Depressive disorders (DD) are highly prevalent and debilitating. A leading cause of global disease burden, the prevalence of DD increases dramatically in adolescence and emerging adulthood (Rohde, Lewinsohn, Klein, Seeley, & Gau, Reference Rohde, Lewinsohn, Klein, Seeley and Gau2013), especially for females, and is associated with impaired functioning and increased mortality (Malhi & Mann, Reference Malhi and Mann2018). Although numerous risk markers have been identified as predictors of DD onset, most studies assume risk markers are static and assess them only once, without considering how risk changes over time (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021; Nelson, McGorry, Wichers, Wigman, & Hartmann, Reference Nelson, McGorry, Wichers, Wigman and Hartmann2017). In fact, many risk markers have been shown to be highly dynamic in nature and change substantially over time (e.g. Goldstein, Perlman, Eaton, Kotov, & Klein, Reference Goldstein, Perlman, Eaton, Kotov and Klein2020; Klein, Kotov, & Bufferd, Reference Klein, Kotov and Bufferd2011; Roberts, Walton, & Viechtbauer, Reference Roberts, Walton and Viechtbauer2006), especially during adolescence (Roberts & DelVecchio, Reference Roberts and DelVecchio2000; Trzesniewski, Donnellan, & Robins, Reference Trzesniewski, Donnellan and Robins2003).
It is largely unknown how the patterning of changes in risk factors is related to risk for DD onset. Two competing hypotheses have been proposed (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021). The risk escalation hypothesis posits that DD onset is most likely following an escalation in levels of risk factors. Several studies have provided support for this hypothesis, demonstrating that increases in risk factors predicted subsequent increases in the severity of depressive symptoms (Mu, Luo, Nickel, & Roberts, Reference Mu, Luo, Nickel and Roberts2016; Steiger, Allemand, Robins, & Fend, Reference Steiger, Allemand, Robins and Fend2014) and probability of DD onset (Laceulle, Ormel, Vollebergh, Van Aken, & Nederhof, Reference Laceulle, Ormel, Vollebergh, Van Aken and Nederhof2014). Alternatively, the chronic risk hypothesis posits that sustained elevation of risk over time predicts DD onset and that change in risk over time contributes little additional information in predicting DD onset. There has been some indirect support for this hypothesis. For example, chronic stressors predict increases in depressive symptoms (Jenness, Peverill, King, Hankin, & McLaughlin, Reference Jenness, Peverill, King, Hankin and McLaughlin2019) and subsequent DD onset (Hammen, Kim, Eberhart, & Brennan, Reference Hammen, Kim, Eberhart and Brennan2009). Furthermore, using a trait–state decomposition method, some studies have found that the stable component, rather than the state component, of personality traits predicted DD onset (Kendall et al., Reference Kendall, Zinbarg, Mineka, Bobova, Prenoveau, Revelle and Craske2015) and the course of depression (Naragon-Gainey, Gallagher, & Brown, Reference Naragon-Gainey, Gallagher and Brown2013). Finally, adolescents who exhibited subthreshold depression across two consecutive assessment waves had a higher likelihood of developing full-criteria depression at later assessment waves than those with subthreshold depression only at one assessment wave (Klein, Shankman, Lewinsohn, & Seeley, Reference Klein, Shankman, Lewinsohn and Seeley2009).
Although empirical studies have provided evidence that is consistent with both of these hypotheses, only one study, to our knowledge, has directly compared the risk escalation and chronic risk hypotheses (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021). In a sample of never-depressed early adolescent girls who were assessed prospectively across five waves, they evaluated the predictive effect of mean level and change in levels of risk factors over time on the development of a first DD onset. Results showed that mean level of risk factors outperformed increases in levels of risk factors in predicting DD onset, supporting the chronic risk hypothesis. This pattern of findings was consistent across 16 well-established risk markers in four different domains.
The present study extends Mu et al.'s (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) work in several ways. First, while Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) examined risk for DD onset during adolescence, the current study focused on transition to adulthood. DD incidence peaks in transition to adulthood (Rohde et al., Reference Rohde, Lewinsohn, Klein, Seeley and Gau2013), which is an important but less studied developmental phase for understanding the development of DD. Furthermore, adolescence is characterized with rapid changes in many psychological aspects, and trajectories of risk factors in adolescence can have long-term consequences on mental health in adulthood (Steiger et al., Reference Steiger, Allemand, Robins and Fend2014). However, it is unclear how dynamic risk during adolescence predicts DD onset in transition to adulthood. Second, unlike Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021), the current study adopted a more stringent definition of DD by excluding DD NOS and defined DD as major depressive disorder (MDD) or dysthymic disorder, which improves the internal validity of the current study. Third, we adopted an entirely different analytic approach. While Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) assessed change using difference scores between adjacent assessments separated by 9-month intervals, we examined whether more lasting change (i.e. 3 years) in risk factors predict subsequent DD onset. After all, 9 months may not have been long enough to capture meaningful change, as it may fail to distinguish transient fluctuations from lasting change. Hence, we employed latent growth curve modeling instead of cox regression as used in Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021). This approach models changes in levels of risk factors adjusting for measurement error and distinguishing it from transient fluctuations, providing a more precise estimate of change.
The current study aims to provide a rigorous test of the risk escalation and chronic risk hypotheses while addressing the above-mentioned issues. To our knowledge, no research to date has ever examined how dynamic risk during adolescence predicts DD onset in transition to adulthood. Well-established risk markers were examined, including subclinical depression and anxiety symptoms (e.g. Klein et al., Reference Klein, Shankman, Lewinsohn and Seeley2009, Reference Klein, Glenn, Kosty, Seeley, Rohde and Lewinsohn2013), personality traits (e.g. neuroticism, extraversion, and conscientiousness; Bagby, Quilty, & Ryder, Reference Bagby, Quilty and Ryder2008; Goldstein, Kotov, Perlman, Watson, & Klein, Reference Goldstein, Kotov, Perlman, Watson and Klein2018; Klein et al. Reference Klein, Kotov and Bufferd2011), depressogenic cognitive or personality styles (e.g. rumination, self-criticism, dependency; Burkhouse, Uhrlass, Stone, Knopik, & Gibb, Reference Burkhouse, Uhrlass, Stone, Knopik and Gibb2012; Kopala-Sibley, Klein, Perlman, & Kotov, Reference Kopala-Sibley, Klein, Perlman and Kotov2017; Nolen-Hoeksema, Reference Nolen-Hoeksema2000), and social risk markers (e.g. social support, peer/parent–child relationship, school engagement; Shochet, Homel, Cockshaw, & Montgomery, Reference Shochet, Homel, Cockshaw and Montgomery2008; Smith, Nelson, & Adelson, Reference Smith, Nelson and Adelson2019; Starr & Davila, Reference Starr and Davila2008). This study included six assessment waves over 6 years, during which a sample of girls was followed from adolescence into emerging adulthood. We examined how levels and changes in risk factors in adolescence predicted DD onset in transition to adulthood.
Methods
Participants
Data included in the current study were collected as part of the Adolescent Development of Emotions and Personality Traits (ADEPT) project. The original ADEPT project followed 550 adolescent girls from the start of adolescence for 3 years with five waves of assessments, with 9 months between two waves. The extended ADEPT project attempted to re-interview all 550 participants when they entered early adulthood. For the current study, we focused on girls who did not have a first onset of MDD or dysthymic disorder until wave 6, so the 72 participants who had MDD or dysthymic disorder during the first five waves were excluded (see online Supplementary Fig. S1 for a comparison of risk factors between participants who had a first onset between waves 5 and 6 and those who had an onset between waves 1 and 5). Participants who had medically caused DD (n = 1), bipolar disorder (n = 6), and those who were missing at wave 6 (n = 78) were also excluded from the analyses.
The current sample consisted of 393 females. Participants were 13.14–15.63 years old (mean = 14.38, s.d. = 0.62) at enrollment (Table 1; see online Supplementary Table S1 for the age range of participants at each wave). Most were non-Hispanic White (81.9%), and 34.5% came from families where both parents had a bachelor's or higher degree. None of these participants had intellectual disability or lifetime history of MDD or dysthymic disorder before enrollment. Details of the recruitment and inclusion/exclusion criteria can be found in Michelini et al. (Reference Michelini, Perlman, Tian, Mackin, Nelson, Klein and Kotov2021).
Table 1. Baseline demographic characteristics of participants with first DD onset at wave 6 v. participants with no DD onset across all waves
Note. First DD onset, n = 78; never DD onset, n = 315.
Participants completed five diagnostic interviews for DD and assessments of relevant risk factors at 9-month intervals for 3 years. Then, at a 6-year follow-up, participants completed a diagnostic interview for DD. Participants used life events calendar to aid recall of DD symptoms in the 3-year interval. Among the 393 participants, 315 participants never developed DD, and 78 participants had their first DD onsets during the wave 6 interval (MDD, n = 76; dysthymic disorder, n = 2). In onset group, there were no new onsets after 15 March 2020,Footnote †Footnote 1 indicating that the new onset is not the result of the pandemic. Informed assent and consent were obtained from all participants and their parents in waves 1–5 and informed consent was obtained from the participants in wave 6. The study was approved by the Stony Brook University Institutional Review Board.
Assessments
Depression diagnosis
DD (i.e. MDD and dysthymic disorder) were assessed using the Kiddie Schedule for Affective Disorders and Schizophrenia for School Aged Children Present and Lifetime Version (K-SADS-PL; Kaufman et al., Reference Kaufman, Birmaher, Brent, Rao, Flynn, Moreci and Ryan1997). The K-SADS-PL is a semi-structured diagnostic interview designed to assess current and lifetime psychiatric diagnoses in children and adolescents based on the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV). The K-SADS-PL interview was conducted by two well-trained research staff who were supervised by licensed clinical psychologists. An independent rater derived diagnoses from 48K-SADS-PL interview recordings to assess interrater reliability. Kappas for MDD and dysthymic disorder were 0.73 and 0.85, respectively (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021).
Depression and anxiety symptoms
The Inventory of Depression and Anxiety Symptoms (IDAS-II; Watson et al., Reference Watson, O'Hara, Naragon-Gainey, Koffel, Chmielewski, Kotov and Ruggero2012) was used to measure symptoms of depression and anxiety. The IDAS-II contains 18 non-overlapping subscales and a General Depression scale composed of items from six subscales measuring depressive symptoms. Participants were asked to indicate to what extent they experienced a symptom over the past 2 weeks on a Likert scale ranging from 1 (not at all) to 5 (extremely). Based on the relevance to risk for depression onset, the current study included the General Depression scale (20 items) and six subscales measuring ill temper (5 items), panic symptoms (8 items), social anxiety (6 items), traumatic intrusions (4 items), traumatic avoidance (i.e. avoiding reminders of past traumas; 4 items), and mania (5 items). The seven specific depressive symptom subscales were not included to avoid redundancy with the General Depression scale. The claustrophobia subscale, the euphoria subscale, and the three obsessive compulsive subscales were not included due to low relevance to risk for depression onset.
Personality traits
The Big Five Inventory (BFI; John & Srivastava, Reference John and Srivastava1999) was used to measure the personality traits of neuroticism, conscientiousness, and extraversion. The neuroticism and the extraversion subscales each contain eight items, while the conscientiousness subscale contains nine items. Participants were asked to indicate the extent to which they agreed that a characteristic (e.g. moody; a reliable worker; sociable) applied to them on a Likert scale ranging from 1 (disagree strongly) to 5 (agree strongly).
Rumination
The Ruminative Responses Scale (RRS) of the Response Styles Questionnaire (Nolen-Hoeksema, Reference Nolen-Hoeksema1987) contains 22 items. Participants were asked to indicate what they generally thought (e.g. think ‘Why can't I get going?’) when they felt depressed on a four-point Likert scale ranging from 1 (almost never) to 4 (almost always).
Self-criticism
Bagby, Parker, Joffe, and Buis's (Reference Bagby, Parker, Joffe and Buis1994) revision of the self-criticism subscale of the Depressive Experiences Questionnaire (DEQ; Blatt, D'Afflitti, & Quinlan, Reference Blatt, D'Afflitti and Quinlan1976) was used to measure self-criticism. The subscale contains nine items (e.g. ‘I often find that I don't live up to my own standards or ideals’). Participants were asked to rate each item on a five-point Likert scale ranging from 1 (disagree strongly) to 5 (agree strongly).
Dependency
The emotional reliance subscale of the Interpersonal Dependency Inventory (IDI; Hirschfeld et al., Reference Hirschfeld, Klerman, Gouch, Barrett, Korchin and Chodoff1977) has 10 items (e.g. ‘I would be completely lost if I didn't have someone special’). Each item was rated on a four-point Likert scale ranging from 1 (not characteristic of me) to 4 (very characteristic of me).
Social support
The Multidimensional Scale of Perceived Social Support (MSPSS; Zimet, Dahlem, Zimet, & Farley, Reference Zimet, Dahlem, Zimet and Farley1988) has 12 items. Participants indicated the perceived adequacy of social support received from their family, friends, and significant others on a seven-point Likert scale ranging from 1 (very strongly disagree) to 7 (very strongly agree).
School engagement
Three subscales of the School Attitude Assessment Survey – Revised (SAAS-R; McCoach & Siegle, Reference McCoach and Siegle2003) – attitudes toward school, attitude toward teachers, and self-motivation/regulation – were used to measure school engagement. The three subscales were aggregated, totaling 14 items. A sample item was ‘I am glad that I go to this school’. Each item was rated on a seven-point Likert-type agreement scale (1 = very strongly disagree, 7 = very strongly agree).
Bullying
The Victim Version of the Revised Peer Experiences Questionnaire (RPEQ; De Los Reyes & Prinstein, Reference De Los Reyes and Prinstein2004) has three subscales: overt victimization, relational victimization, and reputational victimization. The scales were aggregated, totaling 12 items. Participants indicated how often they were treated in a certain way (e.g. A teen excluded me from his/her group of friends) over the past 9 months on a five-point Likert scale (1 = never, 5 = a few times a week).
Parental criticism
The criticism subscale of the Network of Relationships Inventory (NRI; Furman & Buhrmester, Reference Furman and Buhrmester2009) includes three items each for the participant's mother and father. Participants rated how often each parent interacted with them in a certain way (e.g. How often does this person criticize you?) on a five-point Likert scale ranging from 1 (little or none) to 5 (the most), respectively. These six items were aggregated into a total score.
Psychometric properties of these measures (i.e. Cronbach's alphas and stability coefficients) can be found in Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021).
Statistical analysis
The outcome was whether or not DD onset occurred during the 3-year interval between waves 5 and 6. Predictors were intercepts and slopes estimated by latent growth curve modeling using five waves of assessments for each of 17 risk factors, in turn. We conducted the tests in two steps. First, we specified latent growth curve models for each of the 17 risk factors to estimate the mean and interindividual difference in the level and change of risk. Raw scores of each risk marker were standardized using means and standard deviations across all participants and assessment waves available. Grand mean standardized scores of a risk marker at each of the five assessment waves were used as indicators. For each risk factor, the intercept factor was set at either the first or the fifth assessment wave respectively to estimate both the initial and the proximal level of risk relative to time of onset. We therefore examined two univariate growth curve models for each of the 17 risk factors with the intercept setting at the first or the fifth assessment wave separately. Maximum likelihood estimation with robust standard error was performed.
Next, we address dichotomous wave 6 onset of DD in these models. Specifically, DD onset was regressed on the intercept and linear slope, which were latent factors with random effects estimated using latent growth curve modelling and were allowed to correlate (Fig. 1). The intercept factor was set at the first and the fifth assessment occasions respectively. No covariates were included in the structural model.
Figure 1. Structural equation model predicting first DD onset at wave 6 using (a) intercept (w1) and slope and (b) intercept (w5) and slope for each risk marker estimated from wave 1 to 5.
Note. RM = risk marker; W1 = wave 1; W2 = wave 2; W3 = wave 3; W4 = wave 4; W5 = wave 5; W6 = wave 6.
Missing data were addressed using full-information maximum likelihood estimation. All analyses were conducted using Mplus 8.7 (Muthén & Muthén, Los Angeles, CA, USA). We used the following indices to evaluate the fit of the latent growth curve models: the comparative fit index (CFI), the Tucker–Lewis index (TLI), standardized root mean square residual (SRMR), and root mean square error of approximation (RMSEA). Conventional guidelines (Hu & Bentler, Reference Hu and Bentler1999; Marsh, Hau, & Wen, Reference Marsh, Hau and Wen2004; Hooper, Coughlan, & Mullen, Reference Hooper, Coughlan and Mullen2008) suggest that TLI and CFI values of 0.90 or greater indicate adequate fit and values of 0.95 or greater indicate excellent fit; for SRMR and RMSEA, values of 0.08 or less indicate acceptable and 0.05 or less indicate excellent fit.
Results
Level and change of risk before onset
Descriptive statistics
Among the 393 participants, 78 had a first DD onset between waves 5 and 6 (i.e. onset group), while the remaining participants had no DD onset in their life (i.e. no onset group). Standardized mean scores of each risk marker at each wave for the two groups are presented in Fig. 2, and the raw scores (i.e. means and standard deviations) are presented in online Supplementary Table S2. In general, the onset group exhibited a chronically higher level of risk compared to the no onset group for most risk markers throughout the first five assessment waves.
Figure 2. Standardized scores of each risk marker from wave 1 to 5 by onset group at wave 6.
Estimates from latent growth curve models
To estimate level and change of risk, growth curve models were constructed for each risk marker with the intercept set at the first or the fifth assessment wave. Fit indices for the growth curve models are presented in Table 2. Growth curve models generally had adequate fit.
Table 2. Fit indices of growth curve model with intercept set at wave 5 for each risk marker
Note. RMSEA, the root mean square error of approximation; CFI, comparative fit index; TLI, Tucker–Lewis index; SRMR, standardized root mean square residual; AIC, Akaike information criteria; BIC, Bayesian information criteria. We did not present fit indices with intercept set at wave 1 because no differences in fit indices were observed with intercept set at wave 1 v. wave 5. *p < 0.05; **p < 0.01.
Level and change estimates from the growth curve models are presented in online Supplementary Table S3. The mean of the slope for each risk marker indicates the change rate of risk from wave 1 to 5. In terms of magnitude of change, 11 out of 17 risk markers showed significant change over time based on the means of the slopes (ps < 0.05). In terms of their change direction, the trajectories of 10 risk markers (i.e. general depression, ill temper, social anxiety, panic, traumatic intrusion, traumatic avoidance, mania, self-criticism, dependency, bullying) indicated a significant decrease in risk from wave 1 to 5, whereas the slope of one risk marker (i.e. parental criticism) indicated a significant increase in risk across waves.
The variance of the slope indicates interindividual difference in the intraindividual change of the risk factors over time. The variances of the slopes for all risk markers reached statistical significance (ps < 0.01), except for panic (p = 0.56) and traumatic intrusions (p = 0.33), indicating significant interindividual difference in the change of risk (see online Supplementary Fig. S2 for a spaghetti plot of each participant's risk trajectory in a random subsample of 10% of the current sample).
Predictive utility of level and change in risk
We created a structural model for each risk marker to examine the predictive utility of the level and change of risk factors on first DD onset. The latent intercept and slope of each risk marker estimated using data from wave 1 to 5 were used as predictors of DD onset at wave 6. Model fit information (i.e. log-likelihood value, the Akaike information criteria, and the Bayesian information criteria) is presented in online Supplementary Table S4. Table 3 displays the standardized results of structural models.
Table 3. Model results of structural equation model predicting first DD onset at wave 6 using latent intercept and slope estimated using standardized scores of each risk marker from wave 1 to 5
Note. Standardized results of structural models were presented. Two-sided statistical tests were performed at a level of significance of 5%. Wave 1 = intercepts set at wave 1, wave 5 = intercepts set at wave 5. Red indicates significant and positive prediction; green indicates significant and negative prediction. Lighter color indicates significance test coefficients lower than 0.05, whereas darker color indicates significance test coefficients lower than 0.01.
*p < 0.05; **p < 0.01.
A higher level of risk at wave 5 significantly predicted first DD onset for all risk markers except for traumatic avoidance, traumatic intrusions, bullying, and parental criticism. In contrast, across all risk markers, the change in risk from wave 1 to 5 did not significantly predict DD onset after controlling for the wave 5 level of risk. A similar pattern was observed when controlling for wave 1 risk assessment as baseline. Across all risk factors, risk assessment at wave 1 significantly predicted first DD onset for all risk markers except for traumatic avoidance, traumatic intrusions, bullying, and parental criticism. For only two out of 17 risk markers (i.e. self-criticism; social support), an increase in risk significantly predicted first DD onset.
Discussion
This study extends our prior study (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) – which was the first to directly test the chronic risk and risk escalation hypotheses for depression – by doubling the follow-up period, using latent growth curve models instead of subtraction-based difference scores, and examining the transition from adolescence to emerging adulthood. For 13 out of 17 well-established risk markers for DD, chronic risk elevation during adolescence, but not change in risk, predicted first DD onset in transition to adulthood. Girls who had a first lifetime onset of DD in transition to adulthood have exhibited persistent and elevated levels of risk since entering adolescence. Our sample showed significant changes on almost all of the risk markers from early to late adolescence, but for 15 of the 17 risk factors tested, the change pattern did not significantly differ for those girls who had DD onset v. those who remained healthy. These results provide strong support for the chronic risk hypothesis, and little support for the risk escalation hypothesis.
The current findings advance our understanding of the developmental pathways of DD from adolescence to emerging adulthood. Our results demonstrated that chronic risk elevation, rather than an increase in risk over time, confers susceptibility to DD onset, supporting the chronic risk hypothesis. This pattern of results aligned with Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021), where the mean level of risk across multiple assessment waves outperformed risk escalation when predicting first DD onset in adolescence. Similarly, previous research observed strong links between persisting high levels of risk, such as sustained subthreshold depression, chronic stress, and the stable component of risk factors, and a greater likelihood of DD onset (e.g. Kendall et al., Reference Kendall, Zinbarg, Mineka, Bobova, Prenoveau, Revelle and Craske2015; Klein et al., Reference Klein, Shankman, Lewinsohn and Seeley2009). It is important to note that the current study focused on girls who had their first DD onset during transition to adulthood, whereas Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) focused on girls who had their first DD onset in adolescence.
One natural question that follows is why these girls did not have DD onset earlier, like some other girls who had onset during the first five waves of assessment (Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021). Random exposure to life events may be an important mechanism for the timing of DD onset, as twin studies show that unique environmental variance explains over 60% of the risk of MDD (Sullivan, Neale, & Kendler, Reference Sullivan, Neale and Kendler2000). One related possibility is that some individuals may be more vulnerable to a specific set of stressors than others (Lazarus & Folkman, Reference Lazarus and Folkman1984). The type of stressors adolescents face (e.g. family conflicts, dramatic biological changes brought on by puberty) are different from those faced by emerging adults (e.g. independence, careers). Individual differences in vulnerability to stressors may explain the difference we observed in timing of DD onset. Another possibility pertains to a dose–response association between elevated risk for DD and health outcomes, with the impact of elevated risk accumulating until it reaches a threshold to trigger DD onset (Dunn et al., Reference Dunn, Soare, Raffeld, Busso, Crawford, Davis and Susser2018). From this perspective, individuals with higher sustained levels of risk cross the threshold earlier than those with lower sustained levels of risk.
Unexpectedly, for four of the 17 risk markers (i.e. traumatic avoidance, traumatic intrusions, bullying, parental criticism), neither risk level nor change predicted DD onset, which is inconsistent with previous work supporting the predictive role of these risk markers (Klein et al., Reference Klein, Glenn, Kosty, Seeley, Rohde and Lewinsohn2013; Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021). One possible explanation is that the depressogenic effects of these risk markers are time-limited (Shanahan, Copeland, Costello, & Angold, Reference Shanahan, Copeland, Costello and Angold2011). Thus, levels and changes in these risk markers in adolescence may have a weaker influence on mental health in the transition to adulthood. For example, Shanahan et al. (Reference Shanahan, Copeland, Costello and Angold2011) found that the associations between psychosocial risk factors and DD onset were stronger when they were measured concurrently than at different ages. Similarly, Jaffee et al. (Reference Jaffee, Moffitt, Caspi, Fombonne, Poulton and Martin2002) found that early childhood levels of risk had a stronger predictive effect on juvenile-onset DD than on adult-onset DD. Future research is needed to clarify whether certain risk factors have time-limited depressogenic effects. An alternative possibility is that the influence of some risk factors differs as a function of developmental stage (Jaffee et al., Reference Jaffee, Moffitt, Caspi, Fombonne, Poulton and Martin2002; Shanahan et al., Reference Shanahan, Copeland, Costello and Angold2011). However, this is less likely given that previous research has supported the predictive effect of trauma-related symptoms and interpersonal factors on adult-onset DD.
According to the risk escalation hypothesis, risk should increase throughout adolescence for girls who developed DD. However, for all 17 risk markers examined, change of risk from early to late adolescence did not predict DD onset when the effect of risk level was controlled (worsening in self-criticism and social support no longer predicted DD onset when its level at wave 5 was controlled). This pattern is consistent with Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021), but inconsistent with other past evidence supporting the risk escalation hypothesis (e.g. Laceulle et al., Reference Laceulle, Ormel, Vollebergh, Van Aken and Nederhof2014; Mu et al., Reference Mu, Luo, Nickel and Roberts2016; Steiger et al., Reference Steiger, Allemand, Robins and Fend2014). The current study estimates change using data from five assessment waves, thereby allowing us to separate slope and intercept more precisely as compared to previous studies. With a more precise estimation of change, we still found no evidence supporting the predictive role of change in risk levels. The current study has several limitations. First, the window for DD onsets was wide, that is, up to 3 years after wave 5, allowing the possibility that our model could have missed an escalation shortly before onset. However, this possibility is not likely, given that Mu et al. (Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021) found no evidence supporting the predictive effect of risk change 9-month priors to onset when risk level was controlled. Second, participants were mainly white girls from a particular region of the USA, which limits the generalizability of our findings to other cultures and populations. Replication of the current study in more diverse samples is warranted. Third, the assessment of risk factors relied solely on self-report. Although self-report has been shown to be the best assessment approach for many risk factors (Babor, Brown, & Del Boca, Reference Babor, Brown and Del Boca1990), multiple methods (e.g. behavioral observations, physiological and neurocognitive assessments) should be employed in future studies to provide a more comprehensive picture of risk profiles. Fourth, the current study did not examine an exhaustive list of risk markers, as some other well-established risk markers, such as romantic relationship, which has been shown to be effective for adolescent girls, (Starr et al., Reference Starr, Davila, Stroud, Clara Li, Yoneda, Hershenberg and Ramsay Miller2012) were not included. More studies could be conducted on whether other risk markers fit the same pattern in the future.
Notwithstanding these limitations, the current study has the following strengths. First, we explored the developmental period spanning adolescence to emerging adulthood, which is when DD incidence peaks (Rohde et al., Reference Rohde, Lewinsohn, Klein, Seeley and Gau2013). More importantly, we examined the long-term effects of risk level and change on first DD onset. Second, we estimated the latent intercepts and slopes of risk factors using five assessments over 3 years, whereas most previous studies estimated risk change using data from two assessment waves, rendering it difficult to control for measurement errors or to disentangle lasting change from temporal fluctuations (e.g. Laceulle et al., Reference Laceulle, Ormel, Vollebergh, Van Aken and Nederhof2014; Mu et al., Reference Mu, Li, Tian, Perlman, Michelini, Watson and Kotov2021).
In conclusion, we directly compared the chronic risk and risk escalation hypotheses. Our findings demonstrate that for the established depression risk factors, elevated level of risk, but not risk change, predicts DD onset. They also argue that individual differences in depression risk emerge in early adolescence or even childhood and are maintained in the absence of clinical depression. This suggests that to predict and prevent DD, researchers and practitioners should focus on persistently elevated levels of risk to identify individuals who are most likely to develop DD and to provide targeted DD prevention. In addition, adolescents with high levels of risk factors can be targeted for programs to prevent the development of DD. School-based monitoring of risk level among adolescents would be informative for identifying those who are especially vulnerable for DD.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723002441.
Financial support
This work was supported by the National Institute of Mental Health of the National Institutes of Health (R01MH093479 to R. K.).
Competing interests
None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
