[Skip to Content]
Access to paid content on this site is currently suspended due to excessive activity being detected from your IP address 34.204.191.31. Please contact the publisher to request reinstatement.
[Skip to Content Landing]
Table.  
Model Precision (Positive Predictive Value) and Recall (Sensitivity) Based on Amount of Waze Data Used, California, 2018
Model Precision (Positive Predictive Value) and Recall (Sensitivity) Based on Amount of Waze Data Used, California, 2018
1.
WISQARS (Web-based Injury Statistics Query and Reporting System) Injury Center, Centers for Disease Control and Prevention. https://www.cdc.gov/injury/wisqars/index.html. Published August 1, 2017. Accessed October 31, 2017.
2.
National Center for Statistics and Analysis.  2016 Fatal Motor Vehicle Crashes: Overview. Washington, DC: National Highway Traffic Safety Administration; 2017.
3.
Plevin  RE, Kaufman  R, Fraade-Blanar  L, Bulger  EM.  Evaluating the potential benefits of advanced automatic crash notification.  Prehosp Disaster Med. 2017;32(2):156-164. doi:10.1017/S1049023X16001473PubMedGoogle ScholarCrossref
4.
Mell  HK, Mumma  SN, Hiestand  B, Carr  BG, Holland  T, Stopyra  J.  Emergency medical services response times in rural, suburban, and urban areas.  JAMA Surg. 2017;152(10):983-984. doi:10.1001/jamasurg.2017.2230PubMedGoogle ScholarCrossref
5.
Rumsfeld  JS, Brooks  SC, Aufderheide  TP,  et al; American Heart Association Emergency Cardiovascular Care Committee; Council on Cardiopulmonary, Critical Care, Perioperative and Resuscitation; Council on Quality of Care and Outcomes Research; Council on Cardiovascular and Stroke Nursing; and Council on Epidemiology and Prevention.  Use of mobile devices, social media, and crowdsourcing as digital strategies to improve emergency cardiovascular care: a scientific statement from the American Heart Association.  Circulation. 2016;134(8):e87-e108. doi:10.1161/CIR.0000000000000428PubMedGoogle ScholarCrossref
6.
Young  SD, Torrone  EA, Urata  J, Aral  SO.  Using search engine data as a tool to predict syphilis.  Epidemiology. 2018;29(4):574-578. doi:10.1097/EDE.0000000000000836PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Research Letter
    May 22, 2019

    Crowdsourced Traffic Data as an Emerging Tool to Monitor Car Crashes

    Author Affiliations
    • 1Department of Family Medicine, University of California, Los Angeles
    • 2Department of Emergency Medicine, University of California, Irvine, School of Medicine, Irvine
    • 3University of California Institute for Prediction Technology, Department of Informatics, Donald Bren School of Information and Computer Sciences, University of California, Irvine
    • 4Department of Computer Science, Henry Samueli School of Engineering, University of California, Los Angeles
    JAMA Surg. 2019;154(8):777-778. doi:10.1001/jamasurg.2019.1167

    More than 100 deaths daily and 2.5 million emergency department visits each year result from motor vehicle crashes, making motor vehicle crashes a leading cause of death in the United States.1,2 Reducing ambulance and emergency department treatment response time for crash patients could markedly save lives.3 We evaluated whether data from the crowdsourced traffic application Waze (Google) might be used to detect motor vehicle crashes faster than current reporting methods.

    Methods

    We obtained data on California motor vehicle incident reports (eg, crashes, road hazards, and weather conditions) from June 12, 2018, to August 1, 2018, from the California Highway Patrol (CHP) (https://media.chp.ca.gov/) and the crowdsourced traffic reporting application Waze (https://www.waze.com/). Our analysis focused on the 406 559 Waze user incident reports and the 7776 CHP crash reports (of 18 582 total CHP reports). The University of California, Los Angeles Institutional Review Board waived this study because data are anonymous and aggregated. A support vector machine learning model was created based on a subset of the data (ie, training phase) to identify Waze user−reported alerts corresponding to each CHP-reported crash (ie, crash alerts). The features of the model included report time, location, type of incident reported, and user confidence in the reports. Using 10-fold cross-validation, we assessed precision (positive predictive value) and recall (sensitivity) on the remaining subset of the data (ie, test phase) to determine how well the model estimated CHP-reported crashes. We report results on the mean lead time that Waze reports occurred before CHP reports for each crash broken down by the order of Waze alerts received (eg, first, second, third Waze reports). Analyses were conducted using Python software, version 2.7 (Python Software Foundation).

    Results

    Waze crash alerts occurred a mean of 2 minutes 41 seconds before their corresponding CHP-reported crash (precision, 0.86; recall, 0.87). Multiple crash alerts were often associated with the same CHP-reported crash. The earliest crash alert in each series occurred a mean of 4 minutes 3 seconds before the CHP report (precision, 0.87; recall, 0.88). Waiting for additional confirmatory crash alerts was associated with decreased advance warning time compared with CHP reports, with little improvement in precision and recall (Table).

    Discussion

    Emergency medical service units take a mean of 7 to 14 minutes to arrival on scene after a 911 call.4 Crowdsourced traffic data might help to decrease that time by approximately 20% to 60% by enabling novel, low-cost, and early identification of car crashes. These social data are highly associated with conventional reporting data that are often costly to collect and have reporting lag time.5,6

    The ability to use crowdsourced, user-generated traffic data has several immediate clinical implications for treatment and mortality rates among motor vehicle crash patients as well as for improving efficiency around emergency department operations in the United States. First, emergency medical service systems could use crowdsourced data tools to more efficiently mobilize resources and ambulances, especially for simultaneous collisions. Second, early crowdsourced crash data might be reported to trauma centers and hospitals to allow emergency departments to better prepare for injured patients. Trauma surgeons could be notified earlier, diagnostic testing could be prioritized for crash patients, and blood and life-saving equipment could potentially be made available sooner. These prehospital and hospital-level resources, if activated sooner, could aid in increasing quality and rapidity of patient care and potentially be associated with reduced morbidity and mortality.

    This pilot study is limited by data collection time (50 days), ability to validate Waze report accuracy, and ability to generalize to rural areas and regions outside California. However, the findings suggest the need for further research on integrating crowdsourced traffic data as a tool to monitor car crashes and reduce associated mortality, including exploring a longer period, crash severity and rural or urban areas as potential moderators, and potential risks or harms of implementing this approach.

    Back to top
    Article Information

    Accepted for Publication: March 3, 2019.

    Published Online: May 22, 2019. doi:10.1001/jamasurg.2019.1167

    Correction: This article was corrected on June 26, 2019, to add an affiliation for Sean D. Young, PhD, and to fix an error in the first sentence.

    Corresponding Author: Sean D. Young, PhD, University of California Institute for Prediction Technology, Department of Informatics, 5019 Donald Bren Hall, Irvine, CA 92697-3440 (sean.y@uci.edu).

    Author Contributions: Drs Young and Wang had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Young, Wang.

    Acquisition, analysis, or interpretation of data: All authors.

    Drafting of the manuscript: Young.

    Critical revision of the manuscript for important intellectual content: All authors.

    Statistical analysis: Wang.

    Obtained funding: Young.

    Administrative, technical, or material support: Chakravarthy.

    Supervision: Young, Wang.

    Conflict of Interest Disclosures: Dr Young reported receiving grants from the National Institutes of Health during the conduct of the study and gift funding to the University of California Institute for Prediction Technology from Facebook and Intel. No other disclosures were reported.

    Funding/Support: This study was funded by the National Institute of Allergy and Infectious Diseases and the National Human Genome Research Institute.

    Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    Additional Contributions: The chief technology officer, Jay Song, of the California Highway Patrol and Adam Fried of Waze (Google) assisted in providing data and data support; David Bychkov, PhD, assisted with data collection and partnerships; and Cheng Zheng assisted with data analysis.

    References
    1.
    WISQARS (Web-based Injury Statistics Query and Reporting System) Injury Center, Centers for Disease Control and Prevention. https://www.cdc.gov/injury/wisqars/index.html. Published August 1, 2017. Accessed October 31, 2017.
    2.
    National Center for Statistics and Analysis.  2016 Fatal Motor Vehicle Crashes: Overview. Washington, DC: National Highway Traffic Safety Administration; 2017.
    3.
    Plevin  RE, Kaufman  R, Fraade-Blanar  L, Bulger  EM.  Evaluating the potential benefits of advanced automatic crash notification.  Prehosp Disaster Med. 2017;32(2):156-164. doi:10.1017/S1049023X16001473PubMedGoogle ScholarCrossref
    4.
    Mell  HK, Mumma  SN, Hiestand  B, Carr  BG, Holland  T, Stopyra  J.  Emergency medical services response times in rural, suburban, and urban areas.  JAMA Surg. 2017;152(10):983-984. doi:10.1001/jamasurg.2017.2230PubMedGoogle ScholarCrossref
    5.
    Rumsfeld  JS, Brooks  SC, Aufderheide  TP,  et al; American Heart Association Emergency Cardiovascular Care Committee; Council on Cardiopulmonary, Critical Care, Perioperative and Resuscitation; Council on Quality of Care and Outcomes Research; Council on Cardiovascular and Stroke Nursing; and Council on Epidemiology and Prevention.  Use of mobile devices, social media, and crowdsourcing as digital strategies to improve emergency cardiovascular care: a scientific statement from the American Heart Association.  Circulation. 2016;134(8):e87-e108. doi:10.1161/CIR.0000000000000428PubMedGoogle ScholarCrossref
    6.
    Young  SD, Torrone  EA, Urata  J, Aral  SO.  Using search engine data as a tool to predict syphilis.  Epidemiology. 2018;29(4):574-578. doi:10.1097/EDE.0000000000000836PubMedGoogle ScholarCrossref
    ×