Unit9AssignQDA

Child-, adolescent- and young adult-onset depressions: differential risk factors in development?

L. Shanahan1*, W. E. Copeland2, E. J. Costello2 and A. Angold2

1 Department of Psychology, University of North Carolina at Greensboro, NC, USA 2 Developmental Epidemiology Program, Duke University Medical Center, Durham, NC, USA

Background. Previous research reported that childhood adversity predicts juvenile- onset but not adult-onset

depression, but studies confounded potentially genuine differences in adversity with differences in the recency with

which adversity was experienced. The current study paper took into account the recency of risk when testing for

differences among child-, adolescent- and young adult-onset depressions.

Method. Up to nine waves of data were used per subject from two cohorts of the Great Smoky Mountains Study

(GSMS; n=1004), covering children in the community aged 9–16, 19 and 21 years. Youth and one of their parents were interviewed using the Child and Adolescent Psychiatric Assessment (CAPA) between ages 9 and 16 ; these same

youth were interviewed using the Young Adult Psychiatric Assessment (YAPA) at ages 19 and 21. The most common

psychosocial risk factors for depression were assessed : poverty, life events, parental psychopathology, maltreatment,

and family dysfunction.

Results. Consistent with previous research, most childhood psychosocial risk factors were more strongly associated

with child-onset than with adolescent-/adult-onset depression. When potentially genuine risk differences among the

depression-onset groups were disentangled from differences due to the recency of risk, child- and young adult-onset

depression were no longer different from one another. Adolescent-onset depression was associated with few

psychosocial risk factors.

Conclusions. There were no differences in putative risk factors between child- and young adult-onset depression

when the recency of risk was taken into account. Adolescent-onset depression was associated with few psychosocial

risk factors. It is possible that some adolescent-onset depression cases differ in terms of risk from child- and young

adult-onset depression.

Received 23 September 2010 ; Revised 4 April 2011 ; Accepted 9 April 2011 ; First published online 6 May 2011

Key words : Depression, development, epidemiology, onset, psychosocial risk factors.

Introduction

Do child-, adolescent- and adult-onset depression

have the same risk correlates and precursors

(Kaufman et al. 2001)? The answer to this question is

unclear. Neurobiological and treatment research has

found that usually two, but not all three, of these

depression-onset groups share common correlates

(Kaufman et al. 2001), suggesting a complex picture of

both shared and non-shared pathways to the onset

of depression at different points in development.

If developmental subtypes of depression differed in

terms of risk, examining them separately for purposes

of biosocial research, prevention and intervention

would be important, as has been shown by research on

developmental subtypes of antisocial behaviors (for a

review, see Moffitt, 2006).

Psychosocial risk for child-, adolescent- and young

adult-onset depression

Juvenile-onset depression is associated with a range of

early psychosocial risk factors, including childhood

poverty (Gilman et al. 2003), life events (Jaffee et al.

2002), parental psychopathology (Jaffee et al. 2002),

maltreatment (Jaffee et al. 2002 ; Hill et al. 2004) and

family dysfunction (Hill et al. 2004). Indeed, youth

with early-onset depression seem to be characterized

by pervasive dysfunction throughout life (Jaffee et al.

2002 ; Hill et al. 2004 ; see also Kovacs et al. 1984 ;

Christie et al. 1988 ; Giaconia et al. 1994 ; Rao et al. 1995 ;

Kasch & Klein, 1996; Weissman et al. 1999). By con-

trast, the childhood psychosocial risk factor profile for

adult-onset depression has been found to be ‘similar

to that of the never-depressed’ (Jaffee et al. 2002,

* Address for correspondence : L. Shanahan, Ph.D., University of

North Carolina at Greensboro, Department of Psychology, PO Box

26170, Greensboro, NC 27402, USA.

(Email : [email protected])

Psychological Medicine (2011), 41, 2265–2274. f Cambridge University Press 2011 doi:10.1017/S0033291711000675

ORIGINAL ARTICLE

p. 215; Hill et al. 2004). These findings have been in-

terpreted as indicating that child- and adult-onset de-

pressions are likely to be etiologically distinct.

However, such a conclusion is premature because

such apparent differences in risk might merely reflect

differences in how recently risk factors were experi-

enced. Risk factors in the key studies were typically

assessed in childhood, but the depressogenic effects of

adversities are strongest during the period immedi-

ately following their occurrence (e.g. Brown & Harris,

1978 ; Kessler et al. 1997), so perhaps we should

not be surprised that childhood risk factors exerted

most of their effects in childhood. Fig. 1a illustrates

that potentially genuine differences in risk were con-

founded with differences in the recency of risk occur-

rence because the time elapsed between childhood

risk and juvenile-onset depression (path ‘a’) was

much less than the time elapsed between childhood

risk and adult-onset depression (path ‘b’).

We propose to test a ‘recency hypothesis ’, which

posits that the lack of strong associations between

childhood adversity and adult-onset depression oc-

curs not because child- and adult-onset depressions

genuinely differ in terms of risk, but because at every

age the depressogenic effects of adversities are mostly

time limited. For example, family dysfunction might

have depressogenic effects for a number of months or

years, but not longer. Thus, if measured in childhood,

it would be linked with child-onset depression, and

perhaps with adolescent-onset depression (Hill et al.

2004), but not with young adult-onset depression.

If measured in adolescence or young adulthood,

family dysfunction would, however, be linked with

adolescent- and perhaps with young adult-onset

depression.

Studies have used cut-offs ranging from 14 to

20 years of age to distinguish between juvenile- and

adult-onset depression (e.g. Jaffee et al. 2002 ; Gilman

et al. 2003 ; Hill et al. 2004). However, the major

increase in the prevalence of depression in females

occurs around age 13 in Western populations (e.g.

Angold et al. 2002), and research increasingly suggests

that adolescent-onset depressions may constitute their

own category (e.g. Kaufman et al. 2001 ; Copeland et al.

2009). Thus, placing adolescent-onset depressions

with either the child- or the adult-onset depressions

could mask adversity-onset links.

The present study attempts to eliminate the con-

found between the recency and risk differences

hypotheses by measuring the same psychosocial risk

factors occurring concurrently with and antecedently

to child-, adolescent- and young adult-onset de-

pressions (see Fig. 1b). According to the recency

hypothesis, the odds ratios (ORs) for paths ‘c ’ should

be similar in size to one another, as should the

ORs for paths ‘d’. According to the risk differences

(a)

(b)

Childhood Adolescence Adulthood

Childhood Adolescence

Adolescent risk

Adolescent-onset

Adult risk

Young adulthood

Childhood risk

Childhood risk

Childhood-onset

a b

Juvenile-onset Adult-onset

Adult-onset

Age

Age

c d

c c d

Fig. 1. Timing of risk in relation to depression onset. (a) The design of previous studies, with childhood risk predicting

juvenile- and adult-onset depression. (b) The design of the present study, with concurrent and antecedent risk predicting

child-, adolescent- and young adult-onset depression.

2266 L. Shanahan et al.

hypothesis, ORs for paths ‘c ’ should differ in size from

one another, as should the ORs for paths ‘d’.

Method

Sample and procedures

The Great Smoky Mountains Study (GSMS) is a

longitudinal study of the development of psychiatric

disorders in youth (Costello et al. 1996, 2003). The ac-

celerated cohort (Schaie, 1965), two-phase sampling

design and measures are described in detail elsewhere

(Costello et al. 1996). In brief, a representative sample

of 9-, 11- and 13-year-olds in western North Carolina

was selected using a household equal probability

design. In the screening phase the primary caregiver

completed a questionnaire containing items regarding

behavioral disorders from the Child Behavior Check-

list (Achenbach & Edelbrock, 1983). The interview

phase included all children scoring above a predefined

cut-off on this screen (designed to identify the most

pathological 25% of the population), along with a

10% random sample of the remainder. All age-eligible

American Indian children from the area were also

recruited. Data were collected on one cohort at ages

9 and 10, two cohorts at ages 11, 12 and 13, and all three

cohorts at ages 14, 15, 16, 19 and 21 years. Of the 1777

children recruited, 1420 agreed to participate (80%).

Across waves, an average of 82% of possible inter-

views were completed (75–94%). The present study

focuses on the two youngest GSMS cohorts (first as-

sessed at ages 9 and 11 respectively ; n=1004) because childhood assessments were available for these two

cohorts. Each subject was interviewed up to nine times.

Before each interview began, parent and child signed

informed consent/assent forms approved by the

Institutional Review Boards of Duke University

Medical Center and the Eastern Band of Cherokee

Indians.

Measures

Psychiatric disorders were assessed using (1) the

Child and Adolescent Psychiatric Assessment (CAPA;

Angold & Costello, 1995, 2000) up to age 16, and (2)

the upward extension of the CAPA, the Young Adult

Psychiatric Assessment (YAPA), at ages 19 and 21

(Angold et al. 1999). To minimize recall bias, the time

frame for determining the presence of most psychi-

atric symptoms is the 3 months immediately preced-

ing the interview. Scoring programs for the CAPA and

YAPA, written in SAS (SAS Institute, 2004), combined

information about the date of onset, duration and in-

tensity of each symptom to create diagnoses according

to DSM-IV. A symptom was counted as present if

reported by either parent or child up to age 16 or

by the young adult at ages o19 years. The 2-week test–retest reliability of CAPA diagnoses for 10- to

18-year-olds is comparable to that of other structured

diagnostic interviews (K values for individual dis-

orders range from 0.56 to 1.0 ; Angold & Costello,

1995). Consistent with previous relevant research, we

used age to distinguish among the depression-onset

groups (Jaffee et al. 2002; Hill et al. 2004). Using pub-

ertal status to define these groups resulted in only

minor changes. Child-onset depression was defined

as first reported diagnosis between ages 9 to <13, adolescent-onset as first reported diagnosis between

the ages of 13 to 16, and young adult-onset as first

reported diagnosis at ages 19 or 21. We included major

depression, dysthymia and depression not otherwise

specified (NOS) in our depression category. Table 1

describes the depression-onset groups in terms of sex

Table 1. Characteristics of the depression-onset groups. The percentages for male and female refer to weighted percentages within the

respective depression (or never-depressed) groups

Child-onset

First diagnosed at

age 9 to <13

Adolescent-onset

First diagnosed

at age 13–16

Adult-onset

First diagnosed at

age 19 or 21

Never-

depressed

Total, n (%) 46 (2.5) 55 (5.6) 44 (3.6) 860 (88.3)

M, n (%) 27 (58.8) 24 (35.1) 20 (40.6) 497 (52.8)

F, n (%) 19 (41.2) 31 (64.9) 24 (59.4) 363 (47.2)

OR (95% CI) M/F 1.35 (0.53–3.47) 0.49 (0.19–1.24) 0.63 (0.22–1.80) 1.55 (0.85–2.85)

Depressive disorders, n (%)

Minor depression 42 (2.3) 50 (5.5) 34 (2.5)

Dysthymia 4 (0.2) 15 (1.6) 17 (1.7)

Major depression 9 (0.4) 18 (2.0) 16 (1.3)

M, Male ; F, female ; OR, odds ratio ; CI, confidence interval.

Values given as unweighted n and weighted prevalence (%).

Risk for depression onset 2267

and specific depression diagnoses. Several subjects

had multiple diagnoses of depression within one

developmental period (e.g. depression NOS in one

childhood year, and major depression in another

childhood year).

Other disorders were also assessed in the CAPA/

YAPA. The unweighted n values and weighted

prevalence were 204 (11.6%) for childhood behavioral

disorders, 99 (6.4%) for childhood anxiety disorders,

203 (18.8%) for adolescent behavioral disorders

(including substance disorders), 48 (4.3%) for ado-

lescent anxiety disorders, and 211 (27.7%) for young

adult antisocial personality disorder and substance

use disorders, and 65 (9.3%) for young adult anxiety

disorders.

Psychosocial risk factors were also collected in the

CAPA and YAPA unless otherwise specified. Here,

we included putative psychosocial risk domains that

have been commonly identified for depression across

development : poverty, stressful life events, parental

psychopathology, maltreatment, and family dysfunc-

tion (Birmaher et al. 1996 ; Cicchetti & Toth, 1998 ;

Goodyer, 2001; Harrington, 2006 ; Zalsman et al.

2006). Individual risk factors (e.g. low income, ma-

terial hardship, and low education in the domain

of poverty) were coded as 1 (present) if reported by

either parent or child (CAPA), and as 0 when not

present. During the adult assessments with the YAPA,

the subject was the sole reporter of all risk factors.

With the exception of lifetime parental psychopath-

ology, all risk factors were assessed at the time of

the interview (e.g. poverty) or over the preceding 3

months (e.g. life events), and were aggregated across

childhood (i.e. any observation from ages 9 to <13), adolescence (i.e. any observation from ages 13 to 16),

and young adulthood (e.g. any observation at ages 19

and 21). For example, if the subject had experienced

material hardship at any assessment between the ages

of 9 to<13, they received a 1 on the childhood version of material hardship. Because the time frame for as-

sessing depression was also the 3 months immediately

preceding the interview, temporal overlap between

childhood putative risk factors and depression onset

in the same developmental period was possible (e.g.

childhood risk and child-onset depression). Indeed,

associations between risks and depression onset

within the same developmental period can only es-

tablish putative risk factor status (Kraemer et al. 2001).

To increase the parsimony of our analyses and our

power to detect differences between the depression-

onset groups, we created a sum score for each risk

domain.

The poverty scale ranged from 0 to 3, summing low

income, material hardship, and low education. Low

income was coded when the household income was

below the federal poverty level. Material hardship was

coded when the family (CAPA) or the subject (YAPA)

were unable to meet basic needs, having no health

insurance, financial problems, residential instability,

or no insurance for mental health or substance abuse

care. Low education was coded when the subject’s

parents (CAPA) or the subject (YAPA) did not

graduate from high school.

The loss and violence events scale ranged from

0 to 2, summing the occurrence of loss and violence

events. Loss events included parental divorce/

separation; death of a loved one, sibling, or peer ;

romantic breakup; breakup with or loss of best friend;

pregnancy loss ; and job loss (YAPA only). Violence

events included death of a loved one by violence, war,

terrorism, witness to a violent life event, and cause of

death or severe harm. Details of the construction and

psychometric testing of the Life Events section of the

CAPA are contained elsewhere (Costello et al. 1998).

Lifetime parental psychopathology ranged from

0 to 3 and summed whether biological parents had

ever sought or received treatment for mental health or

drug problems, and whether the parent had been

arrested and/or prosecuted for a crime since parent’s

age 18. [Arrests for driving under the influence (DUI)

and/or drug related charges were not coded here.]

This risk factor was only assessed using a lifetime time

frame.

Maltreatment ranged from 0 to 2 and summed

sexual abuse/violence (including rape) and physical

abuse/captivity. In the YAPA, spousal abuse was in-

cluded in the physical abuse variable. Finally, family

dysfunction ranged from 0 to 3, and included parent–

child conflict, interparental conflict, scapegoating

(CAPA only), and subject’s marital conflict (YAPA

only). Parent–child conflict was coded when children

scored in the top 25% of parent–child conflict within a

given wave. Interparental conflict was coded when the

relationship between parents was characterized by

high conflict, poor communication and/or violence.

Scapegoating (parental differential treatment) was

coded when children were regarded/treated more

negatively by a parent compared to other children in

the family. Subject’s marital conflict was coded when

subjects reported having conflict with a spouse.

Some individual risk factors were assessed in the

CAPA, but not in the YAPA, because they were no

longer relevant in young adulthood. For example,

scapegoating (i.e. parental differential treatment of

children in the home) was no longer coded in the

young adult assessments because many subjects no

longer resided with parents and siblings. Other risk

factors were only age appropriate for young adults,

including subject’s job loss, and marital violence and

conflict. Table 2 describes the depression-onset groups

2268 L. Shanahan et al.

in terms of (putative) risk factors. When identical risk

domain scores across developmental periods were

created or risk domain scores were standardized

within developmental period, our overall findings did

not change systematically.

Statistical analyses

Weighted logistic regression models were estimated

using generalized estimating equations (GEEs) imple-

mented by SAS PROC GENMOD. Robust (sandwich-type)

variance estimates adjusted the standard errors of the

parameter estimates for the design effects. All analyses

included sampling weights that were inversely pro-

portional to selection probability ; therefore, the results

are representative of the population from which the

sample was drawn. First, each depression-onset group

was examined separately, with child-onset versus never-

depressed, adolescent-onset versus never-depressed,

and young adult-onset versus never-depressed vari-

ables serving as outcome variables. Each (putative)

risk factor sum score was examined individually for

each depression-onset group in univariate regression

models. [The results for individual risk factors (as op-

posed to the sum scores) are available from the first

author upon request.] Next, we also directly tested

differences in the effect sizes of psychosocial risk

factors among the depression-onset groups. For ex-

ample, we tested whether recent risk factors were

more strongly associated with child- than with

adolescent-onset depression. To test for these differ-

ences, we stacked childhood, adolescence and young

adulthood data, and tested interaction terms between

risk factor sum scores and dummy variables indicat-

ing the timing of onset in the prediction of depression.

Because we conducted a large number of statistical

tests, we focus on patterns of results rather than on

single significant coefficients. We emphasize coeff-

icients that are significant using two-tailed significance

testing (i.e. at p<0.05). However, considering that the hypotheses are directional in nature (i.e. higher levels

of risk are associated with depression), coefficients

significant at p<0.10 are discussed when they are consistent with a larger pattern of significant results.

Results

Replicating previous findings for adult-onset

depression

To replicate previous findings regarding adult-onset

depression, we combined the adolescent- and young

adult-onset groups into one group, a strategy used

in previous research (see Fig. 1a). Compared to the

never-depressed, childhood poverty was the only

childhood risk domain predicting adolescent-/adult-

onset depression at p<0.05 [OR 1.65, 95% confidence interval (CI) 1.17–2.30, p<0.01 ; see path ‘b’ in Fig. 1a].

Table 2. Weighted means (standard deviations) of child, adolescent and young adult risk factors by depression-onset group

Psychosocial risk factors

Possible

range

Overall

mean

(n=1004)

Child-

onset

(n=46)

Adolescent-

onset

(n=55)

Adult-

onset

(n=44)

Never-

depressed

(n=859)

Childhood risk

Poverty 0–3 1.00 (1.00) 1.72 (0.63) 1.43 (1.10) 1.51 (1.00) 0.93 (0.98)

Loss and violence events 0–2 0.32 (0.53) 0.81 (0.61) 0.40 (0.55) 0.53 (0.50) 0.29 (0.52)

Lifetime parental psychopathology 0–3 0.99 (0.90) 1.73 (0.73) 1.19 (0.99) 1.00 (0.49) 0.95 (0.91)

Maltreatment 0–2 0.10 (0.30) 0.44 (0.37) 0.11 (0.33) 0.13 (0.31) 0.09 (0.29)

Family dysfunction 0–3 0.87 (0.83) 1.48 (0.66) 0.80 (0.73) 1.01 (0.73) 0.85 (0.84)

Adolescent risk

Poverty 0–3 0.82 (0.89) 0.89 (1.04) 0.97 (0.73) 0.79 (0.90)

Loss and violence events 0–2 0.46 (0.62) 0.55 (0.68) 0.78 (0.68) 0.43 (0.61)

Lifetime parental psychopathology 0–3 1.09 (0.85) 1.38 (1.03) 1.18 (0.51) 1.05 (0.85)

Maltreatment 0–2 0.19 (0.39) 0.51 (0.51) 0.20 (0.37) 0.15 (0.36)

Family dysfunction 0–3 0.80 (0.80) 1.28 (0.85) 1.44 (0.96) 0.72 (0.75)

Young adult risk

Poverty 0–3 1.25 (0.90) 1.55 (0.63) 1.18 (0.91)

Loss and violence events 0–2 0.52 (0.56) 0.80 (0.61) 0.48 (0.57)

Lifetime parental psychopathology 0–3 1.10 (0.81) 1.49 (0.72) 1.05 (0.81)

Maltreatment 0–2 0.01 (0.11) 0.03 (0.15) 0.01 (0.09)

Family dysfunction 0–3 0.24 (0.46) 0.63 (0.60) 0.20 (0.45)

A total of 1004 subjects had data on childhood (putative) risk factors ; 877 subjects had data on adolescent (putative) risk

factors ; 837 had data on young adult putative risk factors.

Risk for depression onset 2269

Thus, overall similarities in childhood psychosocial

risk between the adult-onset depressed and the never-

depressed were confirmed. To examine differences

in childhood psychosocial risk between child- and

adolescent-/adult-onset depression, we also tested

interactions between risk factors and the timing of

onset in the prediction of depression. Several factors

were more predictive of child- than of adolescent-/

adult-onset depression, including parental psycho-

pathology (OR 1.94, 95% CI 1.09–3.46, p<0.01 for the interaction term), maltreatment (OR 8.55, 95% CI

1.51–48.48, p<0.05), and family dysfunction (OR 2.36, 95% CI 1.40–4.00, p<0.05), but not childhood poverty and loss and violence events (OR 1.21, 95% CI

0.79–1.86, p>0.10, and OR 2.22, 95% CI 0.77–6.37, p>0.10, respectively). As in previous research, child- onset depression and adolescent-/adult-onset de-

pression were mostly different in terms of childhood

psychosocial risk, a finding previously interpreted as

consistent with the risk differences hypothesis.

Recency versus potentially genuine risk differences

To disentangle differences in predictors among the

depression-onset groups caused by recency from

potentially genuine risk differences, we first examined

links between concurrent putative risk factors and

the respective depression onsets (paths ‘c ’ in Fig. 1b).

Next we examined links between antecedent risk

factors and depression onsets (i.e. childhood risk for

adolescent-onset and adolescent risk for young adult-

onset depression ; paths ‘d’ in Fig. 1b). The results are

shown in Table 3.

Concurrent putative risk factors

According to the recency hypothesis, concurrently

assessed risk factors (paths ‘c ’ in Fig. 1b, shown in the

shaded cells of Table 3) should be similar in size for

the three depression-onset groups, and should have

the strongest and most consistent links with de-

pression onset. That is, childhood risk factors should

Table 3. Psychosocial risk factors predicting depression onset (compared to the never-depressed)

Risk factor

Child-onset Adolescent-onset Young adult-onset

OR (95% CI) p OR (95% CI) p OR (95% CI) p

Poverty

Childhood 2.08 (1.62–2.69) <0.001 1.61 (1.04–2.49) 0.03a 1.71 (1.06–2.76) 0.03 Adolescence 1.12 (0.68–1.88) 0.64 1.23 (0.84–1.81) 0.28

Young adulthood 1.58 (1.10–2.28) 0.01a

Loss and violence events

Childhood 3.53 (1.49–8.35) 0.004 1.43 (0.72–2.83) 0.31 2.03 (1.03–4.03) 0.04

Adolescence 1.35 (0.67–2.71) 0.40 2.16 (1.02–4.58) 0.04

Young adulthood 2.48 (0.91–6.77) 0.08a

Lifetime parental psychopathology

Childhood 2.31 (1.43–3.73) 0.001 1.32 (0.84–2.06) 0.23 1.06 (0.88–1.27) 0.56

Adolescence 1.55 (0.86–2.78) 0.15 1.20 (0.86–1.68) 0.29

Young adulthood 1.96 (1.07–3.62) 0.03

Maltreatment

Childhood 9.28 (2.63–32.78) 0.001 0.25 (0.03–2.14) 0.20 2.95 (0.64–13.65) 0.17

Adolescence 7.24 (1.71–30.62) 0.007 1.65 (0.39–6.94) 0.50

Young adulthood 3.36 (0.84–13.43) 0.09a

Family dysfunction

Childhood 2.34 (1.54–3.54) <0.001 0.88 (0.58–1.35) 0.55 1.25 (0.75–2.08) 0.39 Adolescence 2.42 (1.39–4.20) 0.002 3.02 (1.35–6.75) 0.007

Young adulthood 3.85 (1.66–8.94) 0.002

OR, Odds ratio (unadjusted) ; CI, confidence interval.

A total of 1004 subjects had data on childhood (putative) risk factors ; 877 subjects had data on adolescent (putative) risk

factors ; 837 had data on young adult putative risk factors.

Values in bold were significant at p<0.05. Values in bold and italics were significant at p<0.10. Shaded values represent associations between concurrent risk factors and depression onset.

a No longer significant at p<0.10 or less when co-morbidity (i.e. concurrent anxiety and behavioral disorders) was taken into account.

2270 L. Shanahan et al.

have the strongest links with child-onset depression,

adolescent risk factors should have the strongest links

with adolescent-onset depression, and young adult

risk factors should have the strongest links with young

adult-onset depression.

The pattern of results suggest that, consistent with

the recency hypothesis, all childhood putative risk

factors were associated with child-onset depression,

and young adult risk factors were associated

with young adult-onset depression. Only adolescent

maltreatment and family dysfunction (but not

adolescent poverty, loss and violence events, and life-

time parental psychopathology) were associated with

adolescent-onset depression.

Because several concurrent putative risk factors

were linked with child- and young adult-onset de-

pression, but not with adolescent-onset depression,

we tested for putative risk differences between

adolescent-onset depression and the other two

depression-onset groups. For example, to examine

whether concurrent poverty was indeedmore strongly

associated with child- than with adolescent-onset

depression, we examined the interaction between

poverty and timing of depression onset in the predic-

tion of depression, essentially testing whether the ORs

for concurrent risk factors reported in Table 3 differed

between child- and adolescent-onset depression.

Concurrent poverty was more strongly linked with

child- than with adolescent-onset depression (OR 1.82,

95% CI 1.02–3.46, p<0.01 for the interaction term). Similarly, concurrent loss and violence events were

more strongly linked with child- than with adolescent-

onset depression at the statistical trend level (OR 2.75,

95% CI 0.88–8.58, p<0.10 for the interaction term). No other differences in concurrent risk between child-

and adolescent-onset and adolescent- and young

adult-onset depression were significant. Summarizing

results regarding concurrent putative risk factors

(paths ‘c ’ in Fig. 1b), the child- and young-adult onset

depression groups were similar in terms of concurrent

psychosocial risk. Indeed, follow-up analyses did not

identify significant differences in concurrent risk for

child- versus adult-onset depression. Adolescent-onset

depression, however, seemed to have some differences

in risk from these groups.

Antecedent risk factors

According to the recency hypothesis, some modest

associations would be expected between risk factors

from a previous developmental period and depression

onset. That is, some childhood risk factors may mod-

estly predict adolescent-onset depression, and some

adolescent risk factors may modestly predict young

adult-onset depression (paths ‘d’ in Fig. 1b). The

results showed that childhood poverty predicted

adolescent-onset depression, and that adolescent loss

and violence events and family dysfunction predicted

young adult-onset depression (see Table 3). Analyses

examining potential differences in risk (i.e. differences

in ORs) in antecedent risk factors between adolescent-

and young adult-onset depression showed that

antecedent family dysfunction was more predictive of

young adult-onset than of adolescent-onset depression

(OR 3.12, 95% CI 1.42–7.29, p<0.05). Summarizing the results regarding antecedent risk factors, adolescent-

onset depression and young adult-onset depression

were mostly similar in terms of antecedent psycho-

social risk.

Childhood risk factors and young adult-onset depression

Finally, the recency hypothesis would predict weak

links between childhood risk factors and young adult-

onset depression. In fact, most childhood risk factors

did not predict young adult-onset depression, with the

exceptions of childhood poverty and childhood loss

and violence events (see Table 3).

Follow-up analyses

In multivariate models we included corresponding

risk factors from childhood and adolescence to

predict adolescent-onset depression, and from child-

hood, adolescence and young adulthood to predict

young adult-onset depression. The results show that

when concurrent risk factors were included, the pre-

viously significant corresponding risk factors from

previous developmental periods continued to predict

adolescent- and young adult-onset depression with

similar effect sizes. Thus, the effects of earlier risk

factors were not mediated by identical later risk. In

another set of multivariate analyses we controlled for

concurrent co-morbidity. For example, for adolescent-

onset depression we controlled for adolescent anxiety

and behavioral disorders. Most associations remained

significant (see coefficients marked with superscript

‘a ’ in Table 3 for exceptions).

Discussion

This is the first epidemiological study to focus

specifically on associations of psychosocial adversity

with child-, adolescent- and young adult-onset de-

pression in order to disentangle differences due to

recency from potentially genuine risk differences.

We also used age-of-onset cut-offs for child- and

adolescent-onset depression that correspond with the

points at which changes in the prevalence of major

depression occur (e.g. Angold et al. 2002).

Risk for depression onset 2271

Consistent with previous research, most childhood

psychosocial risk factors were more predictive of

child-onset than of adolescent-/adult-onset depres-

sion. When we attempted to disentangle potentially

genuine differences in risk from differences due to the

recency with which risk factors had been experienced,

our pattern of results was mostly consistent with the

recency hypothesis, particularly for child- and young

adult-onset depression. All childhood putative risk

factors were associated with child-onset depression;

and corresponding young adult putative risk factors

were associated with young adult-onset depression.

Only two of five adolescent putative risk factors were

linked with adolescent-onset depression. Overall,

our findings show that differences in childhood risk

reported in previous studies mostly reflected differ-

ences in the recency with which the risk factors

had been experienced rather than genuine risk differ-

ences.

A few noteworthy inconsistencies with the recency

hypothesis emerged. First, childhood poverty had

long-lasting effects, and did not differentiate child-

from later-onset depression. This finding was not en-

tirely surprising. In the work of Jaffee et al. (2002),

childhood socio-economic status did not differentiate

between child- and adult-onset depression. Gilman

et al. (2003) also found that childhood low socio-

economic status did not differentiate among child-,

adolescent- and adult-onset depression. Our follow-

up analyses that controlled for later corresponding

risk factors showed that the pathway from childhood

poverty to later depression onset was not explained

by poverty in adolescence or in young adulthood.

Childhood may be a sensitive period during which the

experience of poverty creates lasting changes in the

organism’s stress response (Power et al. 1999 ; Danese

et al. 2009 ; Miller et al. 2009), and, thus, vulnerability

to depression. Second, childhood loss and violence

events predicted young adult-onset depression.

Although parental loss predicted juvenile- but not

adult-onset depression in a previous paper (Jaffee et al.

2002), others have described the long-lasting mental

health effects of childhood loss events (Brown &

Harris, 1978).

Third, all differences among the depression-onset

groups involved adolescent-onset depression, sug-

gesting that there could be some genuine differences

in risk between adolescent-onset depression and

the other onset groups. Alternative pathways to

adolescent-onset depression, particularly for females,

have been suggested, including low birthweight

(Costello et al. 2007), early pubertal timing (Copeland

et al. 2010), increases in pubertal hormones (Angold

et al. 2003), and biopsychosocial and cognitive inter-

actions (e.g. Susman, 1997 ; Ge et al. 2001).

Limitations and directions for future research

First, although the study’s focus was limited to

psychosocial risk factors, the findings have important

implications for gene–environment (GrE) interaction research. For example, taking into account that devel-

opmental nuances of environmental risk such as

their timing in relation to depression onset may be

important for increasing rates of replications in GrE research involving the serotonin-transporter-linked

polymorphic region 5-HTTLPR (Canli & Lesch, 2007).

Second, our assessments began at age 9, but we will

have missed cases with depression onset before age 9,

depression onset in the 9 months of the year that

the CAPA/YAPA interviews did not cover, and de-

pression onset during years when interviews were not

conducted. Third, the depression-onset groups were

relatively small, limiting our statistical power. We also

did not distinguish between juvenile-onset groups

with recurrence versus those without recurrence ;

however, previous work had found few early ad-

versity differences between such groups (Jaffee et al.

2002). Fourth, our last available age for this study was

21, so the findings may be specific to the narrow young

adult age range assessed here.

Fifth, several of our risk factors were assessed con-

currently with depression, and therefore could be

indicative only of ‘putative’ risk similarities and dif-

ferences among depression-onset groups. We also

did not assess risk factors antecedent to child-onset

depression. Sixth, our findings are not informative

with respect to causal chains leading to the onset of

depression. Risk factors can also be heterogeneous

in terms of their developmental history, and future

research should examine interactions between risk

factors at different developmental periods in the pre-

diction of depression onset. Finally, to capture each

risk domain in the most age-appropriate, devel-

opmentally valid way, some individual risk factors

included in each risk domain varied somewhat be-

tween childhood/adolescence and young adulthood.

These slight changes in the composition of risk do-

mains could allow for an alternative interpretation

of findings: that apparent similarities in associations

between child/adolescent and young adult risk

factors nevertheless disguise potential risk differences.

Additional analyses showed, however, that when risk

factors were forced to be identical across develop-

mental periods or when risk factors were standardized

within each developmental period, the overall find-

ings did not change.

These limitations were balanced by the prospective

longitudinal design of our study, and the reliability of

CAPA and YAPA symptom assessment. Furthermore,

they were not unique to our study. Indeed, the only

2272 L. Shanahan et al.

other prospective longitudinal study of depression-

onset groups assessed depression at only six waves

per subject, starting at age 10, and interviewed

participants every 2, 3 or 5 years, using 12-month time

frames for symptom assessments (Jaffee et al. 2002).

Future prospective longitudinal studies should aim

for continuous coverage of depression-onset data.

This would determine whether findings are specific to

depression onset at particular ages, and not just to any

diagnosis of depression at these ages.

Despite these limitations, our study shows that,

when potentially genuine risk differences were disen-

tangled from differences in the recency of risk, the

number of putative psychosocial risk differences

among developmentally defined depression-onset

groups is relatively small. Although distinguishing

among developmental subtypes has been useful for

other disorders (Moffitt et al. 2008), our findings sug-

gest that assuming distinctions between child- and

young adult-onset depression based on differences in

psychosocial risk factors is unwarranted. Differences

between adolescent-onset depression and the two

other depression-onset groups may be consistent with

studies showing that adolescent-onset depression is

predicted by biological factors.

Acknowledgments

The work presented here was supported by the

National Institute of Mental Health (MH63970,

MH63671, MH48085), the National Institute on Drug

Abuse (DA/MH11301), and the William T. Grant

Foundation. All authors had full access to all in this

study and Dr Shanahan takes responsibility for the

integrity of the data, and the accuracy of the data

analysis.

Declaration of Interest

None.

References

Achenbach TM, Edelbrock C (1983). Manual for the Child

Behavior Checklist and Revised Child Behavior Profile. Queen

City Printers : Burlington, VT.

Angold A, Costello EJ (1995). A test-retest reliability study

of child-reported psychiatric symptoms and diagnoses

using the Child and Adolescent Psychiatric Assessment

(CAPA-C). Psychological Medicine 25, 755–762.

Angold A, Costello EJ (2000). The Child and Adolescent

Psychiatric Assessment (CAPA). Journal of the American

Academy of Child and Adolescent Psychiatry 39, 39–48.

Angold A, Cox A, Prendergast M, Rutter M, Simonoff E,

Costello EJ, Asher BH (1999). The Young Adult Psychiatric

Assessment (YAPA). Duke University Medical Center :

Durham, NC.

Angold A, Erkanli A, Silberg J, Eaves L, Costello EJ (2002).

Depression scale scores in 8–17-year-olds : effects of age

and gender. Journal of Child Psychology and Psychiatry 43,

1052–1063.

Angold A, Worthman C, Costello EJ (2003). Puberty and

depression. In Gender Differences at Puberty (ed. C.

Hayward), pp. 137–164. Cambridge University Press :

New York, NY.

Birmaher B, Ryan ND, Williamson DE, Brent DA (1996).

Childhood and adolescent depression : a review of the past

10 years, Part I. Journal of the American Academy of Child and

Adolescent Psychiatry 35, 1427–1439.

Brown, GW, Harris TO (1978). The Social Origins of

Depression : A Study of Psychiatric Disorder in Women.

Free Press : New York, NY.

Canli T, Lesch KP (2007). Long story short : the serotonin

transporter in emotion regulation and social cognition.

Nature Neuroscience 10, 1103–1109.

Christie KA, Burke JD, Regier DA, Rae DS (1988).

Epidemiologic evidence for early onset of mental disorders

and higher risk of drug abuse in young adults. American

Journal of Psychiatry 145, 971–975.

Cicchetti D, Toth SL (1998). The development of depression

in children and adolescents. American Psychologist 53,

221–241.

Copeland WE, Shanahan L, Costello EJ, Angold A (2009).

Childhood and adolescent psychiatric disorders as

predictors of young adult disorders. Archives of General

Psychiatry 66, 764–772.

Copeland WE, Shanahan L, Miller S, Costello EJ,

Angold A, Maughan B (2010). Outcomes of early

pubertal timing in young women : a prospective

population-based study. American Journal of Psychiatry

167, 1218–1225.

Costello EJ, Angold A, Burns BJ, Stangl DK, Tweed DL,

Erkanli A, Worthman CM (1996). The Great Smoky

Mountains Study of youth : goals, design, methods, and the

prevalence of DSM-III-R disorders. Archives of General

Psychiatry 53, 1129–1136.

Costello EJ, Angold A, March J, Fairbank J (1998). Life

events and post-traumatic stress : the development of a

new measure for children and adolescents. Psychological

Medicine 28, 1275–1288.

Costello EJ, Mustillo S, Erkanli A, Keeler G, Angold A

(2003). Prevalence and development of psychiatric

disorders in childhood and adolescence. Archives of

General Psychiatry 60, 837–844.

Costello EJ, Worthman C, Erkanli A, Angold A (2007).

Prediction from low birth weight to female adolescent

depression : a test of competing hypotheses. Archives of

General Psychiatry 64, 343–350.

Danese A, Moffitt TE, Harrington H, Milne BJ,

Polanczyk G, Pariante CM, Poulton R, Caspi A (2009).

Adverse childhood experiences and adult risk factors for

age-related disease. Archives of Pediatric and Adolescent

Medicine 163, 1135–1143.

Ge X, Conger RD, Elder Jr. GH (2001). Pubertal transition,

stressful life events, and the emergence of gender

Risk for depression onset 2273

differences in adolescent depressive symptoms.

Developmental Psychology 37, 404–417.

Giaconia RM, Reinherz HZ, Silverman AB, Pakiz B (1994).

Ages of onset of psychiatric disorders in a community

population of older adolescents. Journal of the American

Academy of Child and Adolescent Psychiatry 33, 706–717.

Gilman SE, Kawachi I, Fitzmaurice GM, Buka SL (2003).

Socio-economic status, family disruption and residential

stability in childhood : relation to onset, recurrence and

remission of major depression. Psychological Medicine 33,

1341–1355.

Goodyer IM (2001). Life events : their nature and effects.

In The Depressed Child and Adolescent, 2nd edn (ed. I. M.

Goodyer), pp. 204–232. Cambridge University Press :

New York, NY.

Harrington R (2006). Affective disorders. In A Clinician’s

Handbook of Child and Adolescent Psychiatry (ed. C. Gillberg,

R. Harrington and H. Steinhausen), pp. 110–143.

Cambridge University Press : New York, NY.

Hill J, Pickles A, Rollinson L, Davies R, Byatt M (2004).

Juvenile- versus adult-onset depression : multiple

differences imply different pathways. Psychological

Medicine 34, 1483–1493.

Jaffee SR, Moffitt TE, Caspi A, Fombonne E, Poulton R,

Martin J (2002). Differences in early childhood risk factors

for juvenile-onset and adult-onset depression. Archives of

General Psychiatry 59, 215–222.

Kasch KL, Klein DN (1996). The relationship between age at

onset and comorbidity in psychiatric disorders. Journal of

Nervous and Mental Disease 184, 703–707.

Kaufman J, Martin A, King RA, Charney D (2001). Are

child-, adolescent-, and adult-onset depression one and the

same disorder? Biological Psychiatry 49, 980–1001.

Kessler RC, Davis CG, Kendler KS (1997). Childhood

adversity and adult psychiatric disorder in the US

National Comorbidity Survey. Psychological Medicine 27,

1101–1119.

Kovacs M, Feinberg TL, Crouse-Novac M, Paulauskas SL,

Pollok M, Finkelstein R (1984). Depressive disorders in

childhood. II. A longitudinal study of the risk for a

subsequent major depression. Archives of General Psychiatry

41, 643–649.

Kraemer HC, Stice E, Kazdin A, Offord D, Kupfer D (2001).

How do risk factors work together? Mediators,

moderators, and independent, overlapping, and proxy risk

factors. American Journal of Psychiatry 158, 848–856.

Miller GE, Chen E, Fok AK, Walker H, Lim A, Nicholls EF,

Cole S, Kobor MS (2009). Low early-life social class leaves

a biological residue manifested by decreased

glucocorticoid and increased proinflammatory signaling.

Proceedings of the National Academy of Sciences USA 106,

14716–14721.

Moffitt TE (2006). Life-course-persistent versus adolescence-

limited antisocial behavior. In Developmental

Psychopathology, vol. 3 : Risk, Disorder, and Adaptation,

2nd edn (ed. D. Cicchetti and D. J. Cohen), pp. 570–598.

John Wiley & Sons Inc. : Hoboken, NJ.

Moffitt TE, Arseneault L, Jaffee SR, Kim-Cohen J,

Koenen KC, Odgers CL, Slutske WS, Viding E (2008).

Research review: DSM-V conduct disorder : research

needs for an evidence base. Journal of Child Psychology and

Psychiatry 49, 3–33.

Power C, Manor O, Matthews S (1999). The duration and

timing of exposure : effects of socioeconomic environment

on adult health. American Journal of Public Health 89,

1059–1065.

Rao U, Ryan ND, Birmaher B, Dahl RE (1995). Unipolar

depression in adolescents : clinical outcome in adulthood.

Journal of the American Academy of Child and Adolescent

Psychiatry 34, 566–578.

SAS Institute (2004). SAS/STAT1Software Version 9. SAS

Institute Inc. : Cary, NC.

Schaie KW (1965). A general model for the study

of developmental problems. Psychological Bulletin 64,

92–107.

Susman EJ (1997). Modeling developmental complexity in

adolescence : hormones and behavior in context. Journal of

Research on Adolescence 7, 283–306.

Weissman MM, Wolk S, Wickramaratne P, Goldstein RB,

Adams P, Greenwald S, Ryan ND, Dahl RE, Steinberg D

(1999). Children with prepubertal-onset major depressive

disorder and anxiety grown up. Archives of General

Psychiatry 56, 794–801.

Zalsman G, Brent DA, Weersing VR (2006). Depressive

disorders in childhood and adolescence : an overview:

epidemiology, clinical manifestation and risk factors.

Child and Adolescent Psychiatric Clinics of North America 15,

827–841.

2274 L. Shanahan et al.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.