Abstract
Depression, the most common psychiatric condition, is usually chronic or recurrent [1] and the associated disease burden causes considerable disruption to sufferers and to those around them [2]. The effects of parental depression on children from infancy through adolescence have been extensively reported [3–5], though the long-term effects of parental depression on adult children are more contentious. The great diversity in outcomes of children and the relative resilience of many adult children of depressed patients makes it difficult to understand the relationship between risk and the emergence of psychopathology over time. In a 25-year follow-up of adult children of depressed patients, we found their risk for specific disorders such as anxiety and substance disorder were increased compared to controls, but their rates of overall psychiatric morbidity and quality of intimate relationships were similar to controls [6].
Over the last two decades progress has been made in identifying biological, psychological and socialinterpersonal factors which potentially mediate the adverse effects of parental depression on their children. These factors (not all confirmed) can be grouped as pertaining to the parent's illness, the family and the child (Table 1).
Vulnerability factors to psychopathology in children of patients with depression
Parental illness factors and their relative importance are not well understood and investigations generally conceptualize depression as a homogeneous, unidimensional construct with parents often simply dichotomized into depressed and well groups [5]. Yet, a handful of studies have suggested that parental depression compounded by personality disorder [7] or other psychiatric illness such as anxiety [8] may bemore noxious to children than uncomplicated depression. Intuitively, chronic disorders would appear more likely to impact on child development, and ultimately on child outcome, than single episodes. However, chronicity is confounded by severity and no consensus has been reached as to the relationship between chronicity or severity of parental depression and child outcome. Keller et al. [9] found an association between impaired adaptation and the presence of DSM-III disorder in the children of depressed patients and almost every measure of chronicity and severity of depression (duration of episodes, number of criteria met for major depression with or without psychosis, number of suicide attempts and number of episodes) in the biological parents. However, Weissman et al. [10],[11] reported that depression severity defined by hospitalization, suicidal ideation/attempts or recurrence was not a risk factor for familial transmission of depression. Sameroff et al.[12] found that chronicity of depression in mothers contributed negatively to the developmental course of their children and Campbell et al. [13] reported that infants of mothers with chronic depression from 2 to 6 months post-partum were less positive in face-toface interaction than those of mothers with episodic depression.
Family factors such as parental discord, lack of cohesion and divorce are important risk factors for psychopathology in children of depressed parents [3],[5],[7]. Fathers may also have a role in mediating the effects of maternal depression [4],[7],[14]. Although secure attachments are often formed between depressed parents and children, attachment difficulties, most commonly of the insecure attachment of the resistant/avoidant kind [3],[5],[15], may mediate the development of maladaptive functioning in children of depressives.
Factors pertaining to the child may mediate outcome too, especially the age at which exposure to parental depression occurs. This is complicated further by recurrence and chronicity [15], yet the vast majority of studies concentrate on cross-sectional designs and the effect of parental depressive mood (mainly maternal) at one point in time. Although many studies suggest that even very young children may be vulnerable to the effects of parental depression, other studies suggest that children become more vulnerable as they get older [5]. Gender-linked differences in adjustment of children to depressedmothers are complex. Boys and girls of depressed parents may be differentially vulnerable to maternal depression at specific developmental stages and develop different forms of psychopathology [1],[3],[16]. Temperament may exert its effect directly on susceptibility to parental depression due to the vulnerability of children with difficult temperaments to negative familial characteristics, and indirectly, due to the effects of children's temperament on family discord, parenting impairments and disturbances in the parent–child relationship [5].
Despite hypothesized associations between risk factors (e.g. parent illness factors, child gender and patterns of family relationships), few have explored the cumulative and interactive effect of several of these risk factors [4]. Hammen et al. [17] identified a list of eight potential risk/resilience factors (e.g. child self-concept and maternal current mood) that were significantly associated with adverse outcome in the children of affectively ill parents. Beardslee et al. [18] showed that onset of disorder in adolescents over a 4-year period was predicted by duration of parental affective disorder, number of parental nonaffective diagnoses and total number of prior diagnoses. Although the effects of comorbidity and chronicity of parental disorder appear to be more important than severity, the effects of parental depressive severity, comorbidity and chronicity on children have rarely been examined simultaneously.
The majority of the preceding work has focused on the offspring of depressives when still children and adolescents. The relative importance of such risk factors in determining outcome in adult offspring of patients with depression is poorly understood and has only been explored in a piecemeal fashion [19]. For example, studies have examined the associations between quality of parenting and vulnerability to depression [20], effects of parental cognitions and social adjustment [19] and patterns of familial or genetic transmission of depression [11],[21–23].
Against this background, we aimed to examine factors associated with lifetime psychiatric disorders in adult children of a cohort of patients with depression admitted to hospital. We hypothesized that:
• parent illness variables such as depressive type, chronicity, severity, global outcome of the parent's depression and abnormal personality traits in the depressed parent; • patterns of family relationships represented by the marital relationship of the parents and the early parent–child relationship; and • child variables such as age at time of exposure to parental depression and gender; and
Method
The sample: patients and children
Recruitment and follow-up of the original patient cohort (n = 212) [37],[38], followed up 25 years after their admission with symptoms of depression to the psychiatry ward of a general teaching hospital in Sydney, Australia, has been described [39]. At index assessment, patients were diagnosed as primary depression (i.e. with endogenous/psychotic depression or neurotic depression) or compound depression. This latter group comprised patients with depressive symptoms and hysterical neurosis, anxiety neurosis, phobic neurosis, transient situational disturbance, pan-neurosis or neurosis not otherwise specified (NOS), hypochondriacal neurosis, obsessional compulsive neurosis and alcohol addiction. Over 25 years, in addition to other axis I or II disorders, this group experienced significant depressive symptomatology with severe depressive episodes requiring admission (38.9% hospitalized), treatment with antidepressants (61.1%) or ECT (23.5%) subsequent to their index episode [37]. The primary and compound depressive groups were combined for the purpose of this study and the effect of depressive type was examined as a predictor of child outcome. The clinician who assessed child psychopathology was blind to depressive type.
After attrition due to death (n = 70), refusal (n = 37) or inability to be located (n = 34), 25-year outcomes were available for a total of 71 of the original 212 patients [39]. No differences were found at baseline between index cases followed up or not on demographic and depression severity indices. The 71 patients remaining in the study generated 139 living children, of whom lifetime psychiatric history, personality and relationship function (previously described) could be obtained from 94 (68%), representing children of 45 patients [6]. There were no significant demographic or outcome differences between the patients whose children participated in this study and those whose children did not.
The independent variables and instruments
The independent variables considered in the prediction of child psychopathology related to the parent's depression, aspects of family relationships and characteristics of the child. The 25-year follow-up provided extensive data relating to the severity and chronicity of the parent's depression [39]. A logistic regression was used to examine predictor variables; however, the number of predictor variables entered into the logistic regression was limited by the overall sample size due to the importance of maximizing the ratio of cases to variables [40].
The parent variables considered in the prediction of child psychopathology were parental gender; parental neuroticism (N) score at baseline using the Eysenck Personality Inventory (EPI) which has wellestablished reliability and validity [41]; primary versus compound depressive type; global clinical outcome for the entire 25-year follow-up period; number of hospitalizations for depression; number of episodes of depression and per cent of entire follow-up spent depressed (data were recorded as percentage categories due to inherent problems with recall precision).
Family relationship variables were the child's current perception of the early relationship with the parent measured by the Parental Bonding Instrument (PBI – a reliable and valid, self-report measure of perceived parental characteristics which generates scores of two underlying dimensions: ‘care’ and ‘overprotection’) [42]. The current parental marital relationship was measured by the patient's rating of his or her partner using the ‘care’ score on the Intimate Bonds Measure (IBM), a reliable and valid self-report measure of intimate relationships which generates scores of two underlying dimensions: ‘care’ and ‘control’ [43].
Child variables comprised the child's gender, the child's age when the parent was admitted to hospital for depression and the child's current occupational status on the Daniel Prestige Scale, an ordinal rating scale of occupations (higher scores denoting higher occupational level) devised for Australia [44].
Dependent variable
The outcome or dependent variable was the presence or absence of any clinician-diagnosed, lifetime DSM-III-R psychiatric disorder in the child at the 25-year follow-up. Current and lifetime psychiatric diagnoses were made according to two methods: (i) a clinician diagnosis by an experienced psychiatrist based on the best available information derived from semistructured interviews, family psychiatric history obtained from the parent (and spouse if available) and/or medical records; and (ii) by the psychiatrist administering the Composite International Diagnostic Interview schedule [45] which generates computer algorithm ICD-10 and DSM-III-R diagnoses. The full assessment protocol for generating these diagnoses has been outlined previously [6]. Interrater reliability and diagnostic validity of the clinical diagnoses was assessed by having a second rater (HB) make independent psychiatric diagnoses on a subsample (n = 13); the kappa value for determining the presence or absence of any lifetime psychiatric disorder was 1.00.
Statistics
Logistic regression [40] with a forward stepwise method of entry was used to examine the factors associated with the presence of psychiatric disorder in the adult children of the depressed parent cohort. Associations between candidate predictor variables were investigated before the regression analyses being run. Spearman's correlation coefficient (rs) was used to assess correlations between dimensional variables (e.g. PBI score and child's age) and t-tests to assess associations between categorical and dimensional variables (e.g. parent gender and IBM score). Separate regression analyses were conducted for significantly correlated variables. The differing patterns of missing data necessitated combinations of variables for the regression analyses which maximized the number of subjects included in the analysis, while avoiding combinations which included intercorrelated variables.
Ethics
The study had approval from the South Eastern Sydney Area Health Service Human Research Ethics Committee. Children and parents gave informed consent; children were only contacted once their parents had given consent and patient anonymity has been preserved.
Results
Demographics
The mean age (SD) of the 41 male and 53 female children was 36.1 (13.0) with a range from 10 to 72 years. Their mean age (SD) at the time of the parent's index admission was 15.3 (10.0) with a range from 1 to 37 years. The children's mean (SD) occupational status was 4.8 (1.2) (higherDaniel scores, maximum = 7, indicate more prestigious occupations) and years of education (SD) was 11.7 (2.5).
Factors associated with psychological morbidity
The variables relating to parent depression were all intercorrelated at a significance level of <0.01, except for (a) per cent of follow-up spent depressed and global clinical outcome (p = 0.618); and (b) per cent of follow-up spent depressed and number of hospitalizations (p = 0.069) (Table 2). Of interest were the correlations that suggested that the depressed parent was rated as more overprotective or controlling when children were younger at parent admission (rs = −0.58, p = 0.000) and when the depressed parentwas themother (t = 3.13, df = 48, p = 0.003).
Significance levels for associations between variables considered for logistic regression
The more hospitalizations for (rs = −0.36, p = 0.006) and the more episodes of depression (rs = −0.35, p = 0.008) the parent experienced, the lower the rating on the depressed parent's IBM score (i.e. their perception of their spouse as caring) suggesting that the episodic nature of depression had an adverse effect on quality of intimate relationships or vice versa. There was also an association between female gender of the depressed parent and lower parent IBM care score (t = −3.27, df = 56, p = 0.002).
Five logistic regression models were required to examine which variables predicted child psychiatric morbidity. The combination of variables in each model was selected to avoid correlated variables and to maximize available subject numbers in each analysis. In the first model, the following variables were entered into regression analyses with dichotomous lifetime child clinical diagnosis as the dependent variable:
Using this combination of variables, 63/94 cases (due to missing data) were included in the regression analysis. Per cent of follow-up time depressed was the only variable successfully entered into this logistic regression model (χ2 = 15.44, df = 4, p = 0.004). Specifically, greater time depressed predicted presence of clinical diagnoses in the children. For example, 44.5% of the parents of children who received a clinical diagnosis compared with only 7.4% of the parents of children who did not receive clinical diagnosis spent more than a quarter of the follow-up period depressed. This model correctly predicted the classification (i.e. child with or without a clinical diagnosis) of 66.7% of the sample.
In the second model, the following variables were entered:
Using this combination, 63 cases were included in the analysis and the result was the same, that is, per cent of follow-up time depressed was the only variable successfully entered into the model. This model correctly predicted the classification (i.e. child with or without a clinical diagnosis) of 66.7% of the sample.
In the third model, the following variables were entered:
Using this combination, 63 cases were included and child's age at parent admission was the only variable successfully entered into this logistic regression model (χ2 = 7.80, df = 1, p = 0.005). Specifically, a younger age at the time of parent admission for depression (i.e. 12.51, SD = 8.78 compared with 17.69, SD = 10.28 for children without clinical diagnoses) predicted the presence of clinical diagnoses in the children. This model correctly predicted the classification (i.e. child with or without a clinical diagnosis) of 66.7% of the sample.
In the fourth model PBI care and overprotection scores were entered which resulted in 50 cases being included in the regression analysis. Parental Bonding Instrument overprotection was the only variable successfully entered into this logistic regression model (χ2 = 4.47, df = 1, p = 0.035). Specifically, higher PBI overprotection scores predicted the presence of clinical diagnoses in the children, that is, the mean PBI overprotection score for the children with clinical diagnoses was 12.58 (SD = 5.67) compared with 9.42 (SD = 4.62) for children without clinical diagnoses. This model correctly predicted the classification (i.e. child with or without a clinical diagnosis) of 66.0% of the sample.
In the fifth model depressive type (i.e. primary vs compound) was entered as a separate variable in isolation due to its significant association with predictor variables from the previous four models (e.g. global outcome, per cent time depressed, PBI – parent care). In this model depressive type did not significantly predict the classification. Small numbers with complete data precluded the entry of depressed parent personality variables (i.e. index EPI-N score) as a predictor in the model.
In summary, the variables which successfully predicted the presence of clinical diagnosis in the children were: (i) younger age of children at the time of parent index admission; (ii) the chronicity of the parent's depression (defined by per cent of follow-up time spent depressed); and (iii) the children's perception of the depressed parent asmore controlling (as defined by higher PBI overprotection scores).
Of added interest is that female gender of the depressed parent was associated with the child's perception of the depressed parent as more controlling and the depressed parent's perception of lower care from the spouse (defined by lower parent IBM care scores). Recurrence of the parent's depression, defined by the number of hospitalizations for and episodes of depression, was inversely related to the depressed parent's perception of care from the spouse.
Discussion
We hypothesized that a wide range of child, parent and family variables would predict psychological morbidity in adult children of a cohort of depressed patients. Of parent illness variables such as global clinical outcome over 25 years, number of hospitalizations for and episodes of depression and per cent of entire follow-up spent depressed, only the last predicted the presence of clinical diagnosis in the children. This suggests that duration of exposure to parental depression is more noxious to children than clinical severity and recurrence of parental depression as defined by number of hospitalizations or number of episodes [9–11]. Keller et al. [9] failed to find an association between number of hospitalizations for depression and DSM-III disorder in children, but showed an association between child psychopathology and both total duration and number of episodes of depression in both parents. Hammen et al. [17] reported an association between poor child-outcome and total duration, number of episodes and severity of parental depression. Our interpretation of these findings is that the number of episodes of depression or number of hospitalizations for patients with depression is inadequate to explain the long-term effects on the patients' children. It may be that chronic subthreshold symptoms of depression explain the persistent negative effects on parenting [46],[47].
Age at parent admission predicted subsequent child morbidity, but the child's gender did not. That younger age at parent admission predicted child psychopathology makes intuitive sense from a developmental perspective and suggests that the impact of parental psychiatric illness differs at different developmental stages. Older children may be less vulnerable to parental depression because they are more likely to have mastered maturational tasks and developed competencies for successful coping [4].
Studies of younger children that suggest that boys may be more vulnerable to the effects of parental depression [16],[48–50], do not appear to translate to greater risk of psychopathology in the long-term as adults. Similarly, Murray et al. [51] found that associations between child outcome and maternal postnatal depression were independent of the child's gender.
As hypothesized, several of the variables considered to predict child psychological morbidity were intercorrelated. The correlation between depressive type and several of the predictor variables in the logistic regression suggests that, at least in terms of determining the presence of lifetime disorder in the offspring, the effect of depressive type is mediated by chronicity. Other correlations showed the systemic interactions between parent illness, child psychopathology and family relationships. For example, the depressed parentwas rated as more overprotective or controlling when children were younger at parent admission and when the depressed parent was the mother. Further, the patient's spouse was more likely to be perceived as less caring when the depressed patient was female and when the patients' depression was recurrent. In the light of our previous finding of problematic relationships between children of depressed mothers and their fathers (often perceived by their children as uncaring) [6], we propose that there is a vulnerable triad of younger child, depressed, overprotective and overcontrolling mother and distanced father. These findings further emphasize the need for a systemic family focus in the treatment of depression.
This study should be interpreted in the context of several potentially important methodological limitations:
Notwithstanding the inherent methodological problems in a 25-year follow-up, this study has investigated several of the adult child, parent illness and family factors using a broad, systemic approach which potentially mediate the relationship between depression and child outcome. More precise identification of these factors and their interaction can enhance treatment and more importantly, primary intervention strategies. We have shown that the ‘noxious dose’ of parental depression reflecting more sustained depressive illnesses rather than merely hospitalized depression, may be more relevant in determining the longterm effects on children. This study argues for a need to address relationships within families of patients with depression (with focus on spouses, particularly husbands of depressedmothers) and formore detailed investigation of the exact nature of exposure to parental depression when examining the effects of such depression on children.
Footnotes
Acknowledgements
The authors thank the patients and their families who contributed so willingly and helpfully. Debra Rozea and Kylie Fell assisted with interviewing and data collection, Alisa Green with data entry and Claire Thompson with preparation of the manuscript. We also thank Gordon Parker and Kay Wilhelm for their helpful suggestions. Carmelle Peisah was supported by an Eli Lilly Fellowship in the first year and an Australian Rotary Health Research Fund in the fourth year of this project. The longitudinal study of the depressed patients was supported by a Commonwealth of Australia National Health and Medical Research Council grant.