Patterns, predictors, and outcome of the trajectories of depressive symptoms from adolescence to adulthood

The long‐term trajectory of depressive symptoms has a heterogeneous pattern. Identifying factors associated with different trajectories and outcomes may have important theoretical and clinical implications. This study explored patterns of depressive symptom trajectory from adolescence to adulthood, and their relationship with subsequent psychiatric disorders.

, which makes these periods important for our understanding of the developmental course of depression. Enhanced understanding of time course and key features of depressive symptoms across these critical developmental stages helps to identify adolescents who are at risk of developing depression (Kwong et al., 2019;Yaroslavsky, Pettit, Lewinsohn, Seeley, & Roberts, 2013). Furthermore, different trajectory patterns may be associated with different etiological processes, comorbidity clusters, prognoses, and psychosocial outcomes (Muthén & Muthén, 2000;Wardenaar, Monden, Conradi, & De Jonge, 2015). Therefore, being able to characterize trajectories of depressive symptoms may help to elucidate mechanisms that will better target prevention and intervention programs.
The long-term trajectories of depressive symptoms have been described differently across studies (see Musliner, Munk-Olsen, Eaton, & Zandi, 2016). In some studies, the trajectories of depressive symptoms have been described in terms of their main features, namely, severity (e.g., low, medium, and high; Ferro, Gorter, & Boyle, 2015) and stability (e.g., stable, increasing, and decreasing; Yaroslavsky et al., 2013) For example, Yaroslavsky et al. (2013) found three classes of depressive symptom trajectory: low decreasing, moderate decreasing, and high stable, which represented 24%, 44%, and 32% of the participants, respectively. For all classes, depressive symptoms showed a significant decrease across the first two assessment periods. For the low and moderate decreasing classes, depressive symptoms continued to decrease during the transition from late adolescence to emerging adulthood. However, in the high stable class, depressive symptoms plateaued over time. Rawana and Morgan (2014) investigated the trajectory of depressive symptoms from early adolescence to young adulthood in a nationally representative sample of young Canadians. Their findings showed depressive symptoms to decline slightly at ages 12 through 14, began to increase from ages 14 through 17, and then declined through age 21.
The period of greatest vulnerability was ages 14-17. Ferro et al. (2015) examined trajectories of depressive symptoms among youths aged 12-25 years using latent class growth modeling. Three distinct trajectories were identified during emerging adulthood: minimal (55%), subclinical (39%), and clinical (6%). In all trajectories, the age group with the highest level of depressive symptoms was between 15 and 17 years.
Trajectories of depressive symptoms were predicted by gender (Ferro et al., 2015;Musliner et al., 2016;Rawana & Morgan, 2014;Vannucci & McCauley Ohannessian, 2018), and the presence of personal and parental psychiatric disorders. Kwong et al. (2019) recently compared gender differences in the trajectories of depressive symptoms over eight occasions between 11 and 22 years of age.
Females, compared to males had steeper increases in depressive symptoms to age 20, after which levels of depression first plateaued and then started to show a decrease for both males and females.
Furthermore, females had an earlier age of peak velocity (i.e., ages at which depressive symptoms increased most rapidly) of depressive symptoms than males (13.5 years and 16.4 years, respectively). In Ferro et al. (2015), those in the subclinical and clinical symptoms trajectories tended to be female and participants with low self-concept, low socioeconomic status, poor interpersonal relations, and chronic health conditions. Poor coping skills (Yaroslavsky et al., 2013) and lack of connections with friends (Mazza, Fleming, Abbott, Haggerty, & Catalano, 2010) were also associated with high depressive symptom trajectories.
Specific trajectory patterns during adolescence are associated with specific outcomes during adulthood. High depressive trajectory in adolescence was associated with higher emotional problems (de la Torre-Luque, Fiol-Veny, Balle, Nelemans, & Bornas, 2019;Musliner et al., 2016), lower educational attainment in adulthood (Dekker et al., 2007), and risk-taking behaviors, such as excessive drinking and smoking (Wickrama & Wickrama, 2009). Sabiston et al. (2013) showed an elevated depressive trajectory to be associated with lower participation in physical activity and team sports in young adulthood. In Yaroslavsky et al. (2013), participants whose depressive trajectory were in the moderate and high stable classes had worse outcomes at age 30, including lower level of education, lower annual household income, poorer adjustment on each of psychosocial measures, and met the diagnosis of major depressive disorder (MDD), anxiety, and substance use disorders.
As mentioned above, inconsistent results (especially in relation to symptom trajectory enumeration) have been observed across studies on developmental course of depression in adolescence and adulthood. Some studies have short follow-up periods (e.g., de la Torre-Luque et al., 2019) which hinders the impact of transition from adolescence to adulthood. Other studies considered a unitary depressive symptom course, overlooking individual differences that may lead to heterogeneous trajectories underlying an overall course (e.g., Rawana & Morgan, 2014). Finally, findings of some studies were based on small sample (e.g., Wardenaar et al., 2015).
Given the inconsistent patterns of depressive trajectories reported in previous studies, one of the main aims of this study was to examine the developmental trajectories of depressive symptoms from adolescence to adulthood. The second aim was to identify predictors of each of these trajectories. The following sociodemographic and health-related (i.e., sex, race, self-reported health, and frequency of physical activity) and psychosocial factors (i.e., worry, self-esteem, loneliness, state anxiety, emotional disorders) were included as previous studies have consistently reported them to be associated with depression (e.g., Chaiton et al., 2013;Mazza et al., 2010;Rawana & Morgan, 2014). The third aim was to examine the outcome of each trajectories in terms of the development of mental disorders (i.e., anxiety, depression, and drug use disorders) at age 30. lived with their biological parents (95.6%). All participants and guardians provided a written consent to participate.
Reliability indexes across measurement occasions were acceptable within the sample (Cronbach's α between .90 and .92). Moreover, correlations were moderate across measurement occasions (r from 0.29 to 0.45).

| Follow-up instruments
Some sociodemographic and health-related information was collected at T4 follow-up (i.e., years of education, employment status, marital status, household income). Moreover, a joint administration of the Longitudinal Interval Follow-Up Evaluation interview (LIFE; Keller et al., 1987) and the Structured Clinical Interview for DSM-IV (SCID-I; First, Gibbon, Spitzer, & Williams, 1996) was conducted (see

| Procedure
Detailed information about the OADP has been reported elsewhere (Lewinsohn et al., 1993;Rohde et al., 2007). A baseline assessment (T1; 1987-1989) was conducted on a random sample of adolescents who attended high school in Western Oregon (ages 14-18). The participants who completed the T1 assessment (n = 1,709) were invited to a follow-up assessment (T2; 1988-1990)  The K-SADS and baseline scales (i.e., UCLA, Rosenbaum Selfesteem Scale, ASS, and worry scale) were administered at T1. Participants completed the CESD-20 three times, from T1 to T3. Diagnostic interviews were conducted at T2 and T3. Finally, the LIFE/ SCID-I interview and a set of questionnaires were administered at T4 (i.e., Coping Skills Questionnaire, SAS, RSS, PSS, and UES). Trajectory class membership profile was examined by means of logistic regression. We followed a forward covariate entry and three models were tested: the unconstrained model, a model with baseline sociodemographic and health-related covariates (i.e., sex, race, selfreported health and frequency of physical activity); and another model adding baseline psychosocial factors (i.e., worry, self-esteem, loneliness, anxiety state and diagnosis of emotional disorders). To control for uncertainty in class membership, the posterior probability ESSAU ET AL.

| Data analysis
| 567 of belonging to assigned class was used as a weighting variable.
Model selection relied on the AIC. The relative risk ratio (RRR) was used as an estimate of the probability of being member of the concrete class (in comparison to the normative one) given a concrete covariate.
Finally, prediction of mental disorders (i.e., MDD, anxiety disorders, drug use disorders) over age 30 (T4 follow-up) used biasreduction generalized linear modeling (brglm; Kosmidis, 2014;Kosmidis & Firth, 2009). This approach allows accurate estimates being derived when binary outcomes have a very large separation between categories. A model including the following covariates was performed for each outcome: depressive symptom trajectory membership, T4 sociodemographic (i.e., sex, marital status, working status) and psychosocial factors (i.e., social support, self-esteem, coping skills, unpleasant events, social adjustment). The diagnosis of a psychiatric disorder (i.e., MDD, anxiety, or drug use disorders) at base-

| RESULTS
Descriptive statistics of participants are displayed in Table 1. The mean of depressive symptoms decreased over time. Very few participants reported poor health status at baseline and almost half of the adolescents reported doing physical exercise at a frequent basis. Table 1 also displays attrition analyses between participants who were followed across the four follow-up waves (i.e., sample in analysis) and those who dropped out. Significant differences were found between samples in terms of some T1 sociodemographic (sex, race) and health-related factors (self-reported health, worry, loneliness, anxiety, emotional disorder diagnosis, and depressive symptoms).
However, differences showed marginal or small effect size (d < 0.50 or V < 0.30).

| Trajectory class identification
Model comparison revealed that a model comprising three heterogeneous trajectory classes showed the best data fit (Table S1) Additionally, those who showed higher levels of loneliness and anxiety at baseline were at higher risk to being classified into this class.

| Diagnosis prediction at age 30
The accuracy measure and covariate coefficients for each outcome prediction solutions are displayed in Table 3. All models showed good accuracy in predicting a psychiatric diagnosis, with AUC > 0.80 for the anxiety disorder and major depression solutions.  Specifically, participants with a drug use disorder were likely to be male, not married and had poorer coping skills than those without this disorder.

| DISCUSSION
The aims of this study were to determine the number and nature of trajectories of depressive symptoms from adolescence to adulthood, and examine factors in adolescence that predicted these trajectories.
Three trajectories were identified, which were labeled "decreasing symptom" (characterized with a decreasing trajectory of symptoms), "increasing symptom" (characterized by a decreasing pattern of symptoms but raising after age 20), and "normative symptom" (characterized by consistently low symptom levels). These trajectory patterns are consistent to those reported in previous studies from adolescence across to adulthood (Stoolmiller, Kim, & Capaldi, 2005;Yaroslavsky et al., 2013). Moreover, our results are in line with previous studies (Hankin et al., 1998;Thapar et al., 2012) which showed adolescence to be a sensitive period for the manifestation of depressive symptoms; in the present study, high levels of depressive symptoms were found in two of the three classes among participants age 14-16.
Like previous studies Ferro et al., 2015;Mezulis, Salk, Hyde, Priess-Groben, & Simonson, 2014), most participants (78.8%) experienced low levels of depressive symptoms from adolescence to adulthood. This is not surprising given that participants were from a community sample. Also consistent with previous studies (Costello, Swendsen, Rose, & Dierker, 2008;Rawana & Morgan, 2014;Yaroslavsky et al., 2013), we found a pattern of decreasing depressive symptoms. Interestingly, we found an increasing symptom trajectory with a unique shape which has not been reported elsewhere. This trajectory was present in 6.1% of the participants, which showed a general increase with a slight decrease near age 20.
This increase may have been related to the transition from adolescence to emerging adulthood, a period which is characterized by significant psychological changes (e.g., autonomy, identity) and distinct psychosocial events (e.g., residence, career pursuits). These changes are often associated with both risk and opportunities, which in turn could have a significant impact in mental health (Schulenberg, Sameroff, & Cicchetti, 2004). As reported by Rohde, Lewinsohn, Klein, Seeley and Gau (2013), emerging adulthood is significantly related to both first incidence and recurrence of MDD.
Each of the three trajectories seemed to have different predictors. At baseline, members of the increasing trajectory relative to the normative trajectory showed high levels of loneliness and state anxiety. These findings are in agreement with studies that reported elevated emotional symptoms to be a predictor of anxiety and/or depression in adulthood (Balázs et al., 2013;Essau et al., 2014). To our knowledge the present study is the first to show the impact of low involvement in physical exercise on the long-term trajectory of increasing depressive symptoms. The positive health benefits of physical exercise are well-documented (Biddle & Asare, 2011;Lee et al., 2012). For example, as reported in a large recent U. S. study, exercising was associated with reduced self-reported mental health burden (Chekroud et al., 2018); participants who exercised were approximately 1.5 fewer days of poor mental health in the past month compared to those who did not exercise. Furthermore, those who were engaged in team sports and cycling had the lowest mental health burden; in interpreting these findings, the authors argued that engagement in sports was related to social activity that promoted resilience to stress and reduced depression, and reduced social withdrawal and isolation. Gender (i.e., being female) has been reported in previous studies as a predictor of increasing depressive  Côté et al., 2009;Vannucci & McCauley Ohannessian, 2018), however, in the current study, participants in the decreasing symptoms class were mostly female.
Other studies have shown comparable gender distribution across various patterns of depressive trajectories (Arizaga, Polo, & Martinez-Torteya, 2018). The reason for this inconsistent finding is not clear, although participant's age has been speculated as accounting for gender difference in depressive symptoms (Legerstee et al., 2013).
In terms of outcomes, it is interesting that the increasing trajectory predicted development of anxiety disorders during adulthood, but not the development of MDD. The latter diagnostic outcome was related to lower levels of self-esteem, and poorer coping skills and social adjustment. As found by Yaroslavsky et al. (2013), these findings could be interpreted as supporting the interpersonal models of depression persistence (Pettit & Joiner, 2006).
Major strengths of the OADP data set are its rigorous methodological design (e.g., longitudinal design, systematic data collection protocols, implementation of diagnostic interviews) and lengthy follow-up period, which enabled participants to be followed from childhood and adolescence (14-18 years) to adulthood (30 years) via four assessments conducted over a 16-year period. As such, potential cofounders could be controlled and potential recall bias could be reduced. As the participants were comprised of a large community sample, they do not have the selection bias inherent in clinical samples. Diagnosis of any of these anxiety disorders: panic disorder (with or without agoraphobia), social phobia, specific phobia, separation anxiety, generalized anxiety disorder. b Diagnosis of either drug abuse or drug dependence disorder. *p < .05. **p < .01. ***p < .001.