Research JAMA Surgery | Original Investigation Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma Christopher R. Conner, MD, PhD; Hunter M. Ray, MD; Ryan M. McCormack, MD, PhD; Jacqueline S. Dickey, BSA; Samantha L. Parker, MD; Xu Zhang, PhD; Roberto M. Vera, MD; John A. Harvin, MD, MS; Ryan S. Kitagawa, MD Invited Commentary IMPORTANCE Motor vehicle crashes (MVCs) are an important public health concern. Supplemental content Recent trends suggest that introducing rideshare services has decreased the incidence of MVCs. However, detailed analyses linking rideshare volume, convictions for impaired driving, and nonfatal MVC traumas remain inconclusive. OBJECTIVE To determine if there is an association between rideshare use and MVC traumas and convictions for impaired driving in Houston, Texas. DESIGN, SETTING, AND PARTICIPANTS This multicenter cohort study was conducted between January 2007 and November 2019 with hospital data from the Red Duke Trauma Institute within the Memorial Hermann Hospital–Texas Medical Center and Ben Taub General Hospital. Rideshare data from Uber and Google covered trips taken within Houston, Texas, from February 2014 (the date of deployment of Uber to Houston) to December 2018. Impaired driving convictions included all indictments made by the Harris County, Texas, District Attorney’s office from January 2007 to December 2018. All adults with MVC traumas evaluated at both centers in the study population (individuals >16 years with a mechanism of injury classified under “motor vehicle collision”) were included. Impaired driving incidents were included only if the final legal outcome was conviction. MAIN OUTCOMES AND MEASURES The primary study outcomes were the incident rate ratios for hourly MVC traumas and daily impaired driving convictions. RESULTS A total of 23 491 MVC traumas (involving patients with a mean [SD] age of 37.9 [17.8] years and 14 603 male individuals [62.1%]), 93 742 impaired driving convictions, and more than 24 million Uber rides were analyzed. Following the introduction of Uber in February 2014, MVC traumas decreased by 23.8% (from a mean [SD] of 0.26 [0.04] to 0.21 [0.06] trauma incidents per hour) during peak trauma periods (Friday and Saturday nights). The incident rate ratio of MVC traumas following Uber deployment was 0.33 (95% CI, 0.17-0.67) per 1000 indexed rides (P = .002). Furthermore, rideshare use was associated with a significant, geographically linked reduction in impaired driving convictions between January 2014 to December 2019 (incidence rate ratio, 0.76 [95% CI, 0.73-0.78]; P < .001). CONCLUSIONS AND RELEVANCE In this study, introducing rideshare services in the Houston metropolitan area was associated with significant reductions in MVC traumas and impaired driving convictions. Increased use of rideshares may be an effective means of reducing impaired driving and decreasing rate of MVC traumas. Author Affiliations: Author affiliations are listed at the end of this article. Corresponding Author: Christopher R. Conner, MD, PhD, Department of Neurosurgery, McGovern Medical School at the University of Texas Health Science Center at Houston, 6400 Fannin St, Ste 2800, JAMA Surg. doi:10.1001/jamasurg.2021.2227 Houston, TX 77030 (christopher.r. Published online June 9, 2021. firstname.lastname@example.org). (Reprinted) E1 Research Original Investigation Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma otor vehicle crashes (MVCs) are an important public health concern and a leading cause of death for Key Points 1-3 M people younger than 65 years. Alcohol intoxica- Question By decreasing impaired driving, are rideshare services tion plays a likely role in approximately one-third of MVCs and associated with changes in motor vehicle trauma rates? is associated with more severe injuries than unimpaired Findings This multicenter cohort study obtained hospital data at 4,5 MVCs. Even though public policy initiatives have helped both major trauma centers in Houston, Texas; convictions for decrease alcohol-associated mortality, the prevalence of im- impaired driving from the Harris County, Texas, District Attorney’s paired driving remains high. Because MVCs have an esti- office; and rideshare use data from Uber and Google for Houston. mated $242 billion economic cost in the US alone, all pos- Rideshare volume had a significant negative correlation with the sible means of reducing the incidence of MVC traumas incidence of motor vehicle–associated trauma, and this was most evident in those younger than 30 years; a significant decrease in must be explored. convictions for impaired driving was associated with the In the past decade, individuals had the choice to use a ride- introduction of rideshare services. sharing service while socializing. Ridesharing became a prom- ising alternative to impaired driving, in that 33% of US rides- Meaning By using rideshares to avoid impaired driving, young people may aid in decreasing motor vehicle trauma. hare customers use this resource to avoid driving while 8,9 impaired. Yet, research on the outcomes of ridesharing is lim- ited. Previous studies have failed to capture data from non- Texas Medical Center and Ben Taub General Hospital. Both are 10-13 fatal incidents. Unfortunately, contradictory evidence American College of Surgeons level 1 trauma centers and the exists, in that some studies indicate that rideshares lead to only level 1 trauma centers in the greater Houston area. 12-15 an increase in MVCs, while others report a negative 4,10,16-19 association. Inconclusive results in previous analyses Participants, Variables, and Data Sources are likely secondary to temporal granularity on the order of Patient-level data from the MVC traumas presenting to the 2 months, thereby lacking the detail to control daily fluctua- trauma centers were retrospectively obtained from January tions in rideshare use, diurnal traffic patterns, and frequency 2007 to November 2019. Data were sourced from profession- 1,10-12,17,20 of convictions for impaired driving. ally maintained hospital trauma registries. Patients present- ThisstudyaimedtoquantifythechangeinMVCtraumasand ing with an injury mechanism classified as “motor vehicle impaired driving convictions after the introduction of rideshare trauma” were included regardless of outcome (fatal or nonfa- services. We hypothesized that increased rideshare use would tal). The MVC traumas of all acuity levels were included in correlate with decreased MVC traumas and impaired driving the analysis. Other injury mechanisms (eg, automobile- convictions.Weevaluatedspecifictrendsandchangesinhourly pedestrian crashes) and pediatric patients (<16 years old) were increments using Poisson regression by using highly granular excluded. Extracted data included the time of first medical data directly obtained from a rideshare company, Google, contact, demographic variables, and injury characteristics. andhospitalrecords.Ouranalysisalsoexaminedtheassociation The Injury Severity Score was used as a validated means of of rideshares with the rate of impaired driving convictions grading overall injury after trauma. whileconcurrentlycontrollingforgeography,populationgrowth, Rideshare use data came directly from Uber (Uber Tech- driving patterns, and alcohol consumption. This study was nologies Inc). These data included rides provided per hour from performed in Houston, Texas, because it is a large metropolitan February 2014 (the date of deployment of Uber to Houston) area with a concentrated hospital system that has only 2 major to December 2018. A minimum threshold was applied and data trauma centers: Memorial Hermann Hospital–Texas Medical indexed to May 6, 2014 (an index value of 1), prior to data trans- Center and Ben Taub General Hospital. Data from these 2 fer (threshold and indexing were performed by Uber, and the centers alone incorporate all major trauma in the Houston threshold was not disclosed to the research team). More than metropolitan area over the duration of the study. 24 million rides in the Houston metropolitan area were ana- lyzed. Multiple attempts were made to obtain similar data from the rideshare company Lyft Inc without response. A second rideshare data set was obtained from Google Trends (Google Methods LLC) search volume (http://trends.google.com), using the terms 4,18,23 This work was a multicenter cohort study. This article was writ- Uber and Lyft, with a temporal resolution of 1 month. ten in compliance with the Strengthening the Reporting of A Texas Public Information Act request for impaired driv- Observational Studies in Epidemiology (STROBE) guidelines. ing statistics was filed with the Harris County, Texas, District The Committee for the Protection of Human Subjects, the in- Attorney’s office. Arrest data for driving under the influence stitutional review board at University of Texas, Houston, anddrivingwhileintoxicatedincludeddatesandlocationsfrom approved this project. Informed consent was waived because January 2007 to December 2019. Data were limited to inci- there was minimal risk to included individuals and no proce- dents resulting in impaired driving guilty pleas, convictions, dures were performed. or probation (collectively termed convictions in this report), with a time resolution of days. Data from 2019 were excluded because of pending outcomes. Setting Hospital data were collected in Houston at the Red Duke Notably, MVC traumas and impaired driving convictions Trauma Institute within the Memorial Hermann Hospital– can be biased by vehicle usage and alcohol consumption. E2 JAMA Surgery Published online June 9, 2021 (Reprinted) jamasurgery.com Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma Original Investigation Research Data concerning vehicle miles traveled in Harris County from 2 trauma centers were stable over the study period. The MVC January 2005 to December 2018 were obtained from the Texas traumas had a diurnal pattern, with the greatest numbers Department of Transportation’s report on multiyear roadway occurring on Friday and Saturday nights between 9 PM and data tables (https://www.txdot.gov/). Data regarding alcohol 3 AM. Comparison of total trauma rate from January 2007 to consumption for each county from January 2007 to October December 2013 (the date of Uber deployment) vs January 2019 were gathered from the Texas Open Data Portal (http:// 2014 to November 2019 revealed a 23.8% decrease in data.texas.gov). MVC traumas (from a mean [SD] of 0.26 [0.04] to 0.21 [0.06] Totalsaleswereinflationadjustedusingtheconsumerprice traumas per hour) during peak hours but was otherwise index for Harris County from the Bureau of Labor Statistics unchanged (Figure 1). (https://www.bls.gov/). Alcohol consumption was adjusted per Analysis of the MVC trauma patient demographics dem- capita using census data (https://www.census.gov/). onstrated significant changes in the treated patient popula- tion treated in Houston between January 2007 to December 2013 (pre–Uber introduction) and January 2014 to November Statistical Methods Geographic encoding was performed with GeoPy (version 2019 (post–Uber introduction) (Table). The mean (SD) age 1.21.0 [GeoPy Contributors], Python version 3.6.0 [Python significantly increased from 37.2 (14.4) years in the prerides- Software Foundation]) using the Google Maps API (GoogleV3 hare period to 39.4 (18.3) years after Uber’s introduction server) and displayed in R version 3.6.2 (R Foundation for (P < .001). The total number of patients with MVC trauma in Statistical Computing) using ggmap version 3.0.0 the 4 age groups (<30, 30-50, 51-75, and >75 years) were plot- (CRAN). Both trauma and impaired driving data were ted yearly (Figure 2). This demonstrated that the number of imported to R version 3.6.2 (R Foundation for Statistical patients younger than 30 years with MVC traumas decreased Computing) as a time series. They were binned by the time after 2015 (from 866 patients in 2013 to 529 in 2018 [a 38.9% of the first medical contact with a temporal resolution of 1 decrease]), while other age groups had stable incidence of MVC hour. Two outcome variables were modeled: MVC traumas traumas. and impaired driving convictions. A 0-inflated Poisson The mean (SD) Injury Severity Score significantly regression model was fitted for MVC traumas (using the pscl decreased following the introduction of rideshare services package version 1.5.5 [Political Science Computational (15.1 [11.7] to 14.0 [11.2]; P < .001). There was no difference Laboratory]) because data had an hourly time resolution between prerideshare and postrideshare periods in terms of with many points lacking events; an uninflated Poisson distribution of sex (eg, men: prerideshare, 9475 of 15 157 regression model was fitted for impaired driving convictions [62.3%]; postrideshare, 5128 of 8334 [61.5%]; P = .13) or sur- at a time resolution of days (using GLM version 3.6.2 vival (14 582 [96.2%] vs 8003 [96.0%]; P = .46). [CRAN]). The logarithm of vehicle miles traveled was used as an offset in both models. Year, month, weekday day or night, Impaired Driving Data and rideshare volume (from Google Trends and Uber) were We analyzed a total of 248 485 arrests for impaired driving. independent variables in both models. The model for Only arrests resulting in a guilty plea, conviction, or proba- impaired driving convictions also included inflation- tion were included, leaving 96 520 incidents. Of these, 1294 adjusted alcohol sales per 1000 persons as an independent had inaccurate location data and 1484 were duplicate rec- variable. The significance threshold was set at P < .05, ords, leaving 93 742 impaired driving convictions (a category 2-tailed. Further details can be seen in the eAppendix in the including probation and guilty pleas) for analysis. From Janu- Supplement. ary 2007 to December 2013, daily impaired driving convic- tions were unchanged, with a mean (SD) of 22.5 (10.9) per day. From January 2014 to December 2018, the rate declined to a mean (SD) of 19.0 (10.3) impaired driving convictions per day. Results Impaired driving convictions decreased the most for arrests occurring on Fridays, Saturdays, and Sundays (2007 vs 2018: Trauma Data From January 2007 to November 2019, there were 23 491 Friday, 1323 to 1089 convictions [−17.7%]; Saturday, 1802 to MVC trauma evaluations at the 2 trauma centers (16 024 at 1538 convictions [−14.7%]; Sunday, 1728 to 1714 convictions Memorial Hermann Hospital–Texas Medical Center and 7467 [−1.0%]; Figure 1). In addition to fewer impaired driving con- at Ben Taub General Hospital). Of these, 6920 MVCs evalua- victions, there was a significant shift in the geographic distri- tions were level 1 trauma activations (the highest level of bution of the arrests associated with convictions. Before Janu- acuity) of 28 053 total level 1 traumas from all causes. ary 2014, impaired driving convictions stemmed from arrests Involved individuals had a mean (SD) age of 37.9 (17.8) years; predominately located in Houston’s core (encircled by the 14 603 were male (62.1%). Over this period, all-cause level 1 Interstate 610 loop), while post-2014 (from January 2014 to trauma activations increased by 396 activations (20.2%), December 2018) convictions stemmed from arrests made corresponding with a population increase of 1 023 997 mostly outside of Houston’s core (Figure 3). people (26.5%) in Harris County. In 2007, there were 1911 MVC traumas, compared with 1527 from December 2018 to Seasonal Trends November 2019 (a 20.1% decrease). Regional interhospital Three months (March, May, and June) showed deviations in transfer and emergency medical transport patterns to the the number of MVC traumas. There was a corresponding jamasurgery.com (Reprinted) JAMA Surgery Published online June 9, 2021 E3 Research Original Investigation Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma share services and MVC traumas (eTable 1 in the Supple- Figure 1. Weekly and Hourly Trends of Motor Vehicle Trauma, ment). There was a significant nocturnal association with Impaired Driving, and Uber Usage Friday and Saturday, corresponding with the greatest num- A Hourly rate of trauma ber of MVC traumas (rate ratios: Friday, 1.60 [95% CI, 1.49- 1.72]; Saturday, 1.82 [95% CI, 1.70-1.96]; P < .001). This was a 0.5 yearly trend after the introduction of Uber in February 2014, 2007-2013 with significantly fewer MVC traumas after January 2016 2014-2019 0.4 (rate ratio, 0.63 [95% CI, 0.47-0.85]; P < .001). 0.3 Uber Regression Results Rideshare volume was associated with a 67% reduction of MVC 0.2 traumas following Uber deployment (incident rate ratio, 0.33 [95% CI, 0.17-0.67] per 1000 indexed rides; P = .002) (eTable 2 0.1 in the Supplement). There was a significant downward trend in MVC trauma observed from January 2014 to December 2019 (rate ratio, 0.63 [95% CI, 0.57-0.68]; P < .001). Monday Tuesday Wednesday Thursday Friday Saturday Sunday Moreover, there were significantly more traumas on Fri- day nights (rate ratio, 1.63 [95% CI, 1.57-1.69]; P < .001) and B Total impaired driving convictions Saturday nights (rate ratio, 1.84 [95% CI, 1.78-1.91]; P < .001). Corresponding with seasonal variations, there were significantly fewer MVC trauma in September (rate ratio, 0.91 [95% CI, 0.85-0.98]; P = .02) and more in May (rate ratio, 1.08 [95% CI, 1.00-1.16]; P = .04) and June (rate ratio, 2015 1.09 [95% CI, 1.01-1.17]; P = .02). Regression analysis indi- cated that the volume of Uber rides did not directly affect 2018 the rate of impaired driving convictions. However, we did find a significant decrease in the number of impaired driving convictions for Friday and Saturday nights (the nights with highest arrests resulting in impaired driving convictions; Mon Tues Wed Thurs Fri Sat Sun rate ratio, 0.72 [95% CI, 0.70-0.74]; P < .001; eTable 3 in the C Indexed rideshare ride volume Supplement). There was also a yearly significant downward trend in impaired driving convictions following the intro- duction of rideshare services from February 2014 to Decem- ber 2018 (rate ratio, 0.76 [95% CI, 0.71-0.80]; P < .001; eTable 3 in the Supplement). Discussion Using 3 distinct sources of data (institutional trauma data, rideshare volume, and impaired driving convictions), this Monday Tuesday Wednesday Thursday Friday Saturday Sunday study provides the initial evidence that introducing rides- hare services to the Houston area was associated with a A, Hourly rate of trauma (3-hour moving average), both before and after decrease in the number of MVC traumas and impaired driv- introducing Uber rideshare services in January 2014. B, Total impaired driving ing convictions. The association of rideshares with MVC convictions (driving under the influence [DUIs] and driving while intoxicated traumas is 2-fold: a direct, negative association with rides- [DWIs]) for all days in a given year, with a peak on Friday and Saturday nights. C, Indexed Uber ride volume during the first year (2014) of introduction and last hare volume and a significant yearly trend of decreasing year of analysis (2018). As part of the data use agreement with Uber, we were MVCs traumas post-2014. The associations between rides- prohibited from presenting numerical ranges on the y-axis of this graph, haring and impaired driving included variations by geogra- because this was considered proprietary. phy, driving behavior, and patterns of alcohol consumption. increase in alcohol sales and consumption in these months Previous studies have demonstrated that individuals (Figure 4). Seasonal variation in impaired driving convic- younger than 30 years show increased Uber use rates. Our tions also increased in March and May, which corresponded findings are consistent with those showing significant reduc- with increased alcohol sales. tions in the population younger than 30 years (Table; 2013 vs 2018, 866 vs 529 [38.9%]), which traditionally is the primary age group involved in severe motor vehicle trauma Google Trends Regression Results Poisson regression did not demonstrate a significant correla- (Figure 2). Since MVCs are the number 1 cause of mortality in tion between Google Trends searches for Uber and Lyft ride- this age group, increased use of rideshare services plays a E4 JAMA Surgery Published online June 9, 2021 (Reprinted) jamasurgery.com Total DUIs/DWIs, in 100s Traumas, No./h Rideshare ride volume Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma Original Investigation Research Table. Population Demographics Abbreviation: NA, not applicable. Demographics for the prerideshare Characteristic 2007-2013 2014-2019 P value period (January 2007 to Total motor vehicle crash traumas, No. 15 157 8334 NA December 2013) and postrideshare introduction (January 2014 to Age, mean (SD), y 37.2 (14.4) 39.4 (18.3) <.001 November 2019) are shown, Male, No. (%) 9475 (62.3) 5128 (61.5) including the total number of Female, No. (%) 5682 (37.5) 3206 (38.5) incidents. Total motor vehicle traumas include activations Injury Severity Score, mean (SD) 15.1 (11.7) 14.0 (11.2) .01 of all acuity levels (level 1 and Survival total, No. (%) 14 582 (96.2) 8003 (96.0) .46 non–level 1). and before-and-after treatment analysis. These approaches Figure 2. Age-Associated Changes in Motor Vehicle Crash (MVC) model rideshare volume as a step function. Our data indicate Incidence that these services have gained popularity in a sigmoidal pat- A Total MVC traumas tern over 2 years. In using direct rideshare volume as a con- Age, y <30 51-75 tinuous regressor, there is no implicit assignment of control 30-50 ≥75 or treatment classes, as in difference-in-difference methods. Furthermore, this approach eliminates the temporal hetero- geneity of rideshare use that limits before-after treatment analyses. Continuous data also permit a dose-dependent re- sponse estimate. As an alternative method, we used search volume for rideshare companies. We note that search vol- umes will aggregate rideshare trips at all times, including those not associated with impaired driving (eg, airport trips). This method also assumes that search habits are associated with ride volume. We used both methods (direct rideshare 2007 2011 2015 2019 Year volumes and search volumes) in our analysis. We found that while significant MVC reductions occurred for peak times B ISS scores with direct rideshare data, search-volume data lacked statis- 50-74 tical power. This result indicates the need for data of high 50-74 25-49 temporal resolution to investigate the associations between 16 000 16-24 ridesharing and MVCs. 9-15 In addition, we expected that the incidence of MVC trau- 12 000 1-8 mas would increase during the weekends and warmer sea- sons (given an increase in alcohol consumption and nonsed- entary activities). Indeed, both inflation-adjusted alcohol sales and MVC traumas displayed seasonal variability but with dif- ferent patterns of summer increases (Figure 4). May and June are late spring and summer months in which outdoor activi- 2007-2013 2014-2019 ties increase. In Houston, an increase in MVC traumas during Year March may be traceable to the Houston Livestock Show and Rodeo. This event, known as the rodeo, annually hosts more A, Total annual MVC traumas were calculated for 4 age ranges: younger than 30, than 2.5 million visitors. 30-50, 51-75, and older than 75 years. Stable trauma incidence of trauma was Of these, alcohol sales more closely accounted for the noted across the study period, except for those younger than 30 years. Starting in 2016, there was a significant decrease in the incidence of this age range, seasonal variability in impaired driving convictions, but corresponding to the introduction and increased use of rideshares post-2014 the diurnal MVC traumas on a weekly time scale mirrored (866 in 2013 and 529 in 2018; a 38.9% reduction). B, Injury Severity Score (ISS) the known variation in impaired driving convictions distributions for 2007-2013 (prerideshare) and 2014-2019 (postrideshare) also (Figure 1). We conclude that the decrease in MVC traumas show the overall decrease in MVC incidence as well as relative decreases in higher ISS traumas. The ISS ranges are broken down into minor (1-8), moderate during both peak alcohol consumption hours and after intro- (9-15), serious (16-24), severe (25-49), critical (50-74), and maximal (75). ducing ridesharing services suggests individuals choose the Counts and percentages for ISS distributions are in eTable 4 in the Supplement. ridesharing as a safe alternative to impaired driving. Geography also played a role. Following the introduction role in preventing avoidable injuries. In older populations, of rideshare services, impaired driving convictions in Hous- Uber had no association with motor vehicle trauma, a find- ton’s core saw the greatest decrease. Notably, this area also ing that connects to the significantly lower rideshare use in had the highest Uber service volume (Figure 3). This region’s these populations. lowered convictions still occurred, despite an increase in both However, our study contradicts findings from some prior the population number and density. Impaired driving con- investigations. Previous methods include difference-in- victions in outlying areas remained flat or increased, likely 10,12-14,16,18,19 17 difference comparisons, interrupted time series, because of low rideshare service adoption (potentially jamasurgery.com (Reprinted) JAMA Surgery Published online June 9, 2021 E5 No. of patients Total MVC traumas Δ DUIs/DWIs No. of DUIs/DWIs Relative density Research Original Investigation Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma Figure 3. Geographic Distribution of Impaired Driving Arrests in Houston, Texas A Houston B Rideshare rides BELTWAY C Prerideshare (2009-2013) D Postrideshare (2014-2018) Geographic density maps were generated for Uber rides and impaired driving convictions (driving under the influence [DUIs] and driving while intoxicated [DWIs]) in Houston. A and B, Uber rides were greatest within the core of Houston >0 (within I-610) where population is densest and around the 2 major airports. C and D, Impaired driving convictions (DUIs and DWIs) over equal-duration epochs, 2009-2013 E Postrideshare minus prerideshare (pre-Uber) and 2014-2018 (post-Uber), were visualized over the same map as Uber rides. Pre-Uber, convictions were highest at the city core and along major highways. Post-Uber, the peak of convictions within the core was diminished 1 relative to convictions outside of it. E, A subtraction of post-Uber minus –1 pre-Uber rideshares was computed geographically to better visualize the change. Decreased convictions (in blue) were noted in the core, matching with the region of greatest Uber rides. An increase (in orange) in –12 impaired driving outside the core was seen in areas with less Uber use. because of higher ride costs, low driver availability, and lon- of data from all trauma centers (level 2 and lower levels) is a ger wait times). limitation of the study tempered by the stability in referral pat- terns seen in all-cause trauma. Compared with other metro- politan areas in the US, Houston also has a lower population Limitations This study was limited by its evaluation of only a single city, density, fewer public transit options, and significantly higher 16,30 resulting in potentially limited generalizability. Indeed, Hous- use of personal motor vehicles for transportation. It is pos- ton has many unique features. First, there were only 2 level 1 sible that these characteristics lead to higher rates of im- trauma centers, paired with limited level 2 trauma center ca- paired driving, in that people are more likely to imbibe fur- pacity during this period. This hospital system structure aided ther from home and with fewer options to travel. Nevertheless, the study in that more traumas could be analyzed with ho- the possibility that governments can reduce motor vehicle mogenous data, but it differs from other cities with greater trauma by increasing accessibility to reliable, on-demand trans- numbers of trauma centers and broader distribution. The lack portation should be explored further. E6 JAMA Surgery Published online June 9, 2021 (Reprinted) jamasurgery.com Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma Original Investigation Research Figure 4. Monthly and Seasonal Trends of Motor Vehicle Crashes and Impaired Driving 9.6 Alcohol sales Trauma 9.2 No. of DUIs/DWIs 8.8 Black indicates Harris County, Texas, beer, wine, and liquor sales from 8.4 January 2007 to October 2019, when 19 a mean was calculated by month. 8.0 March consumption is linked to the Houston Rodeo and December consumption to holidays. Blue 7.6 indicates corresponding trends in impaired driving (driving under the influence [DUIs] and driving while 7.2 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec intoxicated [DWIs]). Orange indicates Month motor vehicle crash traumas with a mean calculated by month. decrease in MVC trauma and impaired driving convictions. Future work will focus on the association between demo- Conclusions graphics and socioeconomic status and ridesharing services, Overall, this study indicates that introducing rideshare ser- MVC traumas, and impaired driving in other metropolitan vices to the Houston area was associated with a significant areas. ARTICLE INFORMATION Administrative, technical, or material support: 2. Richard CM, Magee K, Bacon-Abdelmoteleb P, Conner, Ray, Parker, Harvin, Kitagawa. Brown JL. Countermeasures that work: a highway Accepted for Publication: March 3, 2021. Supervision: Conner, Ray, Vera, Harvin, Kitagawa. safety countermeasure guide for state highway Published Online: June 9, 2021. safety offices, 9th ed. Published 2017. Accessed Conflict of Interest Disclosures: None reported. doi:10.1001/jamasurg.2021.2227 May4,2021. https://www.nhtsa.gov/sites/nhtsa. Funding/Support: The study was funded by the Open Access: This is an open access article gov/files/documents/812478_countermeasures- HeadStrong Brain Injury Foundation and Alpha distributed under the terms of the CC-BY License. that-work-a-highway-safety-countermeasures- Omega Alpha. © 2021 Conner CR et al. JAMA Surgery. guide-.pdf Role of the Funder/Sponsor: The funders had no Author Affiliations: Department of Neurosurgery, 3. James SL, Abate D, Abate KH, et al; GBD 2017 role in the design and conduct of the study; McGovern Medical School at the University of Texas Disease and Injury Incidence and Prevalence collection, management, analysis, and Health Science Center at Houston, Houston Collaborators. Global, regional, and national interpretation of the data; preparation, review, or (Conner, McCormack, Dickey, Parker, Kitagawa); incidence, prevalence, and years lived with approval of the manuscript; and decision to submit Department of Cardiothoracic and Vascular Surgery, disability for 354 diseases and injuries for 195 the manuscript for publication. McGovern Medical School at the University of Texas countries and territories, 1990-2017: a systematic Health Science Center at Houston, Houston (Ray); Additional Contributions: The authors would like analysis for the Global Burden of Disease Study University of Texas Health Science Center at to thank Jonathan Hall, PhD, Jonathan Wang, BS, 2017. Lancet. 2018;392(10159):1789-1858. doi:10. Houston School of Public Health, Houston (Dickey); Dana Kraushar, BS, Uber, for assisting with 1016/S0140-6736(18)32279-7 Center for Clinical and Translational Sciences, rideshare data; the Harris County District Attorney’s 4. Graf M. Assessing the impact of ridesharing McGovern Medical School at the University of Texas office for information on impaired driving charges services on public health and safety outcomes. Health Science Center at Houston, Houston and convictions; Rebecca Crocker, BA, Red Duke Published 2017. Accessed May 4, 2021. https:// (Zhang); Department of Surgery, Baylor College of Trauma Institute at Memorial Hermann Hospital– milkeninstitute.org/sites/default/files/reports-pdf/ Medicine, Houston, Texas (Vera); Department of Texas Medical Center, and Robin Garza, BA, Ben 110117-Ridesharing-and-Public-Health.pdf Surgery, McGovern Medical School at the University Taub General Hospital, for supplying institutional 5. Blincoe LJ, Miller TR, Zaloshnja E, Lawrence BA. of Texas Health Science Center at Houston, trauma registry data; and Karl Schmitt, MD, The economic and societal impact of motor vehicle Houston (Harvin). University of Texas, Houston, for thoughtful crashes, 2010 (revised): report No. DOT HS 812 013. discussions on trauma and neurosurgery. We thank Author Contributions: Dr Conner had full access to Published 2010. Accessed May 6, 2021. https:// Life Science Editors, especially Brandi Mattson, all the data in the study and takes responsibility for crashstats.nhtsa.dot.gov/Api/Public/ PhD, for editorial assistance. Dr Mattson was the integrity of the data and the accuracy of the ViewPublication/812013 compensated for her contributions; the other data analysis. named individuals were not. 6. Hingson R, Zha W, Smyth D. Magnitude and Concept and design: Conner, Ray, McCormack, trends in heavy episodic drinking, alcohol-impaired Parker, Vera, Harvin, Kitagawa. Additional Information: Uber has relinquished driving, and alcohol-related mortality and overdose Acquisition, analysis, or interpretation of data: editorial control over the manuscript, methods, and hospitalizations among emerging adults of college Conner, Ray, McCormack, Dickey, Zhang, Harvin, findings. A preliminary draft was provided to the ages 18–24 in the United States, 1998–2014. J Stud Kitagawa. company, but no substantive changes were made Alcohol Drugs. 2017;78(4):540-548. doi:10.15288/ Drafting of the manuscript: Conner, Ray, because of their review. jsad.2017.78.540 McCormack, Dickey, Parker, Harvin, Kitagawa. Critical revision of the manuscript for important REFERENCES 7. Shannon D, Rizzi L, Murphy F, Mullins M. intellectual content: Conner, Ray, McCormack, Exploring the Price of Motor Vehicle Collisions—A 1. International Transport Forum. Road safety Parker, Zhang, Vera, Harvin, Kitagawa. Compensation Cost Approach. Transportation annual report 2019. Published 2019. Accessed May Statistical analysis: Conner, McCormack, Zhang, Research Interdisciplinary Perspectives; 2020:4. 6, 2021. https://www.itf-oecd.org/road-safety- Harvin. annual-report-2019 Obtained funding: Conner, Kitagawa. jamasurgery.com (Reprinted) JAMA Surgery Published online June 9, 2021 E7 Trauma, in 100s No. of DUIs/DWIs, in 100s Alcohol sales, in $10 million Research Original Investigation Association of Rideshare Use With Alcohol-Associated Motor Vehicle Crash Trauma 8. Clewlow RR, Mishra GS. Disruptive 17. Morrison CN, Jacoby SF, Dong B, Delgado MK, 24. The R Project for Statistical Computing. Getting transportation: the adoption, utilization, and Wiebe DJ. Ridesharing and motor vehicle crashes in started. Accessed May 6, 2021. www.R-project.org impacts of ride-hailing in the United States. 4 US cities: an interrupted time-series analysis. Am 25. Kahle D, Wickham H. ggmap: spatial Published 2017. Accessed May 4, 2021. https:// J Epidemiol. 2018;187(2):224-232. doi:10.1093/ visualization with ggplot2. The R Journal. 2013;5(1): escholarship.org/uc/item/82w2z91j aje/kwx233 144-161. doi:10.32614/RJ-2013-014 9. Mothers Against Drunk Driving. Rideshare saves 18. Dills AK, Mulholland SE. Ride-sharing, fatal 26. Political Science Computational Laboratory. lives! Published 2015. Accessed May 6, 2021. crashes, and crime. Southern Economic Journal. atahk/pscl. Updated April 20, 2020. Accessed May https://www.madd.org/the-solution/drunk-driving/ 2018;84(4):965-991. doi:10.1002/soej.12255 11, 2021. http://github.com/atahk/pscl rideshare/ 19. Martin-Buck FPV. Driving safety: an empirical 27. Sarriera JM, Álvarez GE, Blynn K, Alesbury A, 10. Greenwood BN, Wattal S. Show me the way to analysis of ridesharing’s impact on drunk driving Scully T, Zhao J. To share or not to share: go home: an empirical investigation of ride sharing and alcohol-related crime, University of Texas at investigating the social aspects of dynamic and alcohol related motor vehicle homicide, Fox Austin working paper. Published 2017. Accessed ridesharing. Transportation Research Record.2017; School of Business research paper No. 15-054. SSRN. May6,2021. https://www.semanticscholar.org/ 2605(1):109-117. doi:10.3141/2605-11 2015. doi:10.2139/ssrn.2557612 paper/Driving-Safety-%3A-An-Empirical-Analysis- 28. Johnson CW, Woods HD. 2019 Houston of-%E2%80%99-s-on-Martin-Buck/3 11. Dills AK, Mulholland SE. Ride-sharing, fatal Livestock Show and Rodeo economic impact study. f1eb273fcee888441147105882dd12ca811fd35 crashes, and crime. SSRN. 2016. Published 2019. Accessed May 4, 2021. https:// 20. Mallett JM. Trends in public transportation 12. Barrios JM, Hochberg YV, Yi H. The cost of www.rodeohouston.com/Portals/0/Content/ ridership: implications for federal policy. Published convenience: ridesharing and traffic fatalities. SSRN. AboutUs/WhoWeAre/Economic%20Impact/2019- 2018. Accessed May 4, 2021. https://fas.org/sgp/ 2019;3361227. EconomicImpact-Summary.pdf crs/misc/R45144.pdf 13. Kirk DS, Cavalli N, Brazil N. The implications of 29. National Highway and Traffic Safety 21. von Elm E, Altman DG, Egger M, Pocock SJ, ridehailing for risky driving and road accident Administration. Time of day and demographic Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. injuries and fatalities. Soc Sci Med. 2020;250:112793. perspective of fatal alcohol-impaired-driving The Strengthening the Reporting of Observational doi:10.1016/j.socscimed.2020.112793 crashes: traffic safety facts DOT HS 811523. Studies in Epidemiology (STROBE) statement: Published 2011. Accessed May 4, 2021. https:// 14. Huang JY, Majid F, Daku M. Estimating effects guidelines for reporting observational studies. Ann crashstats.nhtsa.dot.gov/Api/Public/ of Uber ride-sharing service on road traffic-related Intern Med. 2007;147(8):573-577. doi:10.7326/ ViewPublication/811523 deaths in South Africa: a quasi-experimental study. 0003-4819-147-8-200710160-00010 J Epidemiol Community Health. 2019;73(3):263-271. 30. Qian Z. Without zoning: urban development 22. Osler T, Baker SP, Long W. A modification of the doi:10.1136/jech-2018-211006 and land use controls in Houston. Cities. 2010;27(1): Injury Severity Score that both improves accuracy 31-41. doi:10.1016/j.cities.2009.11.006 15. Brazil N, Kirk DS. Uber and metropolitan traffic and simplifies scoring. J Trauma. 1997;43(6):922-925. fatalities in the United States. Am J Epidemiol. 2016; doi:10.1097/00005373-199712000-00009 184(3):192-198. doi:10.1093/aje/kww062 23. Hall JD, Palsson C, Price J. Is Uber a substitute 16. Peck JL. New York City drunk driving after or complement for public transit? Journal of Urban Uber: working paper 13. Published 2017. Accessed Economics. 2018;10836-50. doi:10.1016/j.jue.2018. May4,2021. https://academicworks.cuny.edu/cgi/ 09.003 viewcontent.cgi?article=1012&context=gc_econ_wp E8 JAMA Surgery Published online June 9, 2021 (Reprinted) jamasurgery.com Supplemental Online Content Conner CR, Ray HM, McCormack RM, et al. Association of rideshare use with alcohol- associated motor vehicle crash trauma. JAMA Surg. Published online June 9, 2021. doi:10.1001/jamasurg.2021.2227 eAppendix. Materials and Methods: Indexed Uber Rides and Statistical methods eTable 1. Result of ZIP regression of Google Trends ride volume estimate on MVC traumas eTable 2. Result of ZIP regression of hourly Uber rideshare data on MVC traumas eTable 3. Result of Poisson regression of daily Uber rideshare data on drunk driving convictions eTable 4. Injury severity score (ISS) before and after introduction of rideshares This supplemental material has been provided by the authors to give readers additional information about their work. © 2021 American Medical Association. All rights reserved. eAppendix. Materials and Methods Index Uber Rides Hourly rideshare volume data were supplied directly from Uber. Data were supplied from introduction of the service to Houston in February 2014 through the end of 2018. Hourly rides were thresholded above a minimum ride volume and then indexed relative to May 6, 2014 (index value of 100). Statistical Methods Data were imported into R (ver 3.6.2, CRAN and R Foundation for Statistical Computing) as time series and binned by time of first medical contact with a temporal resolution on the order of 1 hour. Two outcome variables were modeled: MVC traumas and drunk driving convictions (DUIs/DWIs). The zero-inflated Poisson (ZIP) regression model was fitted for MVC traumas and the Poisson regression model with a dispersion parameter was fitted for DUIs/DWIs. Logarithm of VMT was used as an offset in both models. Year, month, weekday day/night, and rideshare volume (from Google Trends and Uber) were the regressors in both models. The model for DUIs/DWIs also included inflation-adjusted alcohol sales per 1,000 persons as a regressor. All analyses were performed in R. © 2021 American Medical Association. All rights reserved. eTable 1. Result of ZIP regression of Google Trends ride volume estimate on MVC traumas. Results are reported as regression coefficient and standard error estimate from zero-inflated Poisson model. To estimate the rate ratio, the natural exponent of the regression coefficient is calculated. Variable Regression coefficient estimate p value (standard error) -4 -3 Google trends (Uber + Lyft) -7.1x10 (1.4x10 ) p=0.61 Year effect (reference year = 2007) -2 -2 2008 2.1x10 (3.5x10 ) p=0.54 -2 -2 2009 3.1x10 (3.5x10 ) p=0.37 -2 -2 2010 -4.0x10 (3.5x10 ) p=0.26 -2 -2 2011 -6.3x10 (3.6x10 ) p=0.08 -2 -2 2012 -9.0x10 (3.6x10 ) p=0.01 -2 -2 2013 -3.7x10 (3.6x10 ) p=0.30 -2 -2 2014 -3.6x10 (4.7x10 ) p=0.45 -1 -2 2015 -1.4x10 (8.9x10 ) p=0.12 -1 -1 2016 -2.6x10 (1.2x10 ) p=0.04 -1 -1 2017 -3.7x10 (1.4x10 ) p<0.01 -3 -1 2018 -4.6x10 (1.5x10 ) p<0.002 Month effect (reference month = January) -3 -2 February 4.8x10 (3.8x10 ) p=0.90 -2 -2 March 6.6x10 (3.7x10 ) p=0.07 -2 -2 April 2.7x10 (3.7x10 ) p=0.46 -2 -2 May 7.8x10 (3.7x10 ) p=0.03 -2 -2 June 9.0x10 (3.8x10 ) p=0.02 -2 -2 July -7.6x10 (3.8x10 ) p=0.84 -2 -2 August -5.0x10 (3.9x10 ) p=0.20 -2 -2 September -9.0x10 (3.9x10 ) p=0.02 -2 -2 October -2.0x10 (3.8x10 ) p=0.60 -3 -2 November -6.0x10 (3.8x10 ) p=0.88 -2 -2 December 4.2x10 (3.8x10 ) p=0.22 Weekday effect (reference = Sunday day) -3 -2 Sunday night -1.0x10 (4.0x10 ) p=0.80 -2 -2 Monday day -6.6x10 (4.1x10 ) p=0.10 -1 -2 Monday night -3.7x10 (4.4x10 ) p<0.01 -2 -2 Tuesday day -9.0x10 (4.1x10 ) p=0.03 -1 -2 Tuesday night -3.0x10 (4.4x10 ) p<0.01 -1 -2 Wednesday day -1.4x10 (4.2x10 ) p<0.01 -1 -2 Wednesday night -2.3x10 (4.3x10 ) p<0.01 -2 -2 Thursday day -7.8x10 (4.1x10 ) p=0.06 -2 -2 Thursday night -3.7x10 (4.1x10 ) p=0.36 -2 -2 Friday day 1.0x10 (4.0x10 ) p=0.79 -1 -2 Friday night 4.7x10 (3.7x10 ) p<0.01 -2 -2 Saturday day 3.1x10 (4.0x10 ) p=0.44 -1 -2 Saturday night 6.0x10 (3.6x10 ) p<0.01 © 2021 American Medical Association. All rights reserved. eTable 2. Result of ZIP regression of hourly Uber rideshare data on MVC traumas. Results are reported as regression coefficient and standard error estimates from zero-inflated Poisson model. Variable Regression coefficient estimate p value (standard error) -3 -4 Rideshare volume -1.1x10 (3.5x10 ) p=0.002 Year effect (reference year = 2007) -2 -2 2008 2.1x10 (3.5x10 ) p=0.54 -2 -2 2009 3.2x10 (3.5x10 ) p=0.46 -2 -2 2010 -4.0x10 (3.6x10 ) p=0.26 -2 -2 2011 -6.3x10 (3.6x10 ) p=0.08 -2 -2 2012 -9.0x10 (3.6x10 ) p=0.01 -2 -2 2013 -3.8x10 (3.6x10 ) p=0.28 -2 -2 2014 -4.9x10 (3.6x10 ) p=0.17 -1 -2 2015 -1.6x10 (3.7x10 ) p<0.01 -1 -2 2016 -2.8x10 (3.9x10 ) p<0.01 -1 -2 2017 -3.8x10 (4.3x10 ) p<0.01 -3 -2 2018 -4.7x10 (4.3x10 ) p<0.01 Month effect (reference month = January) -2 -2 February 3.9x10 (3.7x10 ) p=0.92 -2 -2 March 6.5x10 (3.6x10 ) p=0.08 -2 -2 April 2.6x10 (3.7x10 ) p=0.48 -2 -2 May 7.5x10 (3.6x10 ) p=0.04 -2 -2 June 8.5x10 (3.7x10 ) p=0.02 -2 -2 July -1.2x10 (3.7x10 ) p=0.75 -2 -2 August -5.5x10 (3.8x10 ) p=0.14 -2 -2 September -9.1x10 (3.7x10 ) p=0.02 -2 -2 October -2.0x10 (3.7x10 ) p=0.58 -2 -2 November -8.6x10 (3.7x10 ) p=0.81 -2 -2 December 4.2x10 (3.7x10 ) p=0.25 Weekday effect (reference = Sunday day) -3 -2 Sunday night -4.4x10 (4.0x10 ) p=0.91 -2 -2 Monday day -6.6x10 (4.1x10 ) p=0.11 -1 -2 Monday night -3.6x10 (4.4x10 ) p<0.01 -2 -2 Tuesday day -8.9x10 (4.1x10 ) p=0.03 -1 -2 Tuesday night -2.9x10 (4.4x10 ) p<0.01 -1 -2 Wednesday day -1.4x10 (4.2x10 ) p<0.01 -1 -2 Wednesday night -2.2x10 (4.3x10 ) p<0.01 -2 -2 Thursday day -7.7x10 (4.1x10 ) p=0.06 -2 -2 Thursday night -2.2x10 (4.1x10 ) p=0.60 -2 -2 Friday day 1.4x10 (4.0x10 ) p=0.72 -1 -2 Friday night 4.9x10 (3.7x10 ) p<0.01 -2 -2 Saturday day 3.5x10 (4.0x10 ) p=0.38 -1 -2 Saturday night 6.1x10 (3.6x10 ) p<0.01 © 2021 American Medical Association. All rights reserved. eTable 3. Result of Poisson regression of daily Uber rideshare data on drunk driving convictions. Results are reported as regression coefficient and standard error estimates from Poisson regression model with a dispersion parameter. Alcohol sales are $1 per person-year and inflation adjusted. Variable Regression coefficient estimate p value (standard error) -2 -2 Rideshare volume -1.5x10 (1.5x10 ) p=0.29 -2 -3 Alcohol Sales 2.6x10 (5.9x10 ) p<0.01 Year effect (reference year = 2007) -1 -2 2008 1.1x10 (1.9x10 ) p<0.01 -1 -2 2009 1.8x10 (2.1x10 ) p<0.01 -1 -2 2010 1.2x10 (2.0x10 ) p<0.01 -1 -2 2011 1.4x10 (1.9x10 ) p<0.01 -2 -2 2012 7.1x10 (1.9x10 ) p<0.01 -2 -2 2013 6.0x10 (2.0x10 ) p=0.002 -4 -2 2014 -8.0x10 (2.1x10 ) p=0.97 -2 -2 2015 -9.4x10 (2.1x10 ) p<0.01 -1 -2 2016 -2.1x10 (2.5x10 ) p<0.01 -1 -2 2017 -2.5x10 (2.8x10 ) p<0.01 -1 -2 2018 -2.8x10 (2.8x10 ) p<0.01 Month effect (reference month = January) -2 -2 February 9.0x10 (2.0x10 ) p<0.01 -2 -2 March 2.0x10 (2.6x10 ) p=0.45 -3 -2 April 6.9x10 (2.1x10 ) p=0.74 -2 -2 May -2.7x10 (2.4x10 ) p=0.26 -2 -2 June -1.9x10 (2.0x10 ) p=0.35 -2 -2 July -2.1x10 (2.0x10 ) p=0.30 -3 -2 August -6.8x10 (2.0x10 ) p=0.73 -2 -2 September 2.0x10 (2.0x10 ) p=0.31 -2 -2 October -2.9x10 (2.2x10 ) p=0.19 -3 -2 November 9.1x10 (2.0x10 ) p=0.65 -2 -2 December -7.3x10 (3.1x10 ) p=0.02 Weekday effect (reference day = Sunday) -1 -2 Monday -8.2x10 (1.5x10 ) p<0.01 -2 Tuesday -1.2 (1.7x10 ) p<0.01 -2 Wednesday -1.0 (1.6x10 ) p<0.01 -1 -2 Thursday -7.1x10 (1.4x10 ) p<0.01 -1 -2 Friday -3.3x10 (1.3x10 ) p<0.01 -2 -2 Saturday 3.9x10 (1.1x10 ) p=0.001 © 2021 American Medical Association. All rights reserved. eTable 4. Injury severity score (ISS) before and after introduction of rideshares. ISS for the pre-rideshare period (2007-2013) and post-rideshare introduction (2014-2019) are shown in total number of patients. ISS (percent) 2007-2013 2014-2019 1-8 (minor) 4253 (28.1) 2781 (33.4) 9-15 (moderate) 4984 (33.0) 2676 (32.1) 16-24 (serious) 3084 (20.4) 1606 (19.3) 25-49 (severe) 2537 (16.8) 1133 (13.6) 50-74 (critical) 212 (1.4) 101 (1.2) 75 (maximum) 56 (0.4) 37 (0.4) © 2021 American Medical Association. All rights reserved.
JAMA Surgery – American Medical Association
Published: Aug 9, 2021