opioid-related overdose.
Supplementary Material
HIGHLIGHTS.
A wearable biosensor was used to monitor respiratory rate among PWUD.
Respiratory rate was captured for 1626.4 of 1800 possible follow-up hours.
No instances of acute respiratory depression were detected.
With refinement, biosensors could be used to intervene during solitary use events.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
CONFLICT OF INTEREST
No conflict declared.
CRediT authorship contribution statement
AMR conceptualized the project and designed the study. NKT conducted the analysis. AMR and NKT wrote the initial manuscript. All authors contributed to the interpretation of results, provided critical revisions to the manuscript, and approved the final version to be submitted for publication.
REFERENCES
- Ahmad FB, Rossen LM, Sutton P, 2021. Provisional drug overdose death counts. Hyattsville, MD. [Google Scholar]
- Ashford RD, Meeks M, Curtis B, Brown AM, 2019. Utilization of peer-based substance use disorder and recovery interventions in rural emergency departments: Patient characteristics and exploratory analysis. Journal of Rural Mental Health 43, 17–29. 10.1037/rmh0000106 [DOI] [Google Scholar]
- Boyer EW, 2012. Management of opioid analgesic overdose. New England Journal of Medicine 367, 146–155. 10.1056/NEJMRA1202561 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Center for Disease Control and Prevention, 2020. Overdose deaths accelerating during COVID-19 [WWW Document]. URL https://www.cdc.gov/media/releases/2020/p1218-overdose-deaths-covid-19.html (accessed 5.9.21).
- Chen A, Ashburn MA, 2015. Cardiac effects of opioid therapy. Pain Medicine 16, S27–S31. 10.1111/PME.12915 [DOI] [PubMed] [Google Scholar]
- Colledge S, Leung J, Larney S, Peacock A, Grebely J, Hickman M, Cunningham E, Trickey A, Stone J, Vickerman P, Degenhardt L, 2020. Frequency of injecting among people who inject drugs: A systematic review and meta-analysis. International Journal of Drug Policy. 10.1016/j.drugpo.2019.102619 [DOI] [PubMed] [Google Scholar]
- Fuller D, Colwell E, Low J, Orychock K, Tobin MA, Simango B, Buote R, Heerden D. van, Luan H, Cullen K, Slade L, Taylor NGA, 2020. Reliability and validity of commercially available wearable devices for measuring steps, energy expenditure, and heart rate: Systematic review. JMIR mHealth and uHealth 8. 10.2196/18694 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hjorthøj CR, Hjorthøj AR, Nordentoft M, 2012. Validity of timeline follow-back for self-reported use of cannabis and other illicit substances - Systematic review and meta-analysis. Addictive Behaviors. 10.1016/j.addbeh.2011.11.025 [DOI] [PubMed] [Google Scholar]
- Jean R, Bonn M, 2020. Drug-User Groups Innovate to Reduce the Risks of Using While Isolated [WWW Document]. Filter. URL https://filtermag.org/drugs-coronavirus-innovation/ (accessed 9.19.21).
- Kanter K, Gallagher R, Eweje F, Lee A, Gordon D, Landy S, Gasior J, Soto-Calderon H, Cronholm PF, Cocchiaro B, Weimer J, Roth A, Lankenau S, Brenner J, 2021. Willingness to use a wearable device capable of detecting and reversing overdose among people who use opioids in Philadelphia. Harm Reduction Journal 18, 1–14. 10.1186/S12954-021-00522-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Keadle SK, Lyden KA, Strath SJ, Staudenmayer JW, Freedson PS, 2019. A Framework to evaluate devices that assess physical behavior. Exercise and Sport Sciences Reviews 47, 206–214. 10.1249/JES.0000000000000206 [DOI] [PubMed] [Google Scholar]
- Nandakumar R, Gollakota S, Sunshine JE, 2019. Opioid overdose detection using smartphones. Science Translational Medicine 11. 10.1126/scitranslmed.aau8914 [DOI] [PubMed] [Google Scholar]
- Peiper NC, Clarke SD, Vincent LB, Ciccarone D, Kral AH, Zibbell JE, 2019. Fentanyl test strips as an opioid overdose prevention strategy: Findings from a syringe services program in the Southeastern United States. International Journal of Drug Policy 63, 122–128. 10.1016/j.drugpo.2018.08.007 [DOI] [PubMed] [Google Scholar]
- Perri M, Kaminski N, Bonn M, Kolla G, Guta A, Bayoumi AM, Challacombe L, Gagnon M, Touesnard N, McDougall P, Strike C, 2021. A qualitative study on overdose response in the era of COVID-19 and beyond: how to spot someone so they never have to use alone. Harm Reduction Journal 2021 18:1 18, 1–9. 10.1186/S12954-021-00530-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Philadelphia Department of Public Health, 2020. Unintentional Drug Overdose Fatalities in Philadelphia, 2019. CHART 5.
- R Core Team, 2020. R: A language and environment for statistical computing.
- Schwartz DG, Ataiants J, Roth A, Marcu G, Yahav I, Cocchiaro B, Khalemsky M, Lankenau S, 2020. Layperson reversal of opioid overdose supported by smartphone alert: A prospective observational cohort study. EClinicalMedicine 25. 10.1016/j.eclinm.2020.100474 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sherman SG, Cheng Y, Kral AH, 2007. Prevalence and correlates of opiate overdose among young injection drug users in a large U.S. city. Drug and Alcohol Dependence 88, 182–187. 10.1016/j.drugalcdep.2006.10.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- Siegler A, Tuazon E, Bradley O’Brien D, Paone D, 2014. Unintentional opioid overdose deaths in New York City, 2005–2010: A place-based approach to reduce risk. International Journal of Drug Policy 25, 569–574. 10.1016/j.drugpo.2013.10.015 [DOI] [PubMed] [Google Scholar]
- Smith EN, Santoro E, Moraveji N, Susi M, Crum AJ, 2019. Integrating wearables in stress management interventions: Promising evidence from a randomized trial. International Journal of Stress Management 27, 172–182. 10.1037/STR0000137 [DOI] [Google Scholar]
- Walley AY, Xuan Z, Hackman HH, Quinn E, Doe-Simkins M, Sorensen-Alawad A, Ruiz S, Ozonoff A, 2013. Opioid overdose rates and implementation of overdose education and nasal naloxone distribution in Massachusetts: Interrupted time series analysis. BMJ 346. 10.1136/bmj.f174 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.