==== Front Contemp Clin Trials CommunContemp Clin Trials CommunContemporary Clinical Trials Communications2451-8654Elsevier S2451-8654(18)30033-410.1016/j.conctc.2018.07.007ArticleDesign of an experimental protocol to examine medication non-adherence among young drivers diagnosed with ADHD: A driving simulator study Lee Yi-Ching YLEE65@gmu.edua∗Ward McIntosh Chelsea bWinston Flaura cPower Thomas bHuang Patty bOntañón Santiago dGonzalez Avelino ea George Mason University, USAb Children's Hospital of Philadelphia, USAc Children's Hospital of Philadelphia and University of Pennsylvania, USAd Drexel University, USAe University of Central Florida, USA∗ Corresponding author. Department of Psychology, George Mason University, 4400 University Drive, Fairfax, VA, 22030, USA. YLEE65@gmu.edu25 7 2018 9 2018 25 7 2018 11 149 155 26 2 2018 28 6 2018 24 7 2018 © 2018 The Authors. Published by Elsevier Inc.2018This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).The diagnosis of ADHD among teens and young adults has been associated with a higher likelihood of motor vehicle crashes. Some studies suggest a beneficial effect of ADHD medication but the exact efficacy is still being debated. Further, medication adherence, which is low in this age group, can further reduce effectiveness. Our long-term objective is to reduce unsafe driving among drivers with ADHD by detecting medication non-adherence through driver behavior modeling and monitoring. As a first step, we developed the described lab study protocol to obtain reliable driver behavior data that will then be used to design and train behavior models built through machine learning. This experimental study protocol was developed to systematically compare driving behaviors under two medication conditions (before and after intake of medication) among young adults with ADHD and a control group of non-ADHD. A driving simulator was used to examine driving behaviors and interactions with traffic. The primary outcome was speed management for two comparisons (ADHD vs. non-ADHD and before vs. after medication), and secondary objectives involved understanding differences among the participants utilizing self-reported surveys about ADHD symptoms, drivers' knowledge, and perception about safety. The study protocol was designed to maximize participant safety and efficiency of data collection, as multiple measures were collected over two 2-h study visits. The sampled ADHD drivers were demographically and psychosocially similar but clinically different from the non-ADHD group. Overall, this protocol was effective in participant recruitment and retention, allowed staggered data collection, and can be incorporated in a subsequent clinical trial that examines the efficacy of a machine-learning based driver monitoring intervention. Keywords Attention-deficit/hyperactivity disorder (ADHD)DrivingMedication adherenceMachine learningDriving simulator ==== Body 1 Introduction Attention-deficit/hyperactivity disorder (ADHD) is a prevalent neurobehavioral disorder in children and adolescents [16,38]. Approximately two-thirds of patients with a childhood diagnosis of ADHD continue to experience clinical symptoms into adulthood [6], intersecting with a period when many young adults start to drive independently. Drivers who are diagnosed with ADHD have shown significant driving impairments [1,2,19], including higher likelihood of motor vehicle crashes, speeding violations and poorer vehicle control, with a relative risk of 1.23 when controlling for exposure [39]. However, not all individuals with ADHD are affected uniformly [17] and it remains unclear the extent to which measures can be developed to distinguish between low and high risk drivers with ADHD on a group level and between low and high risk characteristics on an individual level [15]. Medication intervention, especially stimulant medication, appears to improve the driving deficits exhibited by ADHD drivers [17], although the exact efficacy is still being debated [3,5,8,11,12,24,37]. While the discrepancies may come from methodological limitations and sample size concerns, one general consensus is that the benefits and effects of medication on individuals with ADHD are not uniform [20], confirming the challenges for developing effective measures to distinguish between low and high risk drivers as well as strategic and sustainable treatment plans. In addition, medication adherence is a major problem in adolescence and young adulthood [18,32], as they transition from parent-managed medication to self-managed medication. Therefore, medication non-adherence can further degrade effectiveness of stimulant medication in reducing crash risk. Despite these safety concerns, very little work has been done to evaluate interventions that can improve ADHD symptom management and driving safety during adolescence and young adulthood, when the crash risk is the highest [33]. To address this gap in knowledge, our long-term objective is to create a machine-learning based monitoring intervention to help manage ADHD symptoms while driving. Such a system is expected to effectively monitor driving behavior in situations where ADHD symptoms are under relative control and when they are not, as determined by levels of medication adherence. As the first step toward fulfilling this long-term goal, we have designed an experimental protocol that involved the use of a driving simulator and other assessment measures and allowed for the comparison of driving behaviors between medication conditions (before and after the consumption of daily stimulant ADHD medication) and groups of participants (with and without ADHD). The current paper reports the design of an experimental protocol for collecting reliable driver behavior data; the recruitment and assessment strategies for the study sample – young adults with and without ADHD; and the comparisons of clinical, psychosocial, and demographic characteristics between the two groups. 2 Research design and methods 2.1 Study design The primary objective is to quantify the differences in vehicle control behaviors between two groups of participants – individuals with and without ADHD – as well as between two medication conditions (before and after medication administration) among individuals with ADHD. Traditional statistical methods and data mining techniques will be used to compare and contrast patterns of driving behaviors. The secondary objectives include conducting exploratory analysis to examine potential mechanisms and covariates that may explain the behavioral differences between the two groups and the two medication conditions. 2.2 Recruitment strategies Participants (individuals with self-reported ADHD and zip code matched individuals without ADHD who served as controls) were recruited through a variety of clinical and community settings. These included posting flyers at universities, university counseling centers, bus stops, coffee shops, grocery stores, psychology department subject pool, emailing listservs of undergraduate students and university disability services office, and word-of-mouth referrals from students and colleagues. The study received Institutional Review Board approval from the first author's university. 2.3 Eligibility criteria ADHD participants: Eligible participants were adults 18–24 years of age who had a confirmed diagnosis of ADHD, had a current prescription for stimulant medication for ADHD, held a restricted or an unrestricted driver's license, had normal or corrected-to-normal vision, and had normal hearing abilities. The exclusion criteria were self-reported pregnancy (females), self-reported neurodevelopmental disorders, intellectual disabilities, psychotic disorders, bipolar disorder, or seizure disorders, as well as participants with a confirmed diagnosis of ADHD but who took non-stimulant medication. Non-ADHD participants: Eligible participants were adults 18–24 years of age who held a restricted or an unrestricted driver's license, had normal or corrected-to-normal vision, and had normal hearing abilities. The exclusion criteria were the same as those of ADHD participants with the addition of self-reported diagnosis of ADHD. In addition, the Motion Sickness History Screening Form (MSHSF) [22,23] was used to assess the likelihood of experiencing simulator sickness (a form of motion sickness). The MSHSF includes questions about the frequency of getting carsick, seasick, and airsick. Based on the reported frequencies (a composite score of 7 or higher), we discouraged further participation. For participants who were discouraged but still wished to continue participation, the potential risks and safeguard measures were explained, and they were ensured they would be checked and monitored closely during the study. 2.4 Outcome measures The primary outcome was participant's average speed while driving in a simulator. This and other associated variables were collected from a high-fidelity, motion-based driving simulator, sampled at 60-Hz. It has an open-cab configuration equipped with a motion-base system capable of a single degree of pitch motion and a 90 ± degree high-quality yaw motion, a 3-channel visual system covering 180-deg forward field-of-view, and a force-feedback steering wheel. The variables derived from the simulator included vehicle control variables (e.g., velocity, throttle, brake, pitch) and vehicle diagnostics variables (e.g., gear, engine RPM). Three cameras capturing the foot movement, over the shoulder view (steering wheel movement), and upper body and face view were also recorded. The traffic scenarios used in the current study as well as the process for computing driving behaviors were previously developed and validated [26,31,34]. To examine secondary objectives, the study included several self-reported surveys:a) Demographic information was collected using questions about socio-demographics, driving experience, driving history (accidents), licensure type, learning-to-drive experience (who taught them to drive?), as well as four validated rating scales: Safe Speed Knowledge Test (SSKT) [27], Driving Anger Scale (DAS) [13], Brief Sensation Seeking Scale (BSSS) [30,41], and Driving Behavior Survey (DBS) [9]. b) ADHD history questionnaire was completed by each participant with ADHD and a friend or family member. These questions collected information about each participant's ADHD medication, symptom onset (age, symptom-related problems), and severity of problems or concerns currently caused by ADHD symptoms (when not taking the medication) in School, Work, Family Relationships, Social Relationships, and Self-Esteem categories. c) Conners' Adult ADHD Diagnostic Interview (CAADID) [14] was administered individually. It produced comprehensive demographic and developmental history to support a categorical diagnosis based upon the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV™) criteria for ADHD, during both adulthood and childhood. For screening purposes, both the quantitative and qualitative responses helped delineate the ADHD medical and symptom history by assessing each participant's demographic history, developmental course, ADHD risk factors, and comorbidity screening questions. d) Conners' Adult ADHD Rating Scale (CAARS) self-report and observer-report, screening versions [10] were administered individually. The screening versions consisted of 30 items about behaviors or problems sometimes experienced by adults. These rating scales were administered on-line via Multi-Health Systems' management program, and the calculated profile reports included normative T scores on inattentive symptoms, hyperactive-impulsive symptoms, total ADHD symptoms, and ADHD index. e) Post-drive survey was used after each driving simulator session. Participants were asked to rate the realism of the simulation and if there were concerns about the traffic scenarios. They were also asked to rate the percentage of time they were speeding, driving inside a lane, and following traffic rules. 2.5 Design of protocol The study protocol (key elements presented in Table 1) was designed to maximize participant safety and efficiency of data collection. The safety aspect of the protocol included screening for proneness to simulator sickness, confirmation of ADHD medication, safety features in the driving simulator, and continuous monitoring of participants. The efficiency of data collection part of the protocol included web-based self-reports and observer reports, and staggered and balanced study procedures. The order of the medication condition was counter-balanced, and the order of the experimental drives in the driving simulator was randomized.Table 1 Key elements of the protocol and their rationale. Table 1Key elements Details Justifications Medication condition Each ADHD participant was observed under two medication conditions - regular medication and delayed medication Because of safety concerns, ADHD participants were to be dropped off and picked up by a friend or family member for the study visits Scheduling of study visit ADHD participants could be observed on two separate days up to 14 days apart (option 1) or on one day (option 2) Two options for scheduling the study visits were offered to accommodate the chauffeuring friend or family member's availability Consumption of ADHD medication The medication was consumed under supervision The intake and route of dosage were confirmed Confirmation of ADHD medication Participant's medication was first confirmed by a list of US Food and Drug Administration-approved stimulant medications. Post study visit, the medications were confirmed by the study pediatrician This two-level of confirmation ensured that the medications were stimulants and that any variation of intake frequency and dosage was documented and confirmed Screening for proneness to simulator sickness and constant monitoring during study visits A validated survey was used. Participants who scored high on the survey were informed of their likelihood of developing simulator sickness while operating the simulator Participants were informed of all safety features in the simulator, including a safety gate, wearing seatbelt, emergency stop buttons as well as experimenter's constant monitoring of sickness indicators. Short breaks and cold water were provided. Use of web-based surveys CAARS and history questionnaire could be completed online or by phone, by participants (self-report) and their friend or family member (observer report) Added flexibility and options as to how and when these surveys could be completed Staggered study procedures CAADID and self-report surveys were conducted during the 1-h waiting period in the delayed medication condition, after the consumption of medication This was to minimize the study duration 2.6 Study procedures When interested individuals contacted the lab via email or phone, the experimenters first confirmed that they met the eligibility criteria. Once enrolled, the experimenters explained the two options for scheduling the study visits to ADHD participants (if option 1 was chosen, the two study visits were up to 14 days apart) and provided details of the lab location. All participants were informed that the purpose of the study is to understand driving behaviors and how the behaviors might be different under various driving and clinical conditions. They were notified that we would collect information about driving behaviors, their knowledge and perceptions about driving, demographic information, and health-related information. They were also informed of the study duration and amount of compensation. Additionally, participants with ADHD were told that they needed to provide the contact information of a friend or family member (observer) that we would then contact and ask to complete health-related surveys about the participant. A reminder email and phone call were used two days prior to each scheduled study visit. Participants with ADHD were reminded that, because of safety concerns, they needed to be dropped off and picked up by a friend or family member. This could be the same observer who completed the surveys. They were also reminded to not take their ADHD medication on the day of the study visits and that they needed to bring the medication bottle with them. Fig. 1 lists the steps of study procedures for each medication condition and study visit scheduling option for ADHD participants. A lab room was used for consenting, confirming eligibility criteria, and completing surveys and interviews. The driving simulator located in another lab room was used for the driving portion of the study visits. When participants entered the driving simulator room, they were immediately informed of the safety features. Once they sat in the driver's seat, the experimenters explained the dashboard, center console, and the layout of the simulator. Participants were instructed that they would see and interact with pre-determined traffic events in the virtual driving environment. The driving environment included rural and urban settings with corresponding buildings, trees, curves, intersections, and construction zones. They were asked to wear the seatbelt, obey traffic laws and drive in the simulator as they normally do in a real car on the road. Participants could adjust the seating to their preference; during this time, the experimenters put up curtains, started video recording, and loaded a practice drive that was used to familiarize participants with the dynamics of the simulator.Fig. 1 Steps of study procedures of regular and delayed medication conditions for ADHD participants. Fig. 1 Once the practice drive had ended and the participants' questions were answered, they proceeded to the main experimental drives. The order of the four experimental drives was randomized, and prior to starting each drive, participants were reminded to drive normally and legally when interacting with traffic. A short break was offered after the completion of each drive. Upon completing the four experimental drives, participants were asked to complete the post-drive survey and, if applicable, other self-reports. They were then debriefed, thanked, and compensated for their time. For non-ADHD participants, the study procedures included consenting, completing MSHSF, going to the driving simulator, performing one practice drive and four experimental drives (following pre-determined, randomized order), and completing the post-drive survey and CAARS-self report. There was only one study visit and the duration was 2 h. 2.7 Analytic plan 2.7.1 Sample size justification An a priori power analysis was performed using G*Power software. Effect size of 0.38 was calculated using average speed during simulator driving from a previous study [19]: 16.7 and 14.9 mph for individuals with and without ADHD, respectively (pooled standard deviation = 2.35). Sample size calculation was based on an alpha level of 0.05, a power level of 0.80, an effect size of 0.38, 2 groups, and 4 repetitions. Based on these parameters, a sample size of 36 participants (18 in each group) was needed to achieve the estimated power level. 2.7.2 Analysis for current paper In this paper, demographic, clinical, and psychological characteristics for participants with and without ADHD were reported. Only one participant chose the option 2 study procedure; therefore, we did not conduct a sensitivity test on comparing the two options of study procedure. All data were analyzed using SPSS for Windows, version 24. All p-values were two-sided. Between-group differences were evaluated using Student's t, Chi-Square, and Fisher's exact tests as appropriate. For interval variables, Levene's test for equality of variances were conducted. All of the variables, except DBS Exaggerated subscale, had a p-value of > .05, suggesting that the variances of the population from which the two groups were drawn are the same. 2.7.3 Analysis of primary outcomes The analysis will include average speed as well as other vehicle control variables (such as longitudinal braking profiles and lateral lane deviation) and incidents of crash and near-crash (defined as being within 2 m to another moving or parked vehicle) between the two groups. Results from the crash and near-crash comparisons were reported elsewhere [4]. In addition, speeding profiles during segments of driving (such as straight and curved road segments, gap negotiation with other vehicles) will be compared between the two medication conditions. Several machine learning and data mining techniques are being used. Specifically, data is being examined at three levels of granularity: (1) whole drive, (2) drive segments (segments of 5–10 s), and (3) full resolution of the collected data (60 Hz). At each granularity, we employ two types of machine learning approaches. First, we employ supervised learning methods such as decision tree learning algorithms [36], support-vector machines (SVMs) [21], and deep neural networks [25] to identify signals in the data that are predictive of ADHD symptoms. At the full resolution time series, we also use time series data mining methods to contrast behavioral patterns during challenging maneuvers, such as lane switching, and the results were reported elsewhere [28]. Second, we use unsupervised learning techniques such as k-means [25] and visualization techniques such as t-SNE [29] to identify regularities in the data and form hypotheses that can later be tested. Moreover, some machine learning approaches can only be applied to the lower levels of granularity (segments and full resolution time series), since considering whole drives as the unit of study results in too small a dataset to detect any interesting trends. Additionally, we have also been using the collected data to develop models of driving behaviors via learning from observation using custom-designed algorithms, such as those reported in our previous work [35,40]. The resulting driving agents with specific characteristics (un-medicated ADHD, medicated ADHD, and control/baseline non-ADHD) are compared in their performance to the participant data to identify discrepancies. 2.7.4 Analysis of secondary outcomes Scores on CAARS, SSKT, BSSS, and DBS as well as post-drive surveys will be used as covariates to further explore differences between groups and medication conditions. 3 Results Recruitment outcome and characteristics of the participants are presented in this section. Results of primary and secondary outcomes will be included in future publications. 3.1 Recruitment flow Fig. 2, Fig. 3, for ADHD and non-ADHD participants, respectively, show the participants' recruitment process and confirmation of study eligibility criteria. Recruitment began in March 2017 and was completed in November 2017.Fig. 2 Enrollment of ADHD participants. Fig. 2Fig. 3 Enrollment of non-ADHD participants. Fig. 3 3.2 Participant characteristics 3.2.1 Demographic characteristics Table 2 reports the demographic information of the participants who met study eligibility criteria. Participants with and without ADHD were similar across all demographic characteristics, except that there were more participants who self-identified being White in the ADHD group.Table 2 Demographic characteristics of participants with and without ADHD. Table 2Data are mean ± SD or n Participants with self-reported ADHD status (n = 21) Participants with self-reported non-ADHD status (n = 17) p-value ADHD vs. non-ADHD Age (years) 21.2 ± 1.5 20.7 ± 2.0 .351 Gender (female) 10 10 .532 Race/Ethnicity Asian 2 5 .207 Black 1 3 .307 White 17 7 .018 Other 3 0 .238 Hispanic 3 5 .426 Years of driving experience 4.5 ± 2.0 3.6 ± 1.6 .147 Driving history (# of participants who had …) License suspended or revoked 2 1 1.000 Fender benders 15 10 .228 Stopped by police 12 6 .322 Traffic tickets 8 5 .631 At-fault crashes 1 1 1.000 Miles driven per week 126.4 ± 136.7 83.5 ± 105.2 .295 Education High school 0 3 .099 Some college 15 8 Associate degree 1 3 Bachelor degree 5 3 Employment status Unemployed 9 4 .227 Part-time 9 12 Full-time 3 1 Student status Not a student 2 2 .321 Part-time 5 1 Full-time 14 14 Note: Significant p-value in bold. Chi-square was used for White, Fender benders, Stopped by police, Traffic tickets, Education, Employment, and Student status. Fisher's Exact Test was used for Gender, Race/ethnicity, License suspended, and At-fault crashes. Student's T was used for Age, Years of driving, and Miles driven. 3.2.2 Clinical characteristics The CAARS scores and ADHD history questionnaire for participants with and without ADHD are listed in Table 3. Participants with ADHD (self-report) had significantly higher T-scores of CAARS inattentive and hyperactivity/impulsive symptoms than participants without ADHD.Table 3 Clinical characteristics of participants with and without ADHD. Table 3Data are mean ± SD Participants with self-reported ADHD status (n = 21) Participants with self-reported non-ADHD status (n = 17) p-value ADHD vs. non-ADHD Self-report Observer report CAARS – Inattentive Symptoms 72.6 ± 12.2 61.9 ± 8.3 53.6 ± 9.4 <.001 CAARS – Hyperactivity/Impulsive Symptoms 66.5 ± 14.3 57.3 ± 10.4 50.8 ± 10.8 <.001 History questionnaire – level of impairment (0 = not at all, 3 = very much) School 2.7 ± .6 2.3 ± .7 Did not collect Work 2.0 ± .9 1.5 ± .8 Family relationship 1.5 ± .9 1.6 ± .9 Social relationship 1.9 ± .7 1.7 ± 1.0 Self-esteem 1.7 ± 1.0 1.6 ± .7 Note: Significant p-value in bold. Student's T was used for comparisons. 3.2.3 Psychosocial characteristics The scores of Safe Speed Knowledge Test (SSKT), Driving Anger Scale (DAS), Brief Sensation Seeking Scale (BSSS), and Driving Behavior Survey (DBS) are listed in Table 4. Participants with and without ADHD had similar scores across the four questionnaires. Individual SSKT items were also compared: #13 and 14 showed significant differences between the two groups, t (36) = 2.80 and 2.27, p = .01 and .03, respectively, indicating that the amount of speed reduction was smaller (less safe speed management strategy) for participants with ADHD than for participants without ADHD.Table 4 Psychosocial characteristics of participants with and without ADHD. Table 4Data are mean ± SD Participants with self-reported ADHD status (n = 21) Participants with self-reported non-ADHD status (n = 17) p-value ADHD vs. non-ADHD Safe Speed Knowledge Test (SSKT) −8.5 ± 2.0 −9.4 ± 2.4 .253 Ranges −18.1 to 0 Ranges −18.2 to −.9 Driving Anger Scale (DAS) 43.5 ± 9.6 39.1 ± 9.4 .168 Brief Sensation Seeking Scale (BSSS) 10.1 ± 4.9 7.5 ± 5.0 .115 Driving Behavior Survey (DBS) Overall 3.5 ± .5 3.4 ± .6 .471 Anxiety-based performance deficits subscale 2.6 ± 1.1 2.5 ± .8 .656 Exaggerated safety/caution behavior subscale 4.8 ± .6 5.0 ± .9 .514 Hostile/aggressive behavior subscale 3.1 ± 1.0 2.7 ± 1.0 .263 Note: Student's T was used for comparisons. 4 Discussion ADHD symptom management among teens and young adults remains a challenging public health issue for families and affected individuals for several reasons: 1) not all individuals with ADHD have the same levels and kinds of impairment, 2) the effect of medication is not uniform, 3) late adolescence and young adulthood is developmentally a period of transition from parent-managed to self-managed treatment strategy, and 4) for the last subset of youth who become independent drivers, ADHD symptoms and impairments generally are manifested in suboptimal vehicle control behaviors [17,20,32,39]. We proposed to design and develop an effective, machine-learning based, in-vehicle monitoring intervention via driver behavior modeling and individuated feedback. As the first step toward the designing of such system, an experimental protocol was developed to allow the collection of reliable and objective driver behavior data and assessments from individuals with and without ADHD as well as before and after intake of medication. As reflected in the results, ADHD and non-ADHD participants were significantly different in clinical characteristics, while being demographically and psychosocially similar. Effective recruitment and enrollment processes (age and zip-code matching), well-defined and standardized eligibility criteria, confirmations of medication, and timely data collection procedures all contributed to achieving this clear distinction between the two groups. We also paid special attention to participant susceptivity to simulator sickness or other factors that might undermine their behaviors during data collection, as reflected by the few participants being terminated or withdrawn. This clear distinction on clinical characteristics is important for the subsequent modeling work. In order to develop reliable and predictive models, we look for similarities among participants as well as dissimilarities across participants within each group. The former will lead to the discoveries of contrastive patterns of unsafe driving behaviors from each group and representative patterns between groups. The latter will lead to establishing ranges of variability from one driver to another. In other words, the datasets from each sample ideally should be as homogeneous as possible while taking into account individual differences and variability as well as sample-based heterogeneity. Having collected distinctive datasets from individuals with and without ADHD via the reported study protocol made a significant step toward achieving the long-term goal of designing an in-vehicle monitoring intervention. Extensive computational modeling work is underway to develop individual-level and group-level predictive models of driver behaviors while using self-reported ADHD assessments as covariates to test and refine the parameters and algorithms. Once the models are developed, further simulator testing and on-road studies will be conducted to evaluate the accuracy and efficacy of the in-vehicle monitoring intervention. Nonetheless, the current study had a few methodological limitations that should be stated. By applying multiple criteria to determine participant eligibility, the evaluable sample size became relatively small; thus, the results should be interpreted with caution. While the eligibility criteria were stringent, they were necessary to target a late adolescent-young adult age group with specifically defined ADHD diagnosis for the study and screen out potential confounding variables such as comorbidity. Relying on self-reported ADHD clinical history, diagnosis, and medication was another limitation. Obtaining a more comprehensive profile of the clinical background (e.g., patient's complete medical records) should be considered for future studies. A third limitation was the use of a driving simulator as an assessment tool. A few participants who met eligibility criteria experienced simulator sickness and had to discontinue the study. Also, even though the simulator used in the study was a high-fidelity, motion-based system, it was still an artificial environment that had the resemblance of a driver's seat and forward roadway scenery. A more naturalistic and ecologically valid testing environment (e.g., on-road naturalistic observations through instrumented vehicles or a test-track environment) should be considered for future studies. Despite these limitations, the current study summarized an experimental protocol that led to successful recruitment and assessments of young adult drivers diagnosed with and without ADHD. These groups differed clinically but were similar demographically and psychosocially, thus providing strong foundation for subsequent behavioral modeling work. The current study served as the first step toward our long-term objective of designing a machine-learning based, in-vehicle monitoring intervention for automatically detecting unsafe driving behaviors for adolescent and young adult drivers diagnosed with ADHD. We envision that this monitoring intervention will be part of a developmentally appropriate treatment plan as the intended users transition from family to self-management of ADHD during adolescence [7]. Acknowledgement This work was supported by the National Science Foundation, grant number #1653624. We thank graduate students Randall Drury and Daniela Barragan and undergraduate students Briana Rudolph and Obinna Ihezue for their efforts on data collection and processing. ==== Refs References 1 Jillani S.A. Kaminer Y. High risk driving in treated and untreated youth with attention deficit hyperactivity disorder: public health implications Adolesc. Psychiatr. 6 2 2016 89 99 2 Barkley R.A. Cox D. A review of driving risks and impairments associated with attention-deficit/hyperactivity disorder and the effects of stimulant medication on driving performance J. Saf. Res. 38 1 2007 113 128 10.1016/j.jsr.2006.09.004 3 Barkley R.A. Murphy K.R. O'Connell T. Connor D.F. Effects of two doses of methylphenidate on simulator driving performance in adults with attention deficit hyperactivity disorder J. Saf. Res. 36 2 2005 121 131 10.1016/j.jsr.2005.01.001 4 Barragan D. Lee Y.-C. Pre-crash driving behaviour of individuals with and without ADHD Presented at the 6th International Conference on Driver Distraction and Inattention, Gothenburg, Sweden 2018 5 Biederman J. Fried R. Hammerness P. Surman C. Mehler B. Petty C.R. Reimer B. The effects of lisdexamfetamine dimesylate on the driving performance of young adults with ADHD: a randomized, double-blind, placebo-controlled study using a validated driving simulator paradigm J. Psychiatr. Res. 46 4 2012 484 491 10.1016/j.jpsychires.2012.01.007 22277301 6 Biederman J. Petty C.R. Clarke A. Lomedico A. Faraone S.V. Predictors of persistent ADHD: an 11-year follow-up study J. Psychiatr. Res. 45 2 2011 150 155 10.1016/j.jpsychires.2010.06.009 20656298 7 Brinkman W.B. Sherman S.N. Zmitrovich A.R. Visscher M.O. Crosby L.E. Phelan K.J. Donovan E.F. In their own words: adolescent views on ADHD and their evolving role managing medication Acad. Pediatr. 12 1 2012 53 61 10.1016/j.acap.2011.10.003 22133501 8 Chang Z. Quinn P.D. Hur K. Gibbons R.D. Sjölander A. Larsson H. D'Onofrio B.M. Association between medication use for attention-deficit/hyperactivity disorder and risk of motor vehicle crashes JAMA Psychiatry 74 6 2017 597 603 10.1001/jamapsychiatry.2017.0659 28492937 9 Clapp J.D. Olsen S.A. Beck J.G. Palyo S.A. Grant D.M. Gudmundsdottir B. Marques L. The driving behavior survey: scale construction and validation J. Anxiety Disord. 25 1 2011 96 105 10.1016/j.janxdis.2010.08.008 20832988 10 Conners C.K. Erhardt D. Sparrow E. Conners' Adult ADHD Rating Scales (CAARS) 1999 Multi-Health Systems, Inc 11 Cox D.J. Davis M. Mikami A.Y. Singh H. Merkel R.L. Burket R. Long-acting Methylphenidate reduces collision rates of young adult drivers with attention-deficit/hyperactivity disorder J. Clin. Psychopharmacol. 32 2 2012 225 10.1097/JCP.0b013e3182496dc5 22367664 12 Curry A.E. Metzger K.B. Pfeiffer M.R. Elliott M.R. Winston F.K. Power T.J. Motor vehicle crash risk among adolescents and young adults with attention-deficit/hyperactivity disorder JAMA Pediatr. 171 8 2017 756 763 10.1001/jamapediatrics.2017.0910 28604931 13 Deffenbacher J.L. Oetting E.R. Lynch R.S. Development of a driving anger scale Psychol. Rep. 74 1 1994 83 91 10.2466/pr0.1994.74.1.83 8153239 14 Epstein J. Johnson D.E. Conners C.K. Conners' Adult ADHD Diagnostic Interview for DSM-iv (CAADID) 2001 Multi-Health Systems, Inc 15 Fried R. Petty C.R. Surman C.B. Reimer B. Aleardi M. Martin J.M. Biederman J. Characterizing impaired driving in adults with attention-deficit/hyperactivity disorder: a controlled study J. Clin. Psychiatr. 67 4 2006 567 574 16 Froehlich T.E. Lanphear B.P. Epstein J.N. Barbaresi W.J. Katusic S.K. Kahn R.S. Prevalence, recognition, and treatment of attention-deficit/hyperactivity disorder in a national sample of US children Arch. Pediatr. Adolesc. Med. 161 9 2007 857 864 10.1001/archpedi.161.9.857 17768285 17 Fuermaier A.B.M. Tucha L. Evans B.L. Koerts J. Waard D. de Brookhuis K. Tucha O. Driving and attention deficit hyperactivity disorder J. Neural. Transm. 124 1 2017 55 67 10.1007/s00702-015-1465-6 26419597 18 Goodman D.W. Sustained treatment effect in attention-deficit/hyperactivity disorder: focus on long-term placebo-controlled randomized maintenance withdrawal and open-label studies Therapeut. Clin. Risk Manag. 9 2013 121 130 10.2147/TCRM.S30762 19 Groom M.J. van Loon E. Daley D. Chapman P. Hollis C. Driving behaviour in adults with attention deficit/hyperactivity disorder BMC Psychiatr. 15 2015 175 10.1186/s12888-015-0566-y 20 Gualtieri C.T. Johnson L.G. Medications do not necessarily normalize cognition in ADHD patients J. Atten. Disord. 11 4 2008 459 469 10.1177/1087054707305314 17934180 21 Hearst M.A. Dumais S.T. Osuna E. Platt J. Scholkopf B. Support vector machines IEEE Intell. Syst. Appl. 13 4 1998 18 28 10.1109/5254.708428 22 Hoffman R.B. Molino J.A. Inman V.W. Driving simulator sickness management at Turner-Fairbank Highway rResearch Center Presented at the Driving Simulation Conference North America, Dearborn, Michigan 2003 23 Kennedy R.S. Lane N.E. Berbaum K.S. Lilienthal M.G. Simulator Sickness Questionnaire: an enhanced method for quantifying simulator sickness Int. J. Aviat. Psychol. 3 3 1993 203 220 10.1207/s15327108ijap0303_3 24 Lange H. Buse J. Bender S. Siegert J. Knopf H. Roessner V. Accident proneness in children and adolescents affected by ADHD and the impact of medication J. Atten. Disord. 20 6 2016 501 509 10.1177/1087054713518237 24470540 25 LeCun Y. Bengio Y. Hinton G. Deep learning Nature 521 7553 2015 436 444 10.1038/nature14539 26017442 26 Lee Y.-C. Winston F.K. Stress induction techniques in a driving simulator and reactions from newly licensed drivers Transport. Res. F Traffic Psychol. Behav. 42 Part 1 2016 44 55 10.1016/j.trf.2016.06.019 27 Legree P.J. Heffner T.S. Psotka J. Martin D.E. Medsker G.J. Traffic crash involvement: experiential driving knowledge and stressful contextual antecedents J. Appl. Psychol. 88 1 2003 15 26 10.1037/0021-9010.88.1.15 12675391 28 Li Q. Zhao L. Lee Y.-C. Lin J. Contrast pattern mining for dynamic feature dependencies in paired multivariate time series CIKM 2018: International Conference on Information and Knowledge Management, Turin, Italy 2018 (Under review) 29 Maaten L. van der Hinton G. Visualizing data using t-SNE J. Mach. Learn. Res. 9 Nov 2008 2579 2605 30 McDaniel S.R. Mahan J.E. An examination of the ImpSS scale as a valid and reliable alternative to the SSS-V in optimum stimulation level research Pers. Indiv. Differ. 44 7 2008 1528 1538 10.1016/j.paid.2008.01.009 31 McDonald C.C. Kandadai V. Loeb H. Seacrist T.S. Lee Y.-C. Winston Z. Winston F.K. Simulated Driving Assessment (SDA) for teen drivers: results from a validation study Inj. Prev. 21 3 2015 145 152 10.1136/injuryprev-2014-041480 25740939 32 Montano C.B. Young J. Discontinuity in the transition from pediatric to adult health care for patients with Attention–Deficit/Hyperactivity Disorder PGM (Postgrad. Med.) 124 5 2012 23 32 10.3810/pgm.2012.09.2591 33 National Highway Traffic Safety Administration Traffic Safety Facts - Young Drivers, 2015 Data 2017 US Department of Transportation Retrieved from: https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812363.pdf 34 Ontañón S. Lee Y.-C. Snodgrass S. Bonfiglio D. Winston F.K. McDonald C. Gonzalez A.J. Case-based prediction of teen driver behavior and skill Case-based Reasoning Research and Development 2014 Springer Cham 375 389 10.1007/978-3-319-11209-1_27 35 Ontañón S. Lee Y.-C. Snodgrass S. Winston F.K. Gonzalez A.J. Learning to predict driver behavior from observation Spring Symposium of the Association for the Advancement of Artificial Intelligence, Palo Alto, California 2017 36 Quinlan J.R. C4.5: Programs for Machine Learning 2014 Elsevier 37 Santosh P.J. Sattar S. Canagaratnam M. Efficacy and tolerability of pharmacotherapies for attention-deficit hyperactivity disorder in adults CNS Drugs 25 9 2011 737 763 10.2165/11593070-000000000-00000 21870887 38 Thomas R. Sanders S. Doust J. Beller E. Glasziou P. Prevalence of attention-deficit/hyperactivity disorder: a systematic review and meta-analysis Pediatrics 135 4 2015 e994 e1001 10.1542/peds.2014-3482 25733754 39 Vaa T. ADHD and relative risk of accidents in road traffic: a meta-analysis Accid. Anal. Prev. 62 Supplement C 2014 415 425 10.1016/j.aap.2013.10.003 24238842 40 Wong J. Hastings L. Negy K. Gonzalez A.J. Ontañón S. Lee Y.-C. Machine learning from observation to detect abnormal driving behavior in humans Proceedings of the 31st International Florida Artificial Intelligence Research Society Conference. Melbourne, Florida 2018 41 Zuckerman M. Kuhlman D.M. Joireman J. Teta P. Kraft M. A comparison of three structural models for personality: the big three, the big five, and the alternative five J. Pers. Soc. Psychol. 65 4 1993 757 768