Trajectories of Risky Driving Among Emerging Adults with Their Mental and Psychosomatic Health Predictors in the 12^th Grade

Abstract

Objective:

To identify trajectory classes of risky driving among emerging adults and examine predictive associations of depressive and psychosomatic symptoms in the 12^th grade with the identified trajectory classes.

Methods:

Data were from the last year in high school (12^th-Grade - Wave 3 [W3]) and years 1-4 after high school (Waves 4-7 [W4-7]) of the NEXT Generation Health Study, a nationally representative study starting with 10^th grade (2009-2010). We measured risky driving with the 21-item Checkpoints Self-Reported Risky Driving Scale (C-RDS). Using C-RDS data from W3-7, the latent class growth modeling (LCGM) was used to identify risky driving trajectory classes. Independent variables were W3 depressive symptoms and W3 psychosomatic symptoms. Covariates included family affluence and urbanicity. The LCGM was conducted with SAS PROC Traj. The multinomial logistic regressions were used to examine the associations between the trajectory classes and independent variables, taking complex survey sampling features into account.

Results:

Three risky driving trajectories were identified: low (N=583, 21.43%, weighted and hereafter), medium (N=1423, 59.22%), and high (N=389, 19.35%) risky driving classes. Compared to the low risky driving class, one unit increase in W3 depressive symptoms was significantly associated with a higher likelihood of belonging to the medium (adjusted odds ratio [AOR]=1.04, 95% CI 1.01, 1.07) and the high (AOR=1.05, 95% CI 1.02, 1.08) risky driving classes, respectfully, when controlling for the covariates. Likewise, compared to the low risky driving class, one unit increase in W3 psychosomatic symptoms was significantly associated with a higher likelihood of belonging to the medium (AOR=1.06, 95% CI 1.00, 1.13) and the high (AOR=1.10, 95% CI 1.04, 1.16) risky driving classes, respectively, when controlling for the covariates.

Conclusions:

High school students with depressive and psychosomatic symptoms were at higher risk of engaging in risky driving in the immediate years after leaving high school. These findings suggest that prevention programs that incorporate screening, referral to treatment, and treatment of mental and psychosomatic symptoms in high school may be important opportunities to reduce risky driving among youth as they transition from adolescence to emerging adulthood.

INTRODUCTION

Motor vehicle crashes are the leading cause of death among the U.S. adolescents (Centers for Disease Control and Prevention (CDC) 2020). In 2018, about 2,500 adolescents aged 13-19 years were killed in motor vehicle crashes (Centers for Disease Control and Prevention (CDC) National Center for Injury Prevention and Control (NCIPC) 2020). In addition, 285,000 adolescents were treated in emergency rooms resulting from motor vehicle crashes (Centers for Disease Control and Prevention (CDC) National Center for Injury Prevention and Control (NCIPC) 2020). Fatal motor vehicle crash rates remain unacceptably high among adolescent drivers. When compared to adult drivers, typically, young drivers are more inexperienced, drive faster, maintain shorter headways, make more errors in driver-based decisions, and are more likely to underestimate hazardous driving situations (Simmons-Morton et al. 2013). The inexperience driving coupled with nighttime and weekend driving, not using seatbelts, and distracted driving may double or triple the risk of motor vehicle crash among adolescent drivers (Simmons-Morton et al. 2013).

Identifying predictive factors for risky driving behaviors may help develop prevention strategies to reduce risky driving among young drivers. Previous study has shown that a low level of parental monitoring and increases in substance use characterized the developmental trajectories of young adult risky drivers (Bingham et al. 2004). Further, lacking parental monitoring may contribute to depression problems in adolescents (Jacobson et al. 2000). A more recent study indicated that alcohol and illicit drug use were associated with increased odds of severe mental health problems among adolescents (Li et al. 2021). Therefore, mental health may be directly or indirectly linked to risky driving behaviors in young drivers. Additionally, evidence has shown that mental health (i.e., depression) and psychosomatic symptoms (i.e., headache) were significantly associated with risky behaviors (i.e., bullying) and unhealthy behaviors (i.e., tobacco use) (Fekkes et al. 2004; Brooks et al. 2002), which indicates depressive and psychosomatic symptoms may link to risky driving behaviors as well. A recent study has reported the cross-sectional associations of depressive and psychosomatic symptoms (i.e., headache, abdominal pain, and musculoskeletal pain) with risky driving among young drivers up to four years after leaving high school (Li et al. 2020). Although the cross-sectional associations have been demonstrated, it remains unclear if depressive and psychosomatic symptoms in high school influences risky driving after high school.

The trajectory patterns of depression from adolescence to adulthood show that depressive symptoms developed in early adolescence reached the peak during late adolescence (Lewinsohn et al. 2003; Schulenberg et al. 2006) and are relatively stable in early adulthood through the second decade of life (Tram et al. 2006). However, the young adult may expect to have stable mental health symptoms because early adulthood is a time when they are starting to understand what their lives may be and what life objectives they chose to pursue. They may be feeling better about themselves and experiencing less emotional turmoil as their psychosocial status becomes mature. Therefore, the mental health symptoms in late adolescence relative to early adulthood may have more severe impacts on their later behavior like risky driving. Similar to the trend of depression from adolescence to adulthood, a systematic review and meta-analysis study has shown that late adolescents encounter more psychosomatic symptoms compared to early adulthood (Potrebny et al. 2017). The externalizing problems seem to be relatively stable with the improvement of health and well-being in the emerging adults (Potrebny et al. 2017). Hence, psychosomatic symptoms during late adolescence compared to early adulthood may contribute more to later risky behaviors like risky driving.

Taken together, depressive and psychosomatic symptoms in high school may affect later risky driving patterns after high school. Therefore, understanding how adolescents’ mental and psychosomatic symptoms affect later risky driving is needed for health professionals to improve adolescent mental health care to reduce risky driving and fatal motor vehicle crash risks. The purpose of this study was to identify the trajectory classes of risky driving during five consecutive years that include the last year of high school (12^th grade) through the first four years after high school and to examine the predictive associations of 12^th-grade depressive and psychosomatic symptoms with the identified trajectory classes of risky driving.

METHODS

Sampling

We used the data from the last year in high school (12^th-grade - Wave 3 [W3]) and years 1-4 after high school (Waves 4-7 [W4-7]) of the NEXT Generation Health Study (NEXT), a nationally representative study starting with 10^th grade (2009-2010). Previous studies have described the sampling strategy for the NEXT study (Li et al. 2013). To examine race/ethnicity disparities, African Americans were oversampled to obtain a large enough sample (N=687) for better population estimates. The NEXT study administered the surveys for each wave in the spring semester. There were 2,785 participants who enrolled in the NEXT study in total. Among the 2,785 participants, 91% (N=2,524), 78% (N=2,177), 79% (N=2,2020), 84% (N=2,306), and 83% (N=2,323) completed the survey from W3 to W7. The study consent was obtained from participants if they were 18 years or older or from participants’ legal guardians if the participants were under 18 years. The study protocol was approved by the Institutional Review Board of the Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Trajectory variable

Checkpoints Risky Driving Scale (C-RDS, W3-7)

We used twenty-one items to assess risky driving behaviors (e.g., On how many days in the last 30 days have you exceed the speed limit in residential or school zones? Or purposely tailgated or followed another vehicle very closely?). The 21 items were derived from a validated C-RDS scale (Simmons-Morton et al. 2013). The internal consistency of the C-RDS scale was excellent as Cronbach’s alphas were 0.92, 0.93, 0.93, 0.92, and 0.92 from W3-7, respectively. We dichotomized the C-RDS responses based on each of the 21 questions (1=≥1 day vs. 0=no day) and then summed up the 21 dichotomous variables, with possible counts ranging from 0 to 21. A high C-RDS score indicates higher risky driving involvement.

Independent variables

Depressive Symptoms (W3)

We used the Pediatric Patient Reported Outcome Measurement Information System (PROMIS) depressive symptom scales to measure depressive symptoms (Pilkonis et al. 2011). To measure depressive symptoms, we asked participants how often they felt about each of following 8 items if these questions were true, including: (1) “I felt like I couldn’t do anything right”, (2) “I felt everything in my life went wrong”, (3) “I felt unhappy”, (4) “I felt lonely”, (5) “I felt sad”, (6) “I felt alone”, (7) “I thought that my life was bad”, and (8) “ I could not stop feeling sad” over the last 7 days, using a 5-point Likert scale, with responses ranging from (1) “Never” to (5) “Almost always”. The internal consistency of the PROMIS scales was excellent as the value of Cronbach’s alpha was 0.96 in the 12^th grade. We converted the PROMIS scores into T-scores according to the distribution of scores in the general U.S. pediatric population (mean=50, SD=10). In most cases, the score 50 equals the mean in the U.S. general population. This metric, the T-score (mean=50, SD=10), has been linked to many other conventional measures. Despite other conventional measures used to assess depression (HealthMeasures 2021), this T-score could be calculated to ensure the comparability of depression across different studies.

Psychosomatic Symptoms (W3)

We asked participants how often have they experienced any following symptoms: headache, stomachache, backache, feeling low, irritability, or bad temper, feeling nervous, difficulties in getting to sleep, and feeling dizzy in the last half year, with possible 5 responses ranging from (1) “Rarely or never” to (5) “About every day” (Currie et al. 2008). We summed up these 8 items with possible range from 8 to 40 to indicate psychosomatic symptoms as a continuous variable. The internal consistency of these 8 items was good as the Cronbach’s alpha was 0.84.

Driving Licensure

We identified driving licensure by asking participants if they had a driving license that allows them to drive independently and unsupervised. We limited the analysis sample to a subsample of participants with a full independent driving licensure.

Demographic Variables and Covariates

We collected participants’ sex, race/ethnicity, parental education level, family affluence, and urbanicity at the baseline visit (second year high school, W1). The biological sex of participants was female or male. The participants’ race/ethnicity was Latino, African Americans, Whites, or Others including Asian, American Indian or Alaska Native, and Native Hawaiian or Other Pacific Islanders. We defined the highest education attainment of participants’ parents as: 1) Less than a high school diploma, a high school diploma or General Educational Diploma; 2) Some college, technical school or associate degree; or 3) Bachelor or graduate degree. We assessed family socioeconomic status using the Family Affluence Scale (Currie et al. 2008). First, we asked participants the number of cars owned, the number of computers owned, whether participants had their own bedroom, and number of family vacations in the last 12 months. We then categorized participants’ family socioeconomic status as low, moderate, or high affluence. Urbanicity (suburban, rural, vs. urban derived from participants’ W1 school location based on the seven National Center for Education Statistics categories).

Statistical analyses

We used Latent Class Growth Modelling (LCGM) to identify a set of discrete, mutually exclusive latent classes of participants based on their responses to risky driving measured by C-RDS questions. The goodness-of-fit of LCGM model was assessed by the Bayesian Information Criterion (BIC) value (Jones et al. 2001). The fit of each nested model is compared using the estimate of the log Bayes factor [2log_e (B10) ≈ 2 (BIC)] (Jones et al. 2001). The log Bayes Factor values greater than 10 are interpreted as very strong evidence of a model fit (Jones et al. 2001). The estimate is approximately equal to two times the difference in the BIC values for the two models being compared. To ensure parsimony, non-significant cubic and quadratic terms were removed from trajectories in a testing model, but linear parameters were retained irrespective of significance. Once non-significant terms were removed, each model was retested yielding a new BIC value. The addition of trajectories continues until there was no significant improvement in model fit compared to the previously tested model. Average posterior probability of 0.70 for the within-group membership was used to determine internal reliability (Jones et al. 2001). The model-fit criteria and average posterior probabilities were used to identify the number of risky driving trajectory classes.

Bivariate association between demographic variables and trajectory classes were tested using Rao-Scott χ² test. Covariates selected into the logistic regressions were based on the χ² test at the significant level of .10 to guard against the type II error. Family affluence and urbanicity were included in the adjusted models.

We conducted multinomial logistic regressions to examine the associations between C-RDS trajectory classes and independent variables (depressive and psychosomatic symptoms, respectively).

Trajectory classes were identified with SAS PROC Traj procedure using the censored normal mixture method as risky driving was measured by scale, with possible counts ranging from 0 to 21 (Jones et al. 2001). Other analyses were performed in SAS software version 9.4 (SAS Institute, Cary, NC) taking complex survey sampling features into account (Agnelli 2014). The statistical significance level was set at p = .05 for the analyses.

RESULTS

One-, two-, three-, and four-class models were estimated from the latent class growth modeling. All model fit statistics of the three-class model were better than the two-class model and the four-class model, respectively. The absolute BIC value (1277.44) of the two-class model and the one-class model was higher than the absolute BIC value (250.24) of the three-class model and the two-class model. To ensure parsimony, non-significant cubic and quadratic terms were removed from trajectories in the testing four-class model while linear parameters are retained irrespective of significance, resulting in going back to the one-class model with one linear parameter. The absolute BIC value (1540.82) of the four-class model and the three-class model was higher than the absolute BIC value (250.24) of the three-class model and the two-class model. Therefore, the three-class model (Figure 1) was the best fit for the analyses. The three classes were low risky driving (low C-RDS scale five consecutive years since 12^th grade), medium risky driving (medium C-RDS scale five consecutive years since 12^th grade), and high risky driving (high C-RDS scale five consecutive years since 12^th grade). Among participants, 583 (21.43%, weighted and hereafter) were categorized as low risky driving, 1423 (59.22%) medium risky driving, and 389 (19.35%) high risky driving (Table 1). Higher risky driving class had a higher C-RDS score compared to lower risky driving class.

Figure 1.

Trajectories of risky driving among young drivers indicated by C-RDS score over five consecutive years beginning in 12^th grade

Table 1.

Proportions and average posterior probabilities of risky driving trajectory classes among young drivers

	N (Weighted %)	Average posterior probabilities (Range)
Low risky driving class	583 (21.43)	0.82 (0.50-1.00)
Medium risky driving class	1423 (59.22)	0.71 (0.50-0.98)
High risky driving class	389 (19.35)	0.83 (0.50-1.00)

Table 2 shows the frequency of 3-class risky driving trajectories (low, medium, and high) by demographic variables. The major proportion of C-RDS trajectory was medium risky driving across family affluence based on Rao-Scott χ² test (χ²=10.46, p<.05). More specifically, 463 (60.93%) low, 681 (61.98%) moderate, and 278 (52.96%) high family affluence participants were categorized as medium risky driving, respectively. Moderate (χ²=8.77, p<.05) and high (χ²=6.55, p<.05) family affluence was significantly different from low family affluence, respectively. No other significant difference was found across other sociodemographic covariates.

Table 2.

Frequency of 3-class risky driving trajectory classes by demographic variables

		Low N (%#)	Medium N (%#)	High N (%#)	χ2	P$
Urbanicity	Urban	212 (25.35)	467 (63.70)	71 (10.96)	9.26	0.06
	Suburban	187 (22.02)	426 (58.41)	132(19.54)
	Rural	113 (18.01)	407 (59.36)	148(22.66)
Sex	Male	256 (21.70)	602 (56.62)	207 (22.69)	4.49	0.11
	Female	327 (22.02)	821 (61.35)	182 (16.65)
Family Affluence	Low	222 (25.88)	463 (60.93)	79 (13.20)	10.46	0.03
	Moderate*	257 (18.33)	681 (61.98)	187 (19.69)
	High*	104 (23.20)	278 (52.96)	123 (23.84)

Note: Please see complete Table 2 including additional demographic variables in the Appendix.

Low: low risky driving class; Medium: medium risky driving class; High: high risky driving class.

^#Weighed %; χ²: Rao-Scott χ²;

^*Indicates significantly different from low affluence.

^$Covariates selected into the logistic regressions were based on the χ² test at the significant level of .10 to guard against the type II error.

Table 3 shows the means of psychosomatic and depressive symptoms by 3-class risky driving trajectories. The T-scores of depressive symptoms were 45.62, 50.11, and 51.60 for low, medium, high risky driving class, respectively. The mean T-scores of medium and high risky driving classes were significantly higher than the mean T-scores of low risky driving class. The mean values of psychosomatic symptoms were 14.32, 16.24, and 17.55 for low, medium and high risky driving classes, respectively. The mean psychosomatic symptoms scores of medium and high risky driving classes were significantly higher than the mean score of low risky driving class.

Table 3.

Means scores of depressive and psychosomatic symptoms by risky driving trajectory classes among young drivers

	Low risky driving			Medium risky driving			High risky driving			Total
	N	Mean#	SE	N	Mean#	SE	N	Mean#	SE	N	Mean#	SE
T-scores of Depressive Symptoms	212	45.62	1.08	706	50.11a	0.82	287	51.60a	1.03	1205	49.85	0.61
Scores of Psychosomatic Symptoms	208	14.32	0.61	702	16.24a	0.37	287	17.55a	0.61	1197	16.31	0.26

Note:

^aindicates significantly different from low risky driving class.

^#weighted. SE: standard error.

Table 4 shows the association of depressive symptoms and psychosomatic symptoms in 12^th grade with the risky driving trajectories classes from unadjusted to adjusted multinomial logistic regression models. Compared to the low risky driving class, one unit increase in the 12^th-grade PROMIS scores of depressive symptoms was significantly associated with higher likelihood of being in the medium (adjusted odds ratio [AOR]=1.04, 95%CI 1.01, 1.07) and the high risky driving classes (AOR=1.05, 95%CI 1.02, 1.08), respectfully, when controlling for the covariates. Likewise, compared to the low risky driving group, one unit increase in 12^th-grade score of psychosomatic symptoms was significantly associated with higher likelihood of involvement in the medium (AOR=1.06, 95%CI 1.00, 1.13) and the high risky driving classes (AOR=1.10, 95%CI 1.04, 1.16), respectively, when controlling for the covariates.

Table 4.

Multinomial logistic regressions of risky driving trajectories on 12^th-grade depressive and psychosomatic symptoms

		Unadjusted model			Adjusted model#
	OR	95%CI	P	AOR	95%CI	P
Depressive symptoms
Low risky driving class	Ref	—	—	Ref	—	—
Medium risky driving class	1.04	1,01,1.07	.01	1.04	1.01,1.07	<.01
High risky driving class	1.05	1.02,1.08	<.01	1.05	1.02,1.08	<.01
Psychosomatic symptoms
Low risky driving class	Ref	—	—	Ref	—	—
Medium risky driving class	1.06	1.00,1.13	<.05	1.06	1.00, 1.13	.04
High risky driving class	1.10	1.04,1.16	<.01	1.10	1.04, 1.16	<.01

Note: OR: odds ratio; AOR: adjusted odds ratio; CI: confidence interval.

^#adjusting for family affluence and urbanicity.

DISCUSSION

To the best of our knowledge, our study is one of the first to examine depressive and psychosomatic symptoms among teens and emerging adults with the aim to identify associated risky driving trajectory classes over five consecutive years beginning with the last year of high school (12^th grade). Our results show that these risky driving trajectories are relatively stable among emerging adults. In addition, those who have high levels of depressive symptoms and psychosomatic symptoms in high school are at higher risk of engaging in risky driving after high school.

Our findings are consistent with previous evidence suggesting that poor mental health issues may increase the likelihood of being involved in risky driving (Li et al. 2020; Farah et al. 2008). While the definitive details of the mechanistic pathway between mental health and risky driving remain unclear, it is clear that alcohol and substance use may be a shared and important link between mental health and risky driving. Specifically, some mental health disorders (e.g., depression and anxiety) are associated with binge drinking (Keith et al. 2015), higher likelihood of substance use disorders (i.e., self-treatment with substance to suppress negative mental health (Li et al. 2021), and limitations in decision making (Rubinsztein et al. 2006). Another study has reported that the increase in substance use among adolescents may predict the developmental trajectories of risky drivers when they turn to young adults (Bingham et al. 2004). Therefore, the upward trend of drinking (Li et al. 2020), substance use (Li et al. 2020), and impaired decision making (Farah et al. 2008) resulting from poor mental health may ultimately heighten the proclivity of involvement in risky driving. However, future research is warranted to investigate the effect of alcohol and drug use on the mechanistic pathway from mental health to risky driving in young drivers.

Our results also show that depressive and psychosomatic symptoms in high school may predict risky driving involvement in early adulthood (i.e., years 1-4 after high school). Further, it is possible that the risky driving patterns formed in high school may remain stable after high school. Therefore, our study highlights the need to identify, refer to treatment, and/or treat adolescents’ mental health disorders in high school. The early detection and treatment of depressive and psychosomatic problems may be an effective strategy to not only improve overall mental health but concurrently reduce or prevent risky driving among U.S. emerging adult drivers.

Currently, mental health care needs of most U.S. teens with mental health disorders do not get met due to limitations in mental health care resources (Cummings et al. 2013). To address this, multi-level strategies are needed. Already, school nurses, health educators, and community health practitioners play critical roles and need greater support in advocating for mental health education, promoting mental health assessment, and early detection of mental health disorders among high school students. Further, school administrators and public health policy makers need to continue to bolster their efforts to make mental health resources available to all U.S. high school teens (LaBrie et al. 2010). If not already part of their curricular activity, individual high schools may consider offering mental health education courses/seminars to their students and their families providing information on maintaining good mental health status, the basic features of mental health disorders, and the importance of treatment to prevent harmful consequence. High school educators/counselors can also help in facilitating and expanding the screening for mental health disorders in youth so that treatment might be more timely. Mental health consulting service should be also highly recommended and made more widely-available to high school adolescents (Hoagwood et al. 2001). Ultimately, mental health education, screening, and consulting service may help high school students early identify their potential mental health disorders and in turn to avoid related mental health issue consequences reducing risky driving behaviors and crashes.

We acknowledged several limitations. First, we cannot exclude self-reporting bias as we measured risky driving based on self-reported questionnaire. Thus, participants might avoid reporting their risky driving as a “bad behavior.” Second, participants’ responses regarding risky driving may change over time due to the changes of their driving skills, experience, and performance. Third, although we observed statistically significant associations of depressive and psychosomatic symptoms measured in 12th grade with trajectory C-RDS classes, it is unclear whether the associations will have any clinical significance and if so, to what extent. Given that this is a nationally representative sample, it is plausible to consider that the prospective associations identified in this study confirm the links between depressive and psychosomatic symptoms in high school students and their risky driving trajectory classes during the transitional period from adolescence to early adulthood to some measurable extent. However, more evidence is needed to justify the clinical significance of these links. Fourth, the different time frames of measuring depressive symptoms (over the last 7 days) and psychosomatic symptoms (e.g., in the last half year) may potentially influence their predictive effects on the risky driving trajectory classes. Specifically, the longer-term measures of psychosomatic symptoms which covered more severe symptoms may be relatively more predictive of risky driving compared to the shorter-term measures of depressive symptoms. Finally, some measures of the psychosomatic symptoms and depression symptoms may overlap. For example, “feeling low” as one of psychosomatic symptom measures may be associated with “I felt bad” as one of depressive symptoms measures. However, the associations of risky driving trajectory classes with T-scores of depressive symptoms and scores of psychosomatic symptoms were evaluated in different models which avoid the potential collinearity due to the overlap.

Despite these limitations, our study has several notable strengths. First, we established the longitudinal associations of depressive and psychosomatic symptoms in high school with risky driving after high school. Depressive and psychosomatic symptoms as indicators in high school can predict future risky driving trajectories several years after high school. Further, our findings affirmed that in this depressive and psychosomatic symptom context, risky driving patterns can be formed in high school and appear to remain relatively stable after high school. Second, we used a nationally representative sample allowing us to make some meaningful generalizations to the U.S. emerging adult populations. Third, the findings can provide added value information for risky driving preventions to lower the high rate of injury and fatal motor vehicle crashes in vulnerable emerging adults.

High school students with high levels of depressive and psychosomatic symptoms were at higher risk of engaging in more risky driving events in the transition after adolescence to early adulthood. The findings suggest that incorporating screening, referral to treatment, and treatment of mental health disorders in high school could yield important added valued and prove to be a vital opportunity to reduce risky driving among emerging adult drivers.