Motor Vehicle Crash Risk Among Adolescents and Young Adults With Attention-Deficit/Hyperactivity Disorder

Study Design and Sample

Individuals for this retrospective cohort study were identified from the 6 New Jersey CHOP primary care practices within CHOP’s network, which encompasses more than 50 locations in Pennsylvania and New Jersey. Clinicians manage all aspects of clinical care using a linked EHR system, which was fully implemented in New Jersey primary care by 2005. We queried CHOP’s EHR database to select individuals who (1) were born from 1987 to 1997; (2) were patients at a New Jersey primary care practice; and (3) had a CHOP network visit (to any location) as a New Jersey resident within 4 years of becoming eligible for their learner’s permit at age 16 years and maintained a New Jersey address through their last network visit to establish New Jersey residency. We identified 19 588 individuals. We excluded 73 individuals with an intellectual disability given that this essentially removed them from the pool of eligible drivers; intellectual disability was defined as the presence of an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis code beginning with 317, 318, or 319 either at a CHOP visit or on the list of known conditions (ie, problem list). Additionally, we limited our population to individuals who had 2 or more primary care visits to minimize ADHD misclassification and those who had their last primary care visit at age 12 years or older to ensure individuals were seen by a primary care clinician at an age old enough to confirm ADHD status (Diagnostic and Statistical Manual of Mental Disorders [Fifth Edition] criteria call for ADHD symptom onset by age 12 years). The study cohort included 18 522 primary care patients (Figure 1). This study was approved by CHOP’s institutional review board, and 1 of the authors (K.B.M.) analyzed the data. Consent was waived by the institutional review board because the study included linking of 2 databases, with a total of over 10 million records.

ADHD Classification

Individuals were classified as having ADHD if their EHR indicated an ICD-9-CM code beginning with 314 either at a CHOP network visit or on their problem list; 2503 individuals with ADHD were identified. We formally validated this algorithm using manual review of EHRs; estimated sensitivity was 0.96, and specificity was 0.98. For 2021 of 2503 (80.7%) of individuals, we found an independent source in the EHR confirming ADHD status (eg, letters, medication or test results). Of 2503 individuals, 2357 with ADHD (94.2%) were classified based on visit-level ICD-9-CM codes.

Licensing and Crash Data

We obtained records for all individuals who received a New Jersey driver’s license through December 2014 from the New Jersey Motor Vehicle Commission’s Licensing Database, including exact dates of birth and licensure. The minimum licensure age in New Jersey is 17 years. Detailed data on police-reported crashes occurring in New Jersey from 2004 to 2014 were obtained from the New Jersey Department of Transportation’s Crash Database. These databases were linked in a previous study. In total, 98% of New Jersey drivers who were involved in crashes were linked to a licensing record. The linkage process was validated, and we concluded that the process was conducted with high quality.

Linkage of EHR and Driving Data

Data from the EHR for a larger cohort of CHOP patients who were New Jersey residents were individually linked with licensing-crash data via a hierarchical deterministic linkage. Data elements included date of birth, full name, sex, zip code, city, and street number of residence. Based on a hand review of 3348 records, we estimated the true match rate (ie, true matches/original matches) to be 99.95%. The false nonmatch rate (ie, true matches that were not found), assessed by searching for the 5 most likely potential matches for a sample of 474 unmatched individuals, was estimated to be 1.5%.

Variables

Main outcome measures were time to acquisition of a New Jersey driver’s license and first involvement as a driver in a New Jersey police-reported crash. A crash is reportable if it results in an injury or more than $500 in property damage.

Sex, date of birth, race/ethnicity, and insurance payor at last visit were determined from the EHR. We used ICD-9-CM codes to determine the presence of seizure disorder (345x) and disruptive behavior disorders (conduct disorder [312x] and oppositional defiant disorder [313.81]). For each participant with ADHD, we determined from the EHR whether an ADHD medication was prescribed by a CHOP clinician within 12 months of licensure (a proxy for ADHD severity) and within 30 days of licensure (given that ADHD stimulant medications prescriptions cover 30 days and cannot be refilled). We used the 2010 US Census Gazetteer Files and 2007 to 2011 American Community Survey data to categorize residence zip code at last visit into quintiles of population density and median household income, respectively.

Statistical Analysis

Analyses were restricted to 2479 individuals with ADHD and 15 865 individuals without ADHD who had at least 1 full month of follow-up after becoming age-eligible for licensure (Figure 1). Kaplan-Meier survival curves were used to estimate cumulative probability of licensure and crash involvement over time; the log-rank test was used to compare differences. Multivariable Cox regression models were used to estimate adjusted hazard ratios (aHRs). The proportionality assumption was assessed using Kolmogorov-type supremum tests and by considering the interaction of ADHD status with follow-up time. The association between ADHD and time to licensure was observed to vary temporally in males (supremum test: P < .001), but not females (P = .10); thus, to account for time dependence among males only, final Cox models that used linear and quadratic terms determined to provide the best model fit by likelihood ratio tests. The association between ADHD and time to first crash did not vary over time (P = .40). Adjusted models include potential confounders chosen a priori based on known or suspected associations with ADHD (or its clinical diagnosis) and driving outcomes including sex, race/ethnicity, insurance payor, disruptive behavior and seizure disorders, licensure age (for crash outcome), primary care practice, birth year, and zip code–level income and population density. Multiplicative interaction was assessed using likelihood ratio tests. We examined the association between medication and crashes among the 1036 licensed drivers with ADHD who had a CHOP network visit in the year before licensure. We compared crash hazards in the 2 years after licensure for 3 groups: those not prescribed ADHD medication within the year before licensure; those prescribed medication in the year before licensure but not in the 30 days before licensure; and those prescribed medication within 30 days of licensure. Estimates were considered significant at the .05 level, and hypothesis tests were 2-sided. Analyses were conducted in SAS, version 9.3 (SAS Institute Inc).

Study Population

Individuals had a median of 19 primary care visits (interquartile range [IQR], 10-33), had their last visit at a median age of 18.0 years (IQR, 16.0-18.9), and were a median age of 22.2 years (IQR, 19.7-24.8) at the end of the study period (Table 1). Individuals with ADHD had a median of 2 primary care visits (IQR, 1-5) that included an ADHD-specific ICD-9-CM code. Table 1 shows comparisons of relevant characteristics.

Licensing Rates

Overall, 222 of 691 females with ADHD (32.1%) and 3874 of 8323 females without ADHD (46.5%) were licensed in their first month of eligibility, and 311 of 391 (79.5%) and 4502 of 5210 (86.4%), respectively, were licensed by age 21 years. Table 2 shows the unadjusted cumulative probability of licensure by select ages. Licensing rates among males were lower in ADHD and non-ADHD groups but followed a similar pattern. Unadjusted curves of time to licensure by ADHD status and sex are shown in Figure 2. Among males, adjusted Cox models indicated that differences in licensure between ADHD and non-ADHD groups diminished over time (by age 17.5 years: aHR, 0.65; 95% CI, 0.61-0.70 vs by age 19 years: aHR, 1.00; 95% CI, 0.87-1.14). Among females, the difference remained constant over time (aHR, 0.64; 95% CI, 0.58-0.70). Among individuals ultimately licensed, females with ADHD were older at licensure than females without ADHD (median age, 17.2 years; IQR, 17.0-17.8 years vs 17.0 years; IQR, 17.0-17.5 years; P < .001) and males with ADHD were older than males without ADHD (median age, 17.4 years; IQR, 17.0-18.0 years vs 17.1 years; IQR, 17.0-17.6 years; P < .001).

Crash Rates

Crash analyses were limited to those licensed: 1785 individuals with ADHD and 13 221 without ADHD. Of these, 764 with ADHD (42.8%) and 4715 without ADHD (35.7%) crashed by the end of the study period. Unadjusted cumulative probabilities of crash involvement by select postlicensure months are shown in Table 2. Figure 3 shows the proportion of licensed drivers who remained crash-free over time by ADHD status and sex; both males and females with ADHD were more likely to crash earlier and at higher frequencies than those without ADHD. In adjusted Cox models, the overall aHR for crash risk among adolescent drivers with ADHD was 1.36 (95% CI, 1.25-1.48). Further, there was no evidence that the ADHD-crash relationship varied on the multiplicative scale by licensing age (χ² = 0.26; P = .97) or sex (χ² = 2.05; P = .15). The crash hazard among males with ADHD was 1.42 times higher (95% CI, 1.28-1.56) than among males without ADHD and 1.25 times higher (95% CI, 1.08-1.45) for females with ADHD than females without ADHD.

Of 1063 licensed individuals with ADHD and a CHOP network visit in the year before licensure, 267 (25.1%) were prescribed ADHD medication in that year and 129 (12.1%) were prescribed ADHD medication by a CHOP clinician in the 30 days before licensure. There were no significant differences in 2-year postlicensure risk among the 3 groups compared: medication in the previous year but not just before licensure vs none (aHR, 0.95; 95% CI, 0.64-1.41); medication just before licensure vs none (aHR, 1.17; 95% CI, 0.80-1.71); and medication just before licensure vs in the previous year but not just before licensure (aHR, 1.24; 95% CI, 0.78-1.97).

We conducted sensitivity analyses by reclassifying 482 individuals with ADHD who did not have independent confirming evidence of ADHD into the non-ADHD group. Results did not meaningfully change; estimated aHRs for crash were attenuated by less than 1%. We also reweighted the time variable in Cox regression models to assess potential bias owing to differences in driving exposure (mileage) and estimated that individuals with ADHD would have to drive a mean of 48% more miles than those without ADHD before the 95% CIs of aHRs included 1.0.

Limitations

There are several limitations of this study. Diagnoses relied on assessment by primary care clinicians rather than rigorous testing using criterion standard as set by Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition); however, we validated ADHD diagnosis codes with high sensitivity and specificity, and most individuals continued to have primary care visits after becoming eligible for licensure. Given that our sample includes individuals who were identified before adolescence, a certain percentage may not have impairments owing to ADHD symptoms by the time of licensure; thus, this sample more appropriately reflects individuals with a lifetime history of ADHD. Additionally, we did not directly measure driving exposure. Young adults with ADHD reported driving more in 2 of 3 previous studies that assessed mileage. However, these assessments involved surveys of usual driving patterns at the time of survey, potentially years after crash occurrence, or total mileage driven in the 3 to 5 years since licensure and may be subject to positive illusory bias. Our analyses inherently accounted for time since licensure, which has been used as a proxy for exposure and itself is a critically important variable given rapid declines in crash rates over the first few years of driving. Adolescents and young adults with ADHD may also be less likely to attend college, and hence move out of state after high school, than their counterparts without ADHD; however, significant associations were constant over time and include the period just after licensure when most individuals were still in high school. Finally, there may be limits to generalizability. New Jersey has the oldest licensing age in the United States and is highly urbanized. Further, the prevalence of ADHD in our primary care cohort (14%) was somewhat higher than US-based estimates, perhaps reflecting parental decisions to seek care within the CHOP network.