Multiple domains of risk factors for first onset of depression in adolescent girls

Abstract

Background:

First onsets of depression are especially common in adolescent females and often develop into chronic/recurrent illness. Surprisingly few studies have comprehensively evaluated multiple domains of etiologically-informative risk factors for first onset in adolescents from the community. We investigated whether clinical, cognitive, personality, interpersonal, and biological risk factors prospectively predict a first onset of depressive disorder (DD), and of DD with a chronic/recurrent course, in a community sample of adolescent girls.

Methods:

479 girls (13.5–15.5 years) with no history of DD completed baseline assessments of risk factors and five diagnostic assessments over 3 years. Baseline measures were analyzed separately and jointly to prospectively predict first-onset DD and first-onset chronic/recurrent DD.

Results:

Most risk factors predicted first-onset DD (n=93), including depressive symptoms, anxiety disorders, rumination, personality traits, blunted neural response (late positive potential [LPP]) to unpleasant pictures, peer victimization, parental criticism, and parental mood disorder. Depressive symptoms, rumination, parental mood disorder, and parental criticism were independently associated with first onsets. Nearly all measures, including a blunted neural response to rewards (reward positivity [RewP]), also predicted first-onset chronic/recurrent DD (n=52), with depressive symptoms, low extraversion, poor peer relationships, and blunted RewP emerging as independent risk factors.

Limitations:

This study focused on adolescent females and therefore does not provide information on males.

Conclusions:

Multiple domains of risk factors in early adolescence are prospectively associated with first-onset DD and chronic/recurrent DD. A smaller subset of risk factors uniquely contributing to first onsets may represent core vulnerabilities for adolescent-onset depression and promising prevention targets.

INTRODUCTION

First-onsets of depressive disorders (DD) are common in adolescence and often presage a chronic/recurrent illness (Klein and Allmann, 2014; Rohde et al., 2013). Sex differences in DD also emerge in adolescence, with rates in females (~30%) being more than twice as high as in males (Breslau et al., 2017; Lewinsohn et al., 1998). Elucidating the risk factors for first-onset DD in early adolescence, when youth enter the period of high risk, remains essential for refining etiological theories of depression, identifying at-risk individuals, and informing cost-effective preventive strategies.

A broad range of vulnerabilities, spanning clinical, personality, cognitive/dispositional, interpersonal, and biological domains, have been hypothesized to play a role in the pathogenesis of DD (Forbes and Dahl, 2012; Hankin, 2012; Kendler et al., 2002; Klein et al., 2011; Nolen-Hoeksema, 1987; Rottenberg et al., 2005; Rudolph et al., 2008). Yet, prospective studies investigating first-onsets of DD in adolescence, especially in community samples and samples not selected for being at high risk (e.g., offspring of depressed parents), are limited, as most longitudinal studies of depression risk do not distinguish first-onsets from recurrent episodes. Since different risk factors have been shown to predict first and recurrent onsets (Lewinsohn et al., 1999), elucidating which factors prospectively predict first-onsets is crucial for informing early identification and prevention strategies.

The most-studied risk factors are clinical antecedents and family history. Parental mood disorder is among the best-established risk factors (Elsayed et al., 2019; Starr et al., 2014; Weissman et al., 2016). Subthreshold depressive symptoms, anxiety, irritability, and behavioral disorders have also been shown to predict later first-onset of DD in several studies (Copeland et al., 2009; Kim-Cohen et al., 2003; Klein et al., 2013; Rice et al., 2017).

A number of studies have shown that a variety of trait variables, including broad personality traits (especially high neuroticism/negative affectivity, low extraversion/positive affectivity, and low conscientiousness) (Goldstein et al., 2018), negative cognitive styles (e.g., rumination) (Cohen et al., 2019b; Mac Giollabhui et al., 2018; Stange et al., 2016), and other dispositional constructs (e.g., self-criticism and dependency) prospectively predict adolescent first-onsets of DD (Kopala-Sibley et al., 2017). However, these variables have been most often studied in isolation, both from each other and from other domains of risk factors.

Interpersonal factors have long been posited to play an important role in the development of youth depression (Rudolph et al., 2008). Interpersonal variables commonly considered risk factors for depression in adolescence include problematic family relationships, parental criticism, low parental warmth, lack of family and peer social support, and peer victimization (Bowes et al., 2015; Burkhouse et al., 2012; Griffith et al., 2019; Schwartz et al., 2014; Shanahan et al., 2011; Silk et al., 2009; Stice et al., 2004; Van Voorhees et al., 2008). However, there are very few data on whether interpersonal factors prospectively predict adolescent first-onsets in community samples (Griffith et al., 2019; Schwartz et al., 2014; Shanahan et al., 2011).

A number of biological processes have been posited to play a role in the pathophysiology of DD (Foland-Ross et al., 2013; Forbes and Dahl, 2012; Proudfit et al., 2015; Toenders et al., 2019), but few studies have tested whether they predict first-onsets, especially in adolescence. Available studies have reported that elevated levels of cortisol in the morning, especially the cortisol awakening response (CAR), predict first-onsets of DD, including in adolescents (Adam et al., 2010; Foland-Ross et al., 2013), although this effect may decay over time (Vrshek-Schallhorn et al., 2013). In addition, studies using monetary reward tasks have reported that lower striatal activation (Keren et al., 2018) and blunting of the reward positivity (RewP) component of the event-related potential (ERP) prospectively predicted adolescent first-onsets (Bress et al., 2013; Nelson et al., 2016). Finally, some theorists have posited that DD is associated with diminished reactivity to negative, as well as positive, emotional stimuli (Rottenberg et al., 2005). Consistent with this, recent studies have reported that a blunted late positive potential (LPP), an ERP marker of emotional engagement, is associated with familial risk for depression (Nelson et al., 2015) and predicts an increase in depressive symptoms (Levinson et al., 2019) in adolescents, but no study has tested whether it prospectively predicts first-onset DD.

A likely explanation for these multiple perspectives on the pathogenesis of DD is that depression is a complex, multifactorial disorder characterized by equifinality (i.e., multiple risk pathways leading to the same condition) (Kendler et al., 2002; Thapar et al., 2012). Since many of the risk factors are interrelated, a wide panel of risk indicators from multiple domains should be studied to clarify whether they reflect independent equifinal processes or shared vulnerabilities. Yet, most of the studies investigating risk factors for adolescent first-onsets have examined only a small number of variables in a few domains. A recent study, which investigated the unique contribution of multiple risk factors from the broadest set of domains to date in a community sample, found that rumination, social/academic impairment, and negative affectivity independently contributed to first-onsets (Cohen et al., 2019b). However, like most first-onset studies, this study relied solely on interviews and questionnaires, and did not incorporate laboratory tasks and biological predictors. Initial evidence indicates that integrating easily-obtainable biological indicators (e.g., EEG, pupil dilation) may help further understand the processes leading to first-onsets (Cohen et al., 2019a; Nelson et al., 2016). Therefore, more studies investigating putative biological vulnerabilities along with broad assessments of other domains of risk are warranted.

Another issue with existing studies is that, with rare exceptions (Shanahan et al., 2011; Vrshek-Schallhorn et al., 2013), they have not considered the interval over which risk factors exert their effects on first-onset. While some factors may have persisting effects over time (e.g., family history of psychopathology), the effects of other risk factors may change over time (e.g., due to developmental effects) (Laceulle et al., 2014). Risk factors with more time-limited effects would thus require more frequent re-assessments. This information is critical for selecting assessments and follow-up intervals in future longitudinal and intervention studies.

Finally, adolescent first-onset studies have rarely examined the course of DD after first-onset (Wilson et al., 2014), and therefore could not identify risk factors that predict the beginning of a chronic/recurrent course. Since chronic/recurrent DD is associated with particularly negative outcomes, including suicidality (Hammen et al., 2008), the identification of risk factors able to distinguish, when youths are still unaffected, those subsequently developing these more malignant forms of DD from those without a subsequent DD onset is a high priority.

The present study is the first to investigate the prospective association of a range of etiologically-relevant risk factors from multiple domains, sources, and units of analysis (including clinical interviews, self- and parent-report, coding of speech samples, and hormonal and neural measures) with first-onset of DD in a multi-wave community sample of adolescent females. Previous studies in this cohort investigated a few baseline measures (depressive symptoms, dispositional traits, family history, RewP) as risk factors for first-onset DD over 18 months (Goldstein et al., 2018; Kopala-Sibley et al., 2017; Nelson et al., 2016). Here, we extend previous studies in this and other adolescent samples in three major ways. First, we investigated a broader set of putative risk factors for first-onset over a 3-year period spanning adolescence, including several risk indicators not previously examined in relation to adolescent-onset DD (peer victimization, parental criticism, low social support and parental warmth, and LPP). As we selected variables based on theoretical associations with the development of DD, we hypothesized that they would each significantly predict first-onsets. Moreover, we investigated which risk factors predict DD independently of others. Second, we examined whether each risk factor predicted first-onset only in the shorter term, or for the full duration of the study, with the tentative hypothesis that more persistent and trait-like risk factors (e.g., family history, dispositional variables) would show the most enduring associations. Due to absence of guidelines in the literature, we arbitrarily divided the follow-up period at the mid-point (18 months) and investigated the prospective associations of risk factors with first-onset up to 18 months vs. no onset and first-onset after 18 months vs. no onset. Finally, we sought to identify risk factors that would distinguish, in a sample of unaffected young adolescents, those who would develop an adolescent first-onset DD with a chronic or recurrent course from those without a first-onset. Based on the literature on chronic and recurrent depression, conducted primarily in adults with depression (Klein and Allmann, 2014), we hypothesized that antecedent anxiety disorder, high neuroticism/negative affectivity, low extraversion/positive affectivity, self-criticism, rumination, interpersonal risk factors, and parental mood disorder would predict the first-onset of a chronic/recurrent form of DD in adolescence.

METHODS

Sample

Participants were 550 girls aged 13.5–15.5 (mean=14.39, SD=.63) from the Adolescent Development of Emotions and Personality Traits (ADEPT) cohort, who completed assessments with a biological parent (93.1% mothers; aged 31–59, mean=46.52, SD=4.52). Participants were recruited from the community in Suffolk County, New York using a commercial mailing list of homes with 13–15-year-old girls, word of mouth, local referral sources (e.g., school districts), online advertisements, and community postings. Inclusion criteria were fluency in English, ability to complete questionnaires, and a biological parent willing to participate. Exclusion criteria were intellectual disability and lifetime history of major depressive disorder (MDD) or dysthymic disorder, as the study aimed to predict first-onset DD. Adolescents and parents provided written informed consent after receiving a complete description of the study and were financially compensated for participating. The study was approved by Stony Brook University’s Institutional Review Board.

In-person visits were conducted at baseline (wave 1), month 18 (wave 3) and 36 (wave 5), and phone assessments at month 9 (wave 2) and 27 (wave 4). Each follow-up was completed with high participation rates (95.8%, 94.5%, 92.4%, and 92.2%). Participants with a history of DD not otherwise specified (DD-NOS) at baseline (n=35) were excluded for the current analyses aiming to predict first-onset DD, including DD-NOS (see ‘DD diagnosis’). Furthermore, participants who completed baseline assessments, did not report a first-onset in subsequent waves, but did not participate in the final wave (n=36) were excluded, as it was impossible to know if they had an onset in the 3-year period. The final sample consisted of 479 adolescent participants without a history of DD at baseline.

Measures

Baseline risk factors

The ADEPT study was explicitly designed to investigate risk factors for depression in adolescent girls, and therefore maximized baseline assessments of female-specific risk indicators in adolescence. A full description of all measures and their psychometric properties is provided in Supplementary Material 1.1. Briefly, prior history of any DSM-IV anxiety disorders or behavioral disorder (oppositional defiant, conduct, and attention-deficit/hyperactivity disorder) was ascertained with the Kiddie Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL) conducted with the participants (Kaufman et al., 1997). Adolescents further completed questionnaires on depressive symptoms (Inventory of Depression and Anxiety Symptoms, expanded version; Watson et al., 2012), irritability/hostility (Buss-Perry Aggression Scale; Buss and Perry, 1992), rumination (Response Styles Questionnaire; Nolen-Hoeksema and Morrow, 1991), self-criticism (Depressive Experiences Questionnaire; Blatt et al., 1995), interpersonal dependency (Interpersonal Dependency Inventory; Hirschfeld et al., 1977), personality traits (neuroticism/negative affectivity, extraversion/positive affectivity, and conscientiousness) (Big Five Inventory; Soto and John, 2017), relationships with their parents and best friend (Network of Relationship Inventory - Relationship Qualities Version; Furman and Buhrmester, 2009), perceived social support (Multidimensional Scale of Perceived Social Support; Zimet et al., 1990), parental warmth (Parental Bonding Instrument; Parker, 1979), and peer victimization (Revised Peer Experiences Questionnaire; De Los Reyes and Prinstein, 2004). Parental lifetime history of DSM-IV mood disorder was assessed through interviews with participating parents about themselves (Structured Clinical Interview for DSM-IV Axis I Disorders; First et al., 2002) and the non-participating parents (Family History Screen; Weissman et al., 2000). Parental criticism was assessed through the Five-Minute Speech sample with participating parents (Magana et al., 1986). Participant’s EEG was recorded with a 34-electrode system during a reward sensitivity task (Doors task) (Nelson et al., 2016) and an emotional picture-viewing task (Nelson et al., 2015). We measured the RewP in response to monetary gains relative to losses (gain-loss difference score) from the Doors task and LPP to pleasant and unpleasant pictures (analyzed separately) from the emotional picture-viewing task (Figure 1). CAR was measured as the difference between salivary cortisol at waking and 30 minutes after.

Figure 1.

Grand average of the reward positivity (RewP) (top) and late positive potential (LPP) (bottom) in the chronic/recurrent, non-chronic single-episode, and no-onset groups.

DD diagnosis

A lack of DD history at study baseline was established with the K-SADS-PL. Informant report, elicited from the participating parent, was used to complement the adolescent’s report. The K-SADS-PL was also used to assess and precisely date DD onsets at subsequent assessments at 9-month intervals (waves 2–5). In line with previous studies (Nelson et al., 2016; Shanahan et al., 2011), we included DSM-IV diagnoses of major depressive episode (MDE), dysthymic disorder, and DD-NOS in our first-onset DD group. DD-NOS was operationalized as a clinically significant depressive episode, characterized by presence of cardinal symptoms along with suicidality, clinically significant impairment or need for treatment, that did not quite meet the full criteria for MDE or dysthymic disorder based on DSM-IV (e.g., too brief duration, such as 10 months of dysthymic disorder, or too few symptoms, such as 4 symptoms of an MDE). Since evidence indicates that the majority of adolescents with DD-NOS subsequently meet full criteria for MDE or dysthymic disorder (Klein et al., 2009), investigation of risk factors predating a DD-NOS episode in adolescence is especially important. If participants did not participate in one or more waves, but returned for a subsequent wave, the period since the last completed wave was used to maintain full coverage over 3 years. All interviews were administered by trained interviewers closely supervised by clinical psychologists. Inter-rater reliability for any DD diagnosis based on 48 audio-recorded interviews independently scored by a second rater at wave 1 or wave 3 was excellent (kappa=.81). Among first-onset DD participants, we also identified those with a chronic/recurrent course (defined as having experienced more than one MDE or DD-NOS episode, an MDE or DD-NOS lasting ≥12 months, or dysthymic disorder). These forms of chronic/recurrent depression were grouped together based on previous evidence of substantial overlap in correlates and risk factors for chronic and recurrent DD (Klein and Allmann, 2014; Schramm et al., 2020). For further details, see Supplementary Material 1.1.

Statistical analyses

All analyses were conducted with SPSS 25 (IBM, Armonk, N.Y.) and R. The hypothesized risk factors were tested individually in bivariate logistic regressions predicting first-onset DD vs. no onset. Exploratory analyses tested the endurance of the prospective association of each measure with multinomial logistic regressions contrasting first-onset up to 18 months and over 18 months after baseline vs. no onset as the reference group.

Measures significantly associated with first-onset after multiple-testing correction (5% false discovery rate [FDR]) in bivariate logistic regressions were standardized and forward-entered into a multivariate logistic regression to identify risk factors uniquely associated with first-onsets. The combined effect of the resulting unique risk factors was evaluated with the area under the curve (AUC) in a receiver operating characteristic analysis, where AUC values of .56, .64, and .71 correspond to small, medium, and large effects, respectively (Rice and Harris, 2005) and Nagelkerke R². We further examined whether the combined unique risk factors predicted time until onset in a survival analysis (Cox regression).

Multinomial logistic regressions were also used to predict first-onset of chronic/recurrent DD and non-chronic single-episode DD groups compared to no onset (reference group). We ran these analyses on each risk factor individually, and then FDR-significant measures were entered in multivariate logistic regressions (comparable to those on any first-onset) contrasting each DD group with the no-onset group separately. Eight participants who had a first-onset but did not participate in the final assessment were excluded from these analyses, as it was impossible to determine their course.

RESULTS

Ninety-three (19.42%) participants experienced a first-onset DD (36 MDE, 20 dysthymic disorder, 37 DD-NOS) over 3 years, an average of 14.66 (SD=11.07) months after baseline. Among participants with a DD-NOS first-onset, 7 participants met criterial for MDE (n=3), dysthymic disorder (n=2), or both (n=2) at subsequent waves. Participants with and without a first-onset did not differ on baseline age, socio-demographic characteristics (e.g. ethnicity, parental education, household income), pubertal status, and body-mass index (Table 1). These variables were therefore not included in subsequent analyses.

Table 1.

Demographic characteristics and group comparisons between no-onset and first-onset groups.

	No-onset		First-onset		No-onset vs. First-Onset
	Mean	SD	Mean	SD	t	p
Age (baseline)	14.38	.63	14.36	.63	.28	.78
Age (final assessment)	17.49	.66	17.50	.69	−.05	.96
BMI (baseline)	21.73	4.11	22.04	4.02	−.65	.52
Total PDS (baseline)	12.86	1.98	13.18	1.85	−1.40	.16
	N	%	N	%	Chi2	p
European American	344	89.12	80	86.02	.71	.40
Hispanic	36	9.33	14	15.05	2.63	.10
Both parents in home (baseline)	343	88.86	77	82.80	2.55	.11
Annual household income (baseline)					4.50	.11
Less than $60,000	36	10.29	15	17.05
$60,000 – $120,000	139	39.71	38	43.18
$120,000 or more	175	50.00	35	39.77
Parental education (baseline)					3.96	.27
Both parents completed college	133	34.64	33	35.87
One parent completed college	128	33.33	22	23.91
No parent completed college	123	32.03	37	40.22

Abbreviations: BMI, body mass index; PDS, Pubertal Development Scale total (mean) score; SD, standard deviation. Notes: Data on household income were available for 438 participants, on parental education for 476 participants, on BMI for 462 participants, and on PDS for 471 participants, due to some participants choosing not to provide this information.

In bivariate logistic regressions, first-onsets were prospectively predicted by all investigated predictors (all p<.05 survived FDR-threshold; Table 2), except RewP, LPP to pleasant pictures, CAR, and a history of behavioral disorders.

Table 2.

Descriptive statistics and results of individual logistic regressions on baseline risk factors.

	No-onset		Chronic/recurrent		Non-chronic single-episode		First-Onset vs. No-onset			Chronic/recurrent vs. No-onset			Non-chronic single-episode vs. No-onset
	M/N	SD/%	M/N	SD/%	M/N	SD/%	OR	95% CI	p	OR	95% CI	p	OR	95% CI	p
Clinical/Family history
Depressive symptomsa	1.53	.51	2.05	.75	1.91	.58	1.97	1.58–2.45	<.01	2.18	1.67–2.84	<.01	1.85	1.34–2.54	<.01
History of anxiety disorderb	81	20.98	19	36.54	12	36.36	1.39	1.13–1.72	<.01	1.39	1.07–1.81	.01	1.39	1.01–1.91	.05
History of behavioral disorderb	16	4.15	4	7.69	2	6.06	1.14	.94–1.39	.18	1.15	.90–1.47	.26	1.09	.79–1.51	.61
Parental mood disorderb	84	21.76	14	26.92	16	48.48	1.31	1.06–1.62	.01	1.13	.85–1.50	.40	1.69	1.24–2.30	<.01
Cognitive/Dispositional
Ruminationa	1.53	.52	1.98	.71	1.81	.64	1.80	1.45–2.24	<.01	1.98	1.52–2.57	<.01	1.60	1.15–2.21	<.01
Self-criticisma	1.94	.79	2.53	.96	2.16	.81	1.67	1.34–2.09	<.01	1.92	1.45–2.53	<.01	1.32	.93–1.87	.12
Dependencya	2.21	.67	2.54	.76	2.35	.72	1.46	1.16–1.83	<.01	1.61	1.21–2.14	<.01	1.23	.87–1.75	.24
Irritability/Hostilitya	2.11	.76	2.44	.68	2.54	.80	1.62	1.30–2.04	<.01	1.52	1.14–2.01	<.01	1.70	1.21–2.38	<.01
Personality
Neuroticisma	2.61	.77	3.15	.81	2.95	.70	1.89	1.48–2.41	<.01	2.02	1.49–2.75	<.01	1.57	1.09–2.26	.02
Conscientiousnessa	3.75	.65	3.39	.70	3.55	.49	.61	.48–.76	<.01	.57	.43–.77	<.01	.73	.51–1.04	.08
Extraversiona	3.82	.79	3.38	.75	3.77	.74	.71	.57–.89	<.01	.58	.44–.78	<.01	.93	.65–1.33	.70
Interpersonal
Parental criticisma	.18	.55	.32	.71	.32	.65	1.27	1.04–1.55	.02	1.24	.95–1.61	.12	1.24	.91–1.70	.18
Social supporta	5.93	.95	5.35	1.01	5.7	1.05	.68	.55–.85	<.01	.60	.46–.78	<.01	.79	.56–1.11	.18
Poor relationship with parentsa	1.82	.58	2.21	.72	1.92	.58	1.55	1.24–1.94	<.01	1.78	1.36–2.32	<.01	1.19	.83–1.70	.34
Parental warmtha	3.43	.45	3.15	.53	3.39	.39	.70	.56–.88	<.01	.58	.45–.76	<.01	.92	.64–1.32	.66
Poor relationship with best frienda	1.49	.42	1.76	.52	1.47	.28	1.41	1.13–1.75	<.01	1.67	1.29–2.16	<.01	.95	.64–1.40	.79
Peer victimizationa	1.35	.38	1.55	.59	1.44	.43	1.42	1.16–1.75	<.01	1.48	1.16–1.89	<.01	1.25	.90–1.73	.19
Biological
LPP (unpleasant pictures)a	7.93	6.58	6.72	5.11	6.03	6.89	.76	.60–.98	.04	.82	.60–1.13	.23	.73	.49–1.07	.11
LPP (pleasant pictures)a	5.56	6.24	4.96	5.55	3.59	7.08	.80	.63–1.02	.07	.90	.67–1.23	.53	.72	.49–1.05	.09
RewP (gain-loss)a	5.26	5.85	2.96	3.94	5.87	6.35	.81	.63–1.03	.09	.64	.46–.89	.01	1.11	.77–1.60	.60
Cortisol awakening responsea	8.22	5.7	8.89	5.35	6.85	4.85	1.03	.82–1.29	.82	1.12	.84–1.50	.43	.76	.51–1.14	.19

Abbreviations: LPP, late positive potential responses averaged across pleasant and unpleasant pictures (mean activity at parieto-occipital sites between 300–1000 ms); M, mean; RewP, reward positivity (difference mean activity between gains and losses at FCz between 250–350 ms); N, number of subjects with risk factor present; OR, odds ratio; SD, standard deviation; 95% CI, 95% confidence interval around OR.

Notes:

^aFor continuous risk factors, mean and standard deviation are reported.

^bFor categorical risk factors, number and % of subjects with risk present (e.g., history of anxiety disorder) are reported.

Bold indicates p<.05 (two-tailed) significant after multiple testing correction (5% false-discovery rate [FDR]). Descriptive statistics for questionnaires were calculated on average item responses.

A total of 60 participants had a first-onset in the former 18 months of the study (n=35 between waves 1–2, n=25 between waves 2–3), while 33 had an onset in the latter 18 months (n=15 between waves 3–4, n=18 between waves 4–5). All significant risk factors for first-onset DD were significantly associated with first-onsets up to 18 months, except parental history of mood disorders, which did not reach statistical significance (p=.051) (Table 3). In addition, RewP was associated with first-onsets in the first half of the follow-up period. All clinical, cognitive/dispositional, and personality measures were also associated with onsets after 18 months, except extraversion, dependency and parental mood disorder history, which did not show statistically significant effects (Table 3). Conversely, interpersonal and biological measures were significantly associated with first-onsets up to 18 months (except the LPP to pleasant pictures and CAR, which showed no significant effects), but were not predictive after 18 months. Results contrasting the groups with first-onset up to 18 months (n=60) vs. over 18 months (n=33) are reported in Table S2 for completeness.

Table 3.

Multinomial logistic regressions on baseline risk factors comparing participants with no onset with those with first-onset DD up to 18 months and >18 months after baseline.

	First onset up to 18 months		First onset >18 months		No-onset vs. First onset up to 18 months			No-onset vs. First onset >18 months
	M/N	SD/%	M/N	SD/%	OR	95% CI	p	OR	95% CI	p
Clinical/Family history
Depressive symptomsa	2.02	.73	1.92	.63	2.05	1.59–2.63	<.01	1.84	1.34–2.54	<.01
History of anxiety disorderb	21	35.00	13	39.39	1.35	1.05–1.74	.02	1.47	1.07–2.01	.02
History of behavioral disorderb	4	6.67	3	9.09	1.11	.87–1.42	.39	1.20	.91–1.58	.20
Parental mood disorderb	20	33.33	12	36.36	1.29	1.00–1.66	.05	1.36	.99–1.88	.06
Cognitive/Dispositional
Ruminationa	1.97	.71	1.81	.67	1.94	1.51–2.49	<.01	1.58	1.14–2.19	<.01
Self-criticisma	2.49	.94	2.24	.91	1.84	1.42–2.39	<.01	1.43	1.02–2.01	.04
Dependencya	2.48	.74	2.44	.73	1.48	1.13–1.94	<.01	1.41	.99–2.00	.06
Irritability/Hostilitya	2.52	.72	2.49	.76	1.64	1.26–2.13	<.01	1.58	1.13–2.20	<.01
Personality
Neuroticisma	3.14	.78	3.02	.76	1.99	1.49–2.66	<.01	1.70	1.18–2.45	<.01
Conscientiousnessa	3.40	.70	3.47	.43	.60	.45–.78	<.01	.65	.45–.93	.02
Extraversiona	3.47	.75	3.69	.75	.65	.50–.85	<.01	.85	.60–1.20	.36
Interpersonal
Parental criticisma	.37	.92	.34	.65	1.26	1.01–1.57	.04	1.26	.94–1.70	.13
Social supporta	5.81	.83	5.37	1.09	.61	.48–.78	<.01	.88	.62–1.25	.48
Poor relationship with parentsa	2.25	.74	1.83	.52	1.84	1.43–2.36	<.01	1.03	.71–1.50	.88
Parental warmtha	3.17	.53	3.40	.37	.61	.47–.78	<.01	.94	.65–1.35	.72
Poor relationship with best frienda	1.70	.51	1.59	.48	1.53	1.20–1.96	<.01	1.25	.89–1.75	.19
Peer victimizationa	1.57	.56	1.42	.43	1.54	1.22–1.95	<.01	1.20	.85–1.69	.31
Biological
LPP (unpleasant pictures)a	5.83	5.52	7.16	6.32	.70	.51–.95	.02	.89	.61–1.29	.52
LPP (pleasant pictures)a	3.79	5.82	4.96	6.49	.74	.55–1.00	.05	.91	.63–1.31	.60
RewP (gain-loss)a	3.33	5.04	5.52	6.48	.69	.51–.94	.02	1.04	.72–1.50	.82
Cortisol awakening response	8.67	5.32	7.83	5.94	1.08	.82–1.42	.58	.93	.65–1.35	.71

Abbreviations: LPP, late positive potential (mean activity at parieto-occipital sites between 300–1000 ms); M, mean; N, number of subjects with risk factor present; RewP, reward positivity (difference mean activity between gains and losses at FCz between 250–350 ms); OR, odds ratio; SD, standard deviation; 95% CI, 95% confidence interval around OR.

Notes:

^aFor continuous risk factors, mean and standard deviation are reported.

^bFor categorical risk factors, number and % of subjects with risk present (e.g., history of anxiety disorder) are reported.

Bold indicates p values significant after multiple testing correction (5% false-discovery rate [FDR]). Descriptive statistics for questionnaires were calculated on average item responses.

FDR-significant measures individually predicting first-onset (Table 2) were included in a multivariate logistic regression. These risk factors were moderately correlated but not redundant (Table S1). Four risk factors emerged as unique predictors: depressive symptoms, rumination, parental criticism, and parental mood disorder (Table 4). These four unique risk factors combined showed a Nagelkerke R²=.17 and AUC=.73 (95% confidence interval [CI]=.66–.79; Figure S1). Survival analyses showed that higher scores on the composite of risk factors predicted shorter time to first-onset (hazard ratio=1.92, 95% CI=.45–.91, p<.001) (Figure S2).

Table 4.

Results of multivariate logistic regressions of first-onset depressive disorder (top), chronic/recurrent depressive disorder (middle), and non-chronic single episode depressive disorder (bottom).

First-onset vs. No-onset
	Beta	OR	95% CI	p
Depressive symptoms	.47	1.60	1.20–2.15	<.01
Parental mood disorder	.27	1.31	1.04–1.65	.02
Parental criticism	.26	1.30	1.05–1.60	.02
Rumination	.32	1.37	1.02–1.85	.04
Chronic/recurrent vs. No-onset
	Beta	OR	95% CI	p
Depressive symptoms	.66	1.93	1.46–2.56	<.01
Poor relationship with best friend	.44	1.55	1.16–2.07	<.01
Extraversion	−.43	.65	.47–.90	<.01
RewP	−.52	.59	.41–.86	<.01
Non-chronic single-episode vs. No-onset
	Beta	OR	95% CI	p
Depressive symptoms	.59	1.81	1.30; 2.52	<.01
Parental mood disorder	.49	1.63	1.19; 2.24	<.01

Abbreviations: RewP, reward positivity (difference mean activity between gains and losses at FCz between 250–350 ms); OR, odds ratio; 95% CI, 95% confidence interval around OR.

Notes: Bivariate predictors significant based on 5% false discovery rate (FDR) multiple testing correction in Table 2 were standardized and included in the forward entry multivariate logistic regression models. Only significant predictors in the multivariate model are reported here. Multivariate logistic regressions were not run for the comparison between single-episode and chronic/recurrent groups as no measure in simple logistic regressions (Table S3) was significant based on FDR multiple testing correction.

Among participants with a first-onset, 61% (n=52) experienced a chronic (n=35) or recurrent (n=17) DD during the 3-year follow-up period (mean=9.53, SD=7.71 months after baseline; n=26 between waves 1–2, n=19 between waves 2–3, n=6 between waves 3–4, n=1 between waves 4–5). The remaining participants experienced a non-chronic single episode DD (n=33) (mean=22.47, SD=11.74 months after baseline; n=7 between waves 1–2, n=3 between waves 2–3, n=6 between waves 3–4, n=17 between waves 4–5). Risk factors associated with any first-onset were also associated at FDR-significance threshold with first-onset chronic/recurrent DD (vs. no onset), except for LPP to unpleasant pictures, parental criticism, parental mood disorder (Table 2). Blunted RewP (Figure 1) emerged as a further significant risk factor in this analysis. Only four risk factors (depressive symptoms, rumination, irritability/hostility, and parental mood disorder) distinguished the non-chronic single-episode group from the no-onset group after FDR corrections (Table 2). Analyses contrasting chronic/recurrent vs. non-chronic single-episode groups were also run for completeness and repeated controlling for age at first-onset, which may be a confounding factor (Wilson et al., 2014). The lack of statistically significant differences between these groups (Table S3) likely arises from insufficient power in this subsample (n=85), since these effects are expected to be small (Klein and Allmann, 2014).

Multiple logistic regressions on FDR-significant measures identified four independent predictors of first-onset chronic/recurrent DD: depressive symptoms, low extraversion, poor relationship with best friend, and blunted RewP (Table 4). These four unique risk factors combined showed a Nagelkerke R²=.22 and AUC=.79 (95% CI=.73–.85) (Figure S1). Only subthreshold depression and parental mood disorder emerged as unique risk factors for non-chronic single-episode DD vs. no-onset (Table 4); combined, they showed a Nagelkerke R²=.12 and AUC=.72 (95% CI=.64–.81) (Figure S1).

A series of sensitivity analyses were conducted to ascertain the robustness of our findings: excluding two participants who experienced mania after a DD onset (thus reclassified as bipolar disorder) (Supplementary Material 2.1);excluding participants with a first-onset DD-NOS (Supplementary Material 2.2); removing items of the RSQ which may be confounded with depression itself (Treynor et al., 2003) (Supplementary Material 2.3); computing the RewP as a residualized score, rather than a difference score (Supplementary Material 2.4); and controlling for baseline age in analyses of risk factors for first onsets up to 18 months and >18 months vs. no onset (Supplementary Material 2.5). Results of analyses excluding BD cases, using the residual RewP, using a different scoring of the RSQ, and controlling for age (Supplementary Material 2.1, 2.3, 2.4, 2.5) were fully consistent with those of the main analyses. For most risk factors, results excluding DD-NOS participants were consistent with those contrasting any first-onset DD with no onset (Supplementary Material 2.2).

DISCUSSION

In this multi-wave longitudinal study of adolescent girls, we investigated the prospective associations of a comprehensive panel of putative clinical, cognitive/dispositional, personality, interpersonal, and biological risk factors with adolescent first-onset DD. Most hypothesized risk factors, including a number examined for the first time in association with adolescent first-onset DD in community samples, were prospectively associated with first-onset over 3 years. When investigating all risk factors jointly, adolescents’ depressive symptoms, rumination, parental mood disorder history, and parental criticism emerged as independently predicting adolescent first-onset DD. In analyses investigating risk factors associated with a first-onset DD with a chronic/recurrent course vs. no onset, we found that depressive symptoms, low extraversion, poor peer relationships, and blunted neural reward sensitivity may be risk factors independently contributing to this especially clinically significant and burdensome form of depression. The identified risk factors may represent multiple vulnerabilities independently leading to first-onset DD in adolescent girls, and promising targets for multi-component prevention strategies for youth at risk for adolescent-onset depression.

Our study provides key evidence on a diverse set of risk factors individually proposed in etiological theories of depression (Forbes and Dahl, 2012; Hankin, 2012; Kendler et al., 2002; Klein et al., 2011; Nolen-Hoeksema, 1987; Rottenberg et al., 2005; Rudolph et al., 2008), by showing their prospective association with adolescent first-onset DD. Specifically, this study shows novel evidence that interpersonal factors, including peer victimization, poor peer relationships, and low perceived social support, are prospectively associated with adolescent first-onset DD in a community sample. We further found that parental criticism, poor relationships with parents, and low parental warmth predict adolescent first-onsets of DD, in line with initial evidence from previous studies in high-risk offspring or community samples (Burkhouse et al., 2012; Schwartz et al., 2014; Shanahan et al., 2011). With the inclusion of easily obtainable neurobiological markers, we further report, for the first time, that lower brain reactivity (blunted LPP) in response to unpleasant pictures predicts subsequent first-onset DD. A similar pattern, although not reaching statistical significance, emerged also for the LPP to pleasant pictures. These findings are consistent with initial evidence showing associations of blunted LPP with familial risk for depression (Nelson et al., 2015) and increased depressive symptoms following stress (Levinson et al., 2019). Our findings extend this prior literature by showing that lower emotional engagement in early adolescence may predispose girls to an adolescent first-onset DD.

By examining the recency of risk and durability of predictive effects, we found that RewP, a marker of reward sensitivity previously associated with first-onset depression (Keren et al., 2018; Nelson et al., 2016), did not predict first-onsets over 3 years, but only up to 18 months. The current study examined a longer follow-up period compared to previous studies showing that the RewP prospectively predicts first-onset DD (Bress et al., 2013; Nelson et al., 2016). As such, it is possible that the RewP, like other variables only predictive over the first 18 months (i.e., LPP to unpleasant pictures and interpersonal factors), has time-limited predictive effects and requires frequent re-assessments. Another possible explanation is that the group with a first-onset up to 18 months after baseline was twice as large as the group with a first-onset in the remaining 18 months. In contrast, most clinical, cognitive/dispositional, and personality measures predicted first-onsets also in participants with a first-onset in the second half of the study, suggesting potentially more persistent effects.

Our multivariate results help identify the independent processes from multiple domains that uniquely contribute to adolescent first-onset DD, and potentially represent equifinal processes leading to first-onset. We found that depressive symptoms, rumination, parental mood disorder, and parental criticism each contributed unique effects in predicting first-onset DD in adolescent females, and may therefore represent key independent vulnerabilities implicated in the pathogenesis of adolescent-onset depression. Furthermore, higher levels of risk based on these four risk factors was associated with a faster time to first-onset over the 3-year follow-up. These findings highlight the importance of monitoring these personal and parental risk factors in youth. Of note, all other risk factors were not significant in multivariate analyses, indicating that their individual association with first-onset likely reflect shared vulnerability with the measures that emerged as unique risk factors. Although developing a prediction algorithm was not an aim of this study, we note that the four significant variables collectively yielded a large effect size and moderate prediction accuracy based on AUC, suggesting that they are promising measures for future prediction models of adolescent-onset DD in community samples. Yet, these predictors explained a modest amount of variance in first-onset DD, suggesting that additional factors are likely to play a role. Furthermore, these results have clinical implications for indicated or selective preventive interventions for adolescent-onset depression (Ssegonja et al., 2019), which could target a combination of adolescents’ depressive symptomatology and rumination, as well as their parents’ mood and tendency to be critical of their daughters.

Our study further sheds new light on the risk factors that, in early adolescence, predict first-onset DD with a subsequent chronic/recurrent course vs. no onset. Since these forms of DD are associated with particularly high morbidity and suicidality (Greden, 2001; Hammen et al., 2008), it is important to identify risk factors when youth are still unaffected. We therefore investigated baseline risk factors for chronic/recurrent DD in adolescents without prior DD, rather than taking the approach of predicting the course of depression after first-onset more commonly employed in clinical samples (Kovacs et al., 2016). Many risk factors for any first-onset DD, as well as RewP, were individually associated with first-onset chronic/recurrent DD. The significant effect of RewP on chronic/recurrent first-onset DD, but not on any first-onset, is consistent with its significant predictive effects up to 18 months, since most participants with a chronic/recurrent course experienced a first-onset in the first half of the study period. The unexpected non-significant effect of parental mood disorder (Klein and Allmann, 2014) might be explained by our study design: since high family risk and a chronic/recurrent course are both associated with especially early first-onset in childhood or early adolescence (Klein and Allmann, 2014; Rice et al., 2017), our study recruiting adolescents with no prior DD might have excluded a portion of individuals with high familial risk and at risk for developing chronic/recurrent DD. Future studies with larger samples of participants with chronic/recurrent first-onset DD are needed to investigate the unique effect of parental mood disorders, as well as of different forms of parental mood disorder (e.g., maternal vs. paternal, with vs. without recurrence/chronicity), on offspring chronicity/recurrence.

In multivariate analyses, low positive affectivity (extraversion), aberrant reward processing (blunted RewP), poor peer relationships, and subthreshold depressive symptoms emerged as unique risk factors for first-onset of chronic/recurrent DD vs. no onset, with a large effect size based on AUC. The former three might represent unique risk factors for chronic/recurrent DD, but not for first-onset DD in general. These results are consistent with previous studies in adults with depression indicating that chronic depression is associated with lower positive affect (Klein and Allmann, 2014; Kotov et al., 2010). Based on these findings, multi-component preventive strategies for chronic/recurrent forms of adolescent-onset DD should not only aim to reduce subthreshold symptoms and improve interpersonal peer relationships, but also to upregulate positive affectivity and reward sensitivity (Craske et al., 2019).

The following limitations should be considered. First, this study only focused on adolescent females, and therefore does not provide information on males, children or adults. However, the incidence of first-onset DD increases dramatically in adolescence, especially in females (Salk et al., 2017). The high rate of first-onset DD (19.4%, including MDE, dysthymia, and DD-NOS diagnoses) in our adolescent female sample is consistent with other multi-wave longitudinal studies of adolescent community samples (Cohen et al., 2019b; Hankin et al., 1998; Rohde et al., 2013; Shanahan et al., 2011). Therefore, our study provides novel insights into the population group at highest risk - adolescent girls. Second, our sample, despite closely resembling the geographic area from which it was recruited, was predominantly middle-class and European American, which may explain the non-significant differences between first-onset and no-onset groups on socio-economic characteristics. Future studies on more diverse samples are needed. Third, although our five assessments spanned adolescence, and many first-onsets were detected, some participants in the no-onset and non/chronic single-episode groups may, respectively, experience a first-onset and develop a chronic/recurrent DD, in the future. Fourth, chronicity/recurrence and younger age of onset are difficult to disentangle (Wilson et al., 2014), and as expected girls in the chronic/recurrent group had earlier onsets than girls in the non-chronic single-episode group. Future studies should stratify on these variables to distinguish their effects. Fifth, our analyses of risk factors for first-onset up to 18 months (n=60) and >18 months (n=33) and with a chronic/recurrent DD (n=52) and non-chronic single-episode DD (n=33) contrasted the no-onset group (n=386) with these smaller subgroups. Our sample was sufficiently powered (80%) to detect an OR=1.5 (n=386+60), OR=1.6 (n=386+51), and OR=1.75 (n=386+33) as statistically significant (alpha=.05). This may explain why risk factors showing smaller effects, such as parental mood disorder, in analyses predicting a first-onset in each half of the study vs. no onset, a first-onset with a chronic/recurrent DD vs. no onset, or a first onset with a non-chronic/single-episode DD vs. no onset , did not reach statistical significance. Future studies with larger groups of participants with DD are needed to confirm these findings. Future research should also examine the effects of the risk factors identified in this study on trajectories of continuous depressive symptoms. Finally, our broad assessments of risk factors across multiple domains and units of analyses did not include a history of maltreatment or abuse, hopelessness, and negative self-schema, despite including related constructs. Future studies are warranted to examine the effects of additional risk factors on adolescent-onset depression.

In conclusion, our findings provide novel insights into multiple etiologically-informative risk factors for first-onset and chronic/recurrent DD in adolescence. The identified subset of risk factors independently associated with first-onsets may represent core vulnerabilities contributing to the first-onset of DD in adolescent girls. Future efforts should investigate whether preventive strategies simultaneously targeting the multiple identified risk factors in early adolescence may reduce the incidence of depression among adolescents.

HIGHLIGHTS.

• We investigated etiologically-informative risk factors for first-onset depression in adolescent girls.

• Several clinical, cognitive, personality/dispositional, interpersonal, and biological risk factors were prospectively associated with first onsets.

• Subthreshold depressive symptoms, rumination, parental mood disorder, and parental criticism independently contributed to first onsets.

• The identified risk factors may be core vulnerabilities for adolescent-onset depression.