Parental Postpartum Depression and Children’s Socioemotional Development: The Role of Socioeconomic Inequality

Abstract

Objective

Parental postpartum depression (PPD) is a documented risk factor for mental health problems in childhood, but little is known about its interplay with family socioeconomic status (SES). This study tested the interactive effect of SES in the associations of PPD with mental health symptoms in children from infancy to adolescence.

Method

Data used for this study were from the Quebec Longitudinal Study of Child Development. Data included self-reported maternal and paternal depressive symptoms at 5 months postpartum, using the Center for Epidemiologic Studies Depression Scale. Parents, teachers, and children/adolescents reported internalizing/externalizing symptoms in children/adolescents using the Social Behavior Questionnaire (ages 3.5-13 years) and the Mental Health and Social Inadaptation Assessment for Adolescents (ages 15-17 years). Mothers reported SES at baseline. Cross-classified 3-level mixed effects modeling was used to test associations.

Results

After excluding 168 single-parent families, the sample consisted of 1,899 families with usable data. Of these, 314 (16.5%) families reported maternal PPD, and 151 (8.0%) reported paternal PPD. Family SES moderated the association between PPD in both parents and mental health symptoms in children. In low SES families, paternal PDD (β = .27, 95% CI 0.10-0.45, p = .003) and maternal PPD (β = .38, 95% CI 0.25-0.50, p < .001) were associated with greater child mental health problems in children, whereas this was the case only for maternal PPD in average or high SES families (β = 0.20, 95% CI 0.10-0.29, p < .001).

Conclusion

PPD in both parents increases the risk of mental health problems in children, with a greater effect in low SES families.

Plain language summary

This study explored how parental postpartum depression (PPD) and family socioeconomic status (SES) are associated with children's mental health from infancy to adolescence. Using data drawn from the Québec Longitudinal Study of Child Development, the study found that in families with low SES, both maternal and paternal PPD were linked to more significant mental health issues in children. In contrast, in families with average or high SES, only maternal PPD was associated with children's mental health problems. Given the interactions between SES and PPD in pediatric mental health, the authors propose that interventions supporting parents in low SES families might be most helpful in promoting children's mental health and reducing socioeconomic inequalities.

Mental health problems during childhood and adolescence have become a growing global health concern, with worldwide prevalence rates for any disorder ranging from 13% to 22%.^1,2 Both maternal³ and paternal⁴ parental postpartum depression (PPD) are well-documented risk factors for mental health problems in children and adolescents, with studies showing that both genetic and environmental mechanisms play a role in these associations. With 14% of mothers⁵ and 9% of fathers⁶ experiencing PPD, it is crucial to prevent these conditions to improve family well-being and interrupt the intergenerational transmission of mental health problems.

Parental mental health problems may have a greater detrimental effect on children growing up in poverty. Low socioeconomic status (SES) environments, characterized by limited financial resources, food insecurity,⁷ and restricted access to mental health services,⁸ contribute to increased mental health problems for parents and children.^7,9 Internalizing and externalizing problems are more prevalent among children from lower socioeconomic backgrounds,¹⁰ and these health inequalities have grown in the last decades.¹¹ The moderating effect of SES on the intergenerational transmission of mental health problems is plausible, as poverty is associated with various child, familial, and environmental factors known to moderate the association between parental mental health and child outcomes. These factors include child maltreatment,¹² child stress reactivity,¹³ marital conflict,¹⁴ single parenthood,¹⁵ limited social support,¹⁶ weak neighborhood cohesion,¹⁷ the experience of discrimination based on ethnic minority status,¹⁸ and insufficient access to, or poor quality of, childcare services.⁸ Therefore, considering family SES is necessary to provide a better understanding of the phenomenon of intergenerational transmission of mental health problems.

This hypothesized moderating effect of family SES on the association between PPD and mental health outcomes in children could vary for maternal and paternal depression. Prior research suggests that some risk factors are more specific to maternal PPD, such as social and emotional support, whereas other factors are more specific to paternal PPD, such as social class or low gratification at work.¹⁹ Differences between parents have also been observed in the mechanisms explaining the association between parental depression and mental health problems in children.20, 21, 22, 23 A qualitative study noted differences in the experience of PPD for mothers vs fathers.²⁴ Stress and depressive symptoms experienced by mothers were related to internal requirements concerning the child, the family, and the domestic situation, as well as an overwhelming feeling of responsibility. On the other hand, fathers were more concerned about external requirements, such as feelings of inadequacy due to a demanding career, making it challenging to combine work, childcare, and a challenging schedule. Considering the differences between maternal and paternal depression regarding risk factors, manifestations, and the mechanisms involved in the associations with mental health outcomes in children, differences in the factors moderating these associations are also plausible. To verify if the moderating effect of SES varies for maternal vs paternal depression, it is essential to study the interaction between SES and depression in both parents simultaneously.

Studies examining the moderating effect of SES on the association between PPD in both parents simultaneously and mental health symptoms in children are rare and have yielded inconsistent findings.^22,25,26 Two studies using large cohort samples to study the associations between PPD and mental health outcomes in children during preschool years found no moderating effect of parental education²⁶ or socioeconomic adversity.²² However, another large cohort study observed significant maternal and paternal PPD associations with offspring depression at age 18 years only for parents with low education.²⁵ Furthermore, these 3 studies used varying measures of SES; 2 studies used parental education separately, and the other used a measure of economic adversity at the family level based on 5 sources of risk, excluding parental education. More research is necessary to clarify the inconsistencies between these studies.

The present study addresses a frequent source of bias in research on the associations between parental depression and mental health in children: the same-informant bias. Most longitudinal studies rely solely^20,22,26 or mainly²⁷ on mother reports of mental health symptoms in children. A recent meta-analysis highlighted that correlations between parental depression and mental health outcomes in children were larger when the same informants rated both parental depression and children’s problems, even in longitudinal analyses.²⁸ Using both mother and father reports is recommended for assessing different perspectives, especially when parental psychopathology is present.²⁹ Multilevel longitudinal modeling effectively analyzes multi-informant data³⁰ over time, even with missing information.

In the present study, we examined the associations between clinically elevated levels of depressive symptoms in each parent—hereafter referred to as maternal or paternal PPD—and mental health symptoms in their children from 1.5 to 17 years of age, using multi-informant data from a population-based birth cohort in Quebec, Canada. Using multilevel longitudinal modeling, we tested whether the association between PPD in each parent and mental health symptoms in children varied based on family SES. Considering that the prevalence of internalizing problems is higher for girls and the prevalence of externalizing problems is higher for boys, we also explored whether these associations varied by the child’s sex and mental health outcomes.

Method

Participants

The Quebec Longitudinal Study of Child Development (QLSCD)³¹ is a representative birth cohort study conducted by the Institut de la Statistique du Québec (ISQ), covering all of Quebec except for the Northern Territories and Cree, Inuit, and First Nations Reserves (2.2% of births). Singletons born at 24 to 42 weeks’ gestation were selected from the 1997-1998 Quebec birth registry and stratified by geographical area. Children with mothers who spoke neither French nor English or who were already participating in a longitudinal study were excluded (n = 2,120). For the present study, we further excluded children from single-parent families because the objective was to test specific associations for depression in each parent with outcomes in children, while controlling for the other parent’s depression in the same model.

Study Design

Data in the QLSCD include parental depressive symptoms and covariates at 5 months after childbirth. Follow-up assessments of children occurred annually at ages 1.5 to 8 years and every 2 years thereafter to age 17 years. Data were collected by telephone interviews with mothers as well as self-reported questionnaires from mothers, fathers (or the biological mother’s live-in partner [0.5% of cases]), teachers, and children. The QLSCD protocol was approved by the ISQ and the Sainte-Justine Hospital Research Centre ethics committees. Parents provided written informed consent and children older than age 10 provided assent.

Measures

The primary outcome measure was internalizing and externalizing symptoms in the child. For children and younger adolescents, these were reported by mothers (7 measures, ages 1.5-8 years), fathers (8 measures, ages 1.5-13 years), and teachers (6 measures, ages 6-13 years), using the Social Behavior Questionnaire (SBQ).³² The SBQ integrates items from the Rutter Children’s Behavior Questionnaire,³³ the Child Behavior Checklist,³⁴ the Ontario Child Health Study Scales,³⁵ and the Preschool Behavior Questionnaire.³⁶ From ages 10 to 13 years, children self-reported their own mental health symptoms using the SBQ. At ages 15 and 17 years, they self-reported their symptoms using the Mental Health and Social Inadaptation Assessment for Adolescents (MIA).³⁷

The number of items in the SBQ varied by age. Internalizing symptoms were measured by SBQ subscales for emotional problems (3-6 items) and anxiety (3-4 items). Externalizing symptoms were measured by SBQ subscales for hyperactivity (4-6 items), inattention (2-4 items), aggressivity (12-14 items), opposition (3-5 items), and conduct behaviors (5-7 items). All items are rated on a 3-point Likert scale (never = 0; sometimes = 1; often = 2) and have good psychometric properties for assessing internalizing (Cronbach α = .64-.85) and externalizing (Cronbach α = .78-.94) symptoms during childhood³² and acceptable psychometric properties during infancy (internalizing: Cronbach α = .47-.57; externalizing: Cronbach α = .54-.83).³⁸

Internalizing symptoms in adolescence were measured by the MIA subscales for depression (8 items, α = .90), social phobia (8 items, α = .90), and generalized anxiety (9 items, α = .86). Externalizing symptoms were measured by MIA subscales for attention deficit with or without hyperactivity disorder (16 items, α = .89), conduct disorder (16 items, α = .95), opposition disorder (10 items, α = .84), and aggression (18 items, α = .96).³⁷ The total score for each subscale was standardized as 0 to 10 and scores for internalizing and externalizing symptoms were averaged to obtain scores for total mental health symptoms.

Exposure

Maternal and paternal postpartum depressive symptoms were self-reported by each parent when children were 5 months old on a 13-item scale combining a validated short version³⁹ (α = .81) of the Center for Epidemiological Studies Depression Scale (CES-D)⁴⁰ and 1 item from the Edinburgh Postpartum Depression Scale.⁴¹ Responses to each item ranged from 0 (none) to 3 (all the time), for a total score ranging from 0 to 39. Scores were dichotomized using the validated clinical cutoff of 16/60 on the original CES-D,⁴² equivalent to 10/39.

Moderating Variables

Family SES at baseline was created from 3 self-reported variables following a method proposed by Willms and Shields⁴³: years of education of parents, occupational prestige of parents, and household income. Years of education included primary school education up to MD/PhD education level and ranged from 0 to 20 years. Occupational prestige was based on a modified version of a 16-category scale developed by Pineo et al.⁴⁴ that provides a ranking of occupations according to their social standing or prestige. As this ordinal scale cannot assume that the intervals between ranks were equal, the percentage of individuals in each occupation group was considered as a piece of the logistic distribution using logit transformation, as proposed by Mosteller and Tukey.⁴⁵ Each of the 5 variables was standardized to have a mean of zero and SD of 1, and the SES score was then calculated by taking the average of the 5 standardized variables.⁴⁶

Covariates

We selected potential confounding factors associated with either exposure or outcome.47, 48, 49, 50, 51, 52, 53, 54 These included child sex, birth weight, term or preterm (<37 weeks’ gestation) birth, and neonatal risk index (score based on multiple obstetrical problems that reflects the overall health condition at birth), as obtained from obstetrical records; in utero exposure to smoking, alcohol, or drugs reported by the mothers 5 months postpartum; for both parents, immigration status, age group, education level, and occupational status; household information such as family types (intact vs reconstituted), number of siblings, and household income, all reported by the mother during interviews at baseline; and for both parents, lifetime history of depression and antisocial behaviors during adolescence and adulthood, as self-reported by each parent on a 0-to-10 scale adapted from the Diagnostic Interview Schedule III–Revised.⁵⁵

Statistical Analysis

We used cross-classified 3-level mixed effects modeling to test associations between PPD and mental health symptoms in children, with observations (level 1) nested within informants and waves of data collection (level 2), and informants and waves nested within families (level 3) (Figure 1). In addition to its tolerance for missing data, we chose mixed effects modeling primarily to preserve the maximum amount of data available in the QLSCD on children’s mental health. This type of model is often used to analyze longitudinal data, as it allows for the correlation of errors between measures reported at different waves of data collection for the same participant. In the case of this study, this type of model also made it possible to use data from several informants,³⁰ while considering the correlation of errors between the measures reported by each informant for the same wave of data collection.

Figure 1.

Flow Diagram Showing Cross Classified Multilevel Study Population Using Data From the Quebec Longitudinal Study of Child Development^a 1997-1998 Birth Cohort

Note:From the initial longitudinal sample of 2,120 families, 168 single-parent families and 53 families without any measurement of the outcome were excluded. The analyzed sample consisted of the remaining 1,899 families, providing a total of 33,848 observations, forming a cross-classified 3-tiered hierarchical (multilevel) data structure. Level 1 contains all the outcome measures as observed on the questionnaires. These observations are nested in waves of data collection (upper part of the figure), representing variations between informants within each wave, and nested in informants (lower part of the figure), representing longitudinal variations within informants. Waves and informants at level 2 are nested in families at level 3. Level 3 contains all covariates measures at the family level, representing variations between families. Cross-classified multilevel modeling was used to account for the correlation of errors between measures from multiple informants within the same wave, between repeated outcome measures within the same informant, and between measures within the same family. A. = adolescent self-reported; C. = child self-reported; F. = father-reported; M. = mother-reported; ob. = observations; T. = teacher-reported.^aData were compiled from the final master file of the Quebec Longitudinal Study of Child Development (1998-2015), ©Gouvernement du Québec, Institut de la statistique du Québec.

We included PPD and covariates as fixed effects at level 3, with random slopes to account for repeated measures of mental health symptoms experienced by children over time. We conducted exploratory analyses on internalizing and externalizing outcomes separately as well as in different age ranges for assessing children’s outcomes, to confirm the stability of the results. We based our covariates selection for the main analysis on the Akaike information criterion using a stepwise algorithm. We included the covariates selected for most of the imputed datasets in the final models, based on the majority method.⁵⁶ We tested interaction terms between PPD in each parent and family SES in the same model. If the interaction was significant (p < .05) with one or both parents, we further stratified our analyses by low vs medium or high SES using the first tercile of the distribution as the cutoff (after observing no moderating effect on the association of maternal depression when using the median as the cutoff in exploratory analyses). Considering the heterogeneity of the conceptualization and measurement of SES in prior literature,^22,25,26 we tested the interaction separately with the 3 variables used to construct the SES variable (parental years of education, parental occupational prestige, and household annual income). For the interaction with parental education or occupational prestige, we allowed each parent’s depression to interact with each parent’s years of education or occupational prestige, respectively. Finally, we tested the interaction between each parent’s depression and the child’s sex in their association with the child’s mental health. For all tests of interaction, we kept the interaction terms in the models only if they were at least marginally significant (p < .1).

We used multiple imputation to account for missing covariate data at baseline, with a missing at random assumption resulting in 20 imputed datasets. We performed sensitivity analyses on available (nonimputed) data. We applied inverse probability weighting to control for potential bias from selective censoring due to missing data on exposure or outcome variables, using the method described by Weuve et al.⁵⁷ With data from 1 imputed dataset, we used logistic regression to estimate the response probability, which was influenced by various factors, including the baseline characteristics of the index child (sex, birth weight, neonatal risk index, number of siblings), the parents (immigration status, education level, antisocial behaviors, age, lifetime history of depression, postpartum depression, principal occupation), and the household (source of income, income, SES, income sufficiency, neighborhood cohesion, family type). For each observation contributing to the analysis, the weights corresponded to the inverse probability of being uncensored, at specific waves by a particular informant. Covariates balance for each weight is presented in Figure S1, available online. We performed all analyses using R v4.1.0 (R Foundation for Statistical Computing, Vienna, Austria), and the linear mixed effect models (lme4),⁵⁸ multiple imputation by chained equations (MICE),⁵⁹ and tools for multiple imputation in multilevel modeling (mitml)⁶⁰ packages.

Results

Participants

Of the 2,120 families in the QLSCD sample, we excluded 168 single-parent families. Data on depressive symptoms in both parents and at least 1 measure of mental health symptoms in children from 1.5 to 17 years of age were available for 33,848 observations from 1,899 families (Figure 1). Children’s outcomes were reported only by the mothers in 41 of these families, but a multi-informant design was used in all other families.

Parental Postpartum Depression

Postpartum depression affected 314 mothers (16.5%) and 151 fathers (8.0%) in the analyzed sample. We used the term father generically to include the mother’s partner living in the household, which was the case in 0.5% of households. The sample did not include any same-sex parents. Table 1 shows participant characteristics at baseline by maternal and paternal depression after multiple imputation. Proportion of boys, birth weight, and proportion of children exposed to in utero alcohol did not vary between parents with depression and parents without depression, but the proportion of children exposed to in utero smoking increased when parents were depressed. Parents with depression had higher scores of antisocial behaviors were more likely to have a lifetime history of depression, have lower education, have lower occupational prestige, and live in lower socioeconomic households with a lower annual income. Parents born outside Canada or Europe tended to be more depressed, although the differences between groups with or without depression were statistically significant only for mothers. Mothers younger than 25 years old were more likely to be depressed. Fathers older than age 40 tended to be more depressed than fathers younger than age 40, but this was not statistically significant. Fathers living in 2-parent intact families were less likely to have depression than fathers living in reconstituted families.

Table 1.

Baseline Child, Father, Mother, and Household Characteristics in the Quebec Longitudinal Study of Child Development^a by Parental Depression^b

	Mother					Fatherc
	Without depression (n = 1,585)		With depression (n = 314)		p	Without depression (n = 1,748)		With depression (n = 151)		p
	n	(%)	n	(%)		n	(%)	n	(%)
Child characteristics
Boys	791	(49.9)	167	(53.2)	.32	882	(50.5)	76	(50.3)	1
	Mean	(SD)	Mean	(SD)		Mean	(SD)	Mean	(SD)
Birth weight, g	3,420	(495)	3,420	(482)	.87	3,420	(496)	3,400	(462)	.34
	n	(%)	n	(%)		n	(%)	n	(%)
In utero exposure to smoking	365	(23.0)	97	(30.9)	.004	415	(23.7)	47	(31.1)	.05
In utero exposure to alcohol	594	(37.5)	110	(35.0)	.45	655	(37.5)	49	(32.5)	.26
Parent’s characteristicsd					<.001
Ethnic origin										.41
African/Haitian	15	(0.9)	7	(2.2)		27	(1.5)	3	(2.0)
British	66	(4.2)	5	(1.6)		51	(2.9)	4	(2.6)
Canadian	1,029	(64.9)	199	(63.4)		1,159	(66.3)	92	(60.9)
French	256	(16.2)	31	(9.9)		259	(14.8)	26	(17.2)
Native American	7	(0.4)	5	(1.6)		10	(0.6)	3	(2.0)
Other European	84	(5.3)	11	(3.5)		81	(4.6)	6	(4.0)
Other	128	(8.1)	56	(17.8)		161	(9.2)	17	(11.3)
Age at childbirth					.003					.18
<25 y	349	(20.6)	104	(28.5)		154	(8.3)	15	(7.1)
25-39 y	1,307	(77.2)	251	(68.8)		1,506	(81.6)	153	(72.5)
≥40 y	36	(2.1)	10	(2.7)		132	(7.2)	21	(10.0)
	Mean	(SD)	Mean	(SD)		Mean	(SD)	Mean	(SD)
Antisocial behaviorse	0.99	(1.14)	1.31	(1.37)	<.001	1.38	(1.63)	1.66	(1.76)	.03
	n	(%)	n	(%)		n	(%)	n	(%)
Lifetime history of depression	206	(13.0)	94	(29.9)	<.001	199	(11.4)	53	(35.1)	<.001
	Mean	(SD)	Mean	(SD)		Mean	(SD)	Mean	(SD)
Years of education	12.3	(2.89)	11.3	(2.56)	<.001	12.0	(2.94)	11.4	(2.73)	.003
	n	(%)	n	(%)		n	(%)	n	(%)
Occupational prestigef					<.001					.01
Professional/high management	301	(17.8)	26	(7.1)		352	(19.1)	54	(25.6)
Semiprofessional/technician/middle management	367	(21.7)	47	(12.9)		380	(20.6)	57	(27.0)
Supervisor/clerical/sales/services	760	(44.9)	196	(53.7)		232	(12.6)	20	(9.5)
Foreman/skilled manual/farmer	149	(8.8)	42	(11.5)		686	(37.2)	60	(28.4)
Unskilled manual	115	(6.8)	54	(14.8)		196	(10.6)	20	(9.5)
Household characteristics
Biparental intact family unit	1,392	(87.8)	273	(86.9)	.73	1,541	(88.2)	124	(82.1)	.04
Number of siblings					.09					.17
0	723	(42.7)	139	(38.1)		777	(42.1)	85	(40.3)
1-2	880	(52.0)	198	(54.2)		970	(52.5)	108	(51.2)
≥3	89	(5.3)	28	(7.7)		99	(5.4)	18	(8.5)
	Mean	(SD)	Mean	(SD)		Mean	(SD)	Mean	(SD)
Annual incomeg, CAD$1000	49.5	(24.9)	36.2	(22.6)	<.001	48.5	(24.8)	35.5	(24.3)	<.001
Socioeconomic statush	0.10	(0.99)	−0.40	(0.95)	<.001	0.05	(0.98)	−0.38	(1.08)	<.001

Note: Data represent a single imputation of missing data by chained equations. The p values are based on t test or Kruskal-Wallis test for continuous variables and Pearson χ² test for categorical variables. Boldface indicates significant associations at p < .05.

^aData were compiled from the final master file of the Quebec Longitudinal Study of Child Development (1998-2015), ©Gouvernement du Québec, Institut de la statistique du Québec.

^bClinical cutoff 16/60 in original Center for Epidemiological Studies Depression Scale (CES-D), equivalent to 10/39 in this study.

^cIn 11/1,899 families, the mother’s partner living in the household is reported as the father.

^dMothers’ characteristics are presented on the left side and fathers’ characteristics are presented on the right side.

^eScore on 0-to-10 scale based on self-reported antisocial behaviors from adolescence to adulthood adapted from the Diagnostic Interview Schedule III–Revised.

^fStatistics Canada 1980 Standard Occupational Classification of 16 ranks based on social status or prestige, reduced to 5 categories.

^gBased on a 9-level ordinal variable representing annual household income brackets, transformed into a continuous variable using the central value of each bracket.

^hStandardized score built from the following 5 variables: years of education and occupational prestige of each parent and household income.

Mental Health Symptoms in Children From Infancy to Adolescence

Data on mental health symptoms in children from 1 to 17 years of age were available for 33,848 observations from 1,899 families (Figure 1). Figure S2, available online, shows the average mental health symptom scores for each wave of data collection by each informant in the analytic sample. Overall, the scores ranged from 1.0 to 3.6 on a 10-point scale. Pearson correlations between mother and father reports ranged from r = 0.40 to r = 0.43. Correlations between teacher and mother reports ranged from r = 0.24 to r = 0.28, and correlations between teacher and father reports ranged from r = 0.21 to r = 0.29. Correlations between child and father reports ranged from r = 0.27 to r = 0.32, and correlations between child and teacher reports ranged from r = 0.24 to r = 0.32.

Association of Maternal and Paternal Depression With Mental Health Symptoms in Children

The associations of maternal and paternal PPD with total mental health symptoms experienced by children are presented in Table 2. After controlling for family SES and other covariates, maternal PDD (β = .27, 95% CI 0.20 to 0.35, p < .001) and paternal PPD (β = .14, 95% CI 0.04 to 0.25, p = .007) were associated with increased total mental health symptoms in children.

Table 2.

Moderating Effect of Family Socioeconomic Status (SES) on Associations^a of Maternal or Paternal Postnatal Depression With Mental Health Symptoms in Children From 1.5 to 17 Years of Age in the Quebec Longitudinal Study of Child Development^b

Variable	Model 1			Model 2			Model 3
	β	(95% CI)	p	β	(95% CI)	p	β	(95% CI)	p
Maternal depression	0.38	(0.3 to 0.46)	<.001	0.28	(0.2 to 0.36)	<.001	0.25	(0.17 to 0.33)	<.001
Paternal depression	0.18	(0.07 to 0.29)	.001	0.15	(0.04 to 0.26)	.005	0.14	(0.03 to 0.24)	.01
Child’s sex	—		—	—		—	0.12	(0.06 to 0.17)	<.001
In utero exposure to smoking	—		—	—		—	0.01	(0 to 0.01)	.01
In utero exposure to alcohol	—		—	—		—	0.04	(0.01 to 0.08)	.02
Maternal lifetime depression	—		—	—		—	0.13	(0.05 to 0.21)	.001
Paternal lifetime depression	—		—	—		—	0.02	(−0.08 to 0.11)	.73
Maternal age at birth	—		—	—		—	−0.01	(−0.02 to 0)	.001
Paternal age at birth	—		—	—		—	−0.01	(−0.01 to 0)	.02
Maternal antisocial behaviors	—		—	—		—	0.04	(0.01 to 0.06)	.002
Paternal antisocial behaviors	—		—	—		—	0.03	(0.02 to 0.05)	<.001
Biparental intact family	—		—	—		—	−0.11	(−0.19 to −0.02)	.01
SESc	—		—	−0.15	(−0.19 to −0.12)	<.001	−0.08	(−0.12 to −0.05)	<.001
Interaction
Maternal depression × SES	—		—	−0.13	(−0.22 to −0.04)	.004	−0.13	(−0.21 to −0.04)	.003
Paternal depression × SES	—		—	−0.13	(−0.25 to −0.01)	.03	−0.13	(−0.24 to −0.01)	.03

Note: Boldface indicates significant associations at p < .05. β = estimate.

^aAll models were weighted to account for attrition using propensity scores from 24 predictors of dropout.

^bData were compiled from the final master file of the Quebec Longitudinal Study of Child Development (1998-2015), ©Gouvernement du Québec, Institut de la statistique du Québec.

^cStandardized score built from the following 5 variables: years of education and occupational prestige of each parent and household income.

Moderating Role of Family SES

Family SES did moderate the association between both parents’ PPD and children’s mental health symptoms. As familial SES increased, the strength of the associations decreased, with an interaction effect of β = −.13 (95% CI −0.21 to −0.04, p = .003) for maternal depression and an interaction effect of β = −.13 (95% CI −0.24 to −0.01, p = .03) for paternal depression (Table 2). Stratified analyses showed that paternal PPD was associated with increased mental health symptoms in children only in low-SES families (β = .27, 95% CI 0.10 to 0.45, p = .003). Meanwhile, maternal PPD was associated with increased mental health symptoms in children in both low SES (β = .38, 95% CI 0.25 to 0.50, p < .001) and medium or high SES (β = .20, 95% CI 0.10 to 0.29, p < .001) families (Figure 2).

Figure 2.

Associations Between Parental Postpartum Depression and Children’s Total Mental Health Symptoms from 1 to 17 years by Family Socioeconomic Status (SES) in the Quebec Longitudinal Study of Child Development^a

Note:β coefficients are shown of the associations between parental postpartum depression in each parent and mental health symptoms in the child, obtained from mixed effect analyses, and stratified by family SES. Effect sizes are larger in low SES families than in medium or high SES families. The association is not significant for paternal depression in medium or high SES families.^aData were compiled from the final master file of the Québec Longitudinal Study of Child Development (1998 – 2015), ©Gouvernement du Québec, Institut de la statistique du Québec.

We further tested the interaction between PPD and the 3 dimensions of family SES in association with mental health symptoms in children (Table 3). The parental education model revealed that as maternal years of education increased, the strength of the association between paternal PPD and outcomes in children decreased (β = −.11, 95% CI −0.21 to −0.01, p = .03). Maternal years of education tended to moderate the association between maternal PPD and children’s outcomes as well, but the interaction was only marginally significant (β = −.08, 95% CI −0.16 to 0.00, p = .06). The household income model revealed that as income increased, the association between maternal PPD and mental health symptoms in children weakened, but the interaction was only marginally significant (β = −.07, 95% CI −0.15 to 0.01, p = .097). Parental occupational prestige did not moderate any association.

Table 3.

Moderating Effects of Family Socioeconomic Dimensions on the Associations^a of Maternal or Paternal Postnatal Depression With Mental Health Symptoms From 1.5 to 17 Years of Age in the Quebec Longitudinal Study of Child Development^b

Variables	Parental years of education			Parental occupational prestigec			Household incomed
	β	(95% CI)	p	β	(95% CI)	p	β	(95% CI)	p
Maternal depression	0.26	(0.18 to 0.34)	<.001	0.28	(0.2 to 0.36)	<.001	0.25	(0.17 to 0.33)	<.001
Paternal depression	0.15	(0.04 to 0.25)	.007	0.15	(0.04 to 0.26)	.008	0.13	(0.02 to 0.23)	.02
Maternal years of education	−0.03	(−0.06 to 0.01)	.10	—		—	—		—
Paternal years of education	−0.04	(−0.07 to 0)	.029	—		—	—		—
Maternal occupational prestige	—		—	−0.06	(−0.09 to −0.02)	.001	—		—
Paternal occupational prestige	—		—	−0.03	(−0.06 to 0)	.095	—		—
Household annual income	—		—	—		—	−0.08	(−0.11 to −0.04)	<.001
Interactions
Maternal depression ×
Maternal years of education	−0.08	(−0.16 to 0)	0.06	—		—	—		—
Paternal years of education	—		ns	—		—	—		—
Maternal occupational prestige	—		—	—		ns	—		—
Paternal occupational prestige	—		—	—		ns	—		—
Household annual income	—		—	—		—	−0.07	(−0.15 to 0.01)	.097
Paternal depression ×
Maternal years of education	−0.11	(−0.21 to −0.01)	.03	—		—	—		—
Paternal years of education	—		ns	—		—	—		—
Maternal occupational prestige	—		—	—		ns	—		—
Paternal occupational prestige	—		—	—		ns	—		—
Household annual income	—		—	—		—	—		ns

Note: Boldface indicates significant associations at p < .05. β = estimate; ns = nonsignificant interaction (p > .1).

^aAll models are adjusted for the child’s sex and in utero exposure to smoking and alcohol; parent’s lifetime history of depression, age at birth, and antisocial behaviors; and family type (intact or reconstituted) as fixed effects, with random effects on informants and children, and weighted to account for attrition using propensity scores from 24 predictors of dropout.

^bData were compiled from the final master file of the Quebec Longitudinal Study of Child Development (1998-2015), ©Gouvernement du Québec, Institut de la statistique du Québec.

^cBased on Statistics Canada 1980 Standard Occupational Classification of 16 ranks based on social status or prestige, transformed into a continuous variable and standardized.

^dBased on a 9-level ordinal variable representing annual household income brackets, transformed into a continuous variable using the central value of each bracket and standardized.

Sensitivity Analysis

Table S1, available online, presents the interaction model with raw data (not imputed) for paternal depression or without the weighting. The estimates and CIs were almost identical. Separate analyses for internalizing and externalizing symptoms yielded similar results (Table S2, available online). Sensitivity analyses were also conducted using a sample that excluded families without a multi-informant design (Table S3, available online), yielding nearly identical results as the analyses performed on the entire sample. Furthermore, analyses that examined children’s outcomes up to preschool or school age produced outcomes similar to those obtained in the analyses that examined outcomes in children up to adolescence, but the main and interactive associations of paternal depression tended to diminish as the period was extended (Table S4, available online). Finally, the main analyses were not moderated by child sex.

Discussion

Using a multi-informant, population-based cohort study, we examined the association between clinically elevated levels of maternal and paternal postpartum depressive symptoms and mental health symptoms in children from infancy to adolescence, as well as the moderating role of family SES. PPD in both parents was associated with worse mental health symptoms, and these associations strengthened as family SES decreased. Exploration of the moderating effects of the various dimensions of family SES on these associations revealed that maternal education interacted with maternal and paternal depression in their associations with mental health symptoms in children, whereas paternal education did not.

The moderating effect of SES observed on the associations between both maternal and paternal postpartum depression and mental health from childhood to adolescence contributes to advancing knowledge in this area. First, these results support the theory of the accumulation of risks over the life course, which posits that the exposure to multiple stressors such as PPD in the context of poverty is associated with worse outcomes than the addition of the outcomes related to these risks taken separately.⁶¹ Second, our results contrast with those of Malmberg and Flouri,²² who found no interaction between PPD and socioeconomic disadvantage in their association with mental health in children. However, their measure was based on overcrowding, lack of home ownership, receipt of income support, lack of access to a car or van, and income poverty. The contrasting results of this study with ours could be explained by the fact that this measurement strictly targets the material aspects of SES and excludes the level of education. Indeed, the level of education provides parents with internal and social resources that could possibly help them contribute to meeting their child’s needs despite depression. Therefore, our results fill a gap in prior literature. Among the 3 studies that have already tested this interaction in both parents,^20,22,25 we are the first to our knowledge to use a global measure of SES at the family level, including education, occupation, and income of both parents.

Exploring the interactions between different dimensions of SES and PPD in their association with mental health of children allowed us to contextualize our results with previous studies that specifically focused on the moderating effect of parental education level. Our findings supported the results of Pearson et al.,²⁵ who observed a moderating influence of parental education on the associations between PPD in each parent and the clinical depression diagnosis in their children at 18 years old. However, our results contrasted with the work of Gutierrez-Galve et al.,²⁶ who did not observe such an effect on the associations between PPD and overall mental health symptoms in children at 3.5 years old. It might have been plausible to explain the contrasting results between the work of Pearson et al.²⁵ and Gutierrez-Galve et al.²⁶ by suggesting that the child’s mental health was not measured at the same age. However, our sensitivity analyses across different age ranges indicated consistent trends, dismissing this hypothesis. Furthermore, the divergent results could have been explained by the analyses of Pearson et al.²⁵ incorporating more confounding variables, potentially rendering the associations nonsignificant among high SES groups. Nevertheless, our unadjusted model displays a significant interaction, contradicting this explanation. In conclusion, although our results helped eliminate certain hypotheses to explain these contradictory findings, their underlying causes remain elusive, emphasizing the need to systematically test the moderating effects of SES or parental education level on such associations.

Finally, the fact that the association between paternal PPD and mental health in children was moderated by maternal education, but not by paternal education, is unexpected and novel. This might be explained by the differences in mechanisms specific to maternal or paternal depression. For example, prior studies found that negatively biased perceptions,⁶² warmth,⁶³ parenting,²³ support from the other parent,^20,64 or the other parent’s relationship with the child²² mediated the association of maternal depression with internalizing problems in children, but not of paternal depression. On the other hand, Gutierrez-Galve et al.^20,26 observed that factors related to the mother, such as later maternal depression and couple conflict, explained a greater proportion of the association between paternal depression and outcomes in children than direct paternal involvement with the child. As maternal education is a better predictor of these mother-related factors than paternal education, it is understandable why maternal education would moderate the association of paternal PPD with children’s outcomes, whereas paternal education would not.

Our findings have valuable implications for public health strategies focusing on preventing the intergenerational transmission of mental health problems and reducing socioeconomic inequalities in child mental health. Universal interventions, when offered at the same intensity to the entire population, may inadvertently increase social inequalities in child mental health rather than alleviate it.⁶⁵ A meta-analysis revealed that preventive interventions for children with mentally ill parents showed larger effect sizes in samples with higher SES.⁶⁶ To avoid increasing the social gradient through public health interventions, it might be more effective to target vulnerable groups who are most in need of mental health promotion. Targeted interventions designed to prevent maternal PPD in low SES areas are effective⁶⁷ and could also be efficient in the context of limited resources. Given that our results contrasted with some previous studies, further research is necessary before making recommendations regarding the best strategies to adopt. However, our findings seem to favor the universal proportionate approach, which offers a combination of low-intensity universal interventions, with more intense targeted interventions for high-risk groups.⁶⁸

This study has several strengths and some limitations. We used a large birth cohort sample, representative of the population of Quebec, Canada. We measured outcomes of children from infancy to adolescence up to 17 years old, which allowed us to interpret our results in a life course perspective. We systematically included measures from both parents for the dependent variable and all covariates to consider the whole family system in our analyses. We used a multi-informant design and a solid methodological approach to reduce potential same-informant bias. Discrepancies in the assessment of mental health symptoms in children between mothers and fathers are acknowledged and attributed to their unique perspectives on the child’s behavior.⁶⁹ Recognizing that individual characteristics of informants can impact their reports, researchers advocate for employing multiple informants to offer a comprehensive understanding of the child’s behavior, especially when parental psychopathology is present.²⁹ Likewise, the observations of a child’s behavior in various contexts, such as home vs school, seems to provide different but complementary perspectives.⁷⁰ Finally, we used robust methods to deal with our missing data and mitigate selective censoring bias.

Despite these strengths, this study has limitations. As this is an observational study, it is not possible to draw causal inferences from our results. The mental health assessments of the parents and the children were based on reports from parents, children, and teachers, which are not as accurate as diagnostic interviews with clinicians. However, data on internalizing or externalizing symptoms were available through parents and teachers who saw the child daily, as well as by self-reports in adolescence. Additionally, the observed prevalence of clinically elevated depressive symptoms in mothers (17%) and fathers (8%) in our study is similar to that reported in the literature for mothers (16%-17%)^5,71,72 and fathers (13%)^6,73 in Canada, suggesting that self-reported scales effectively estimated the prevalence of clinical depression. The secondary use of existing data limited the availability of data at certain time points. For example, it was impossible to control for the continuation of paternal depressive symptoms after the postpartum period, as it was not measured after 5 months. As a previous study conducted with the same cohort highlighted the importance of the chronicity of maternal depressive symptoms in relation to child development,⁷⁴ further research should consider this aspect with paternal depression. Furthermore, children’s mental health symptoms were not reported by the father at age 8 and were not reported by the mother at ages 10 and 13. However, the aim of the study was to estimate the association between postpartum parental depression and the average of children’s symptoms across childhood and adolescence, as represented by the complementary perspectives of various informants. The use of multilevel modeling, applying random effects over time and informant, ensured that differences in the timing of measurements among informants did not influence the results. Finally, the respective contributions of each informant are relatively balanced, with the child reporting 5 measurements; the teacher, 6; the father, 7; and the mother, 6. Because only data on lifetime history of depression or antisocial behavior were available, our results might be explained in part by unmeasured confounding factors, such as lifetime history of other psychiatric disorders. Due to the ethnic homogeneity of our sample, our results might not be generalizable to populations with different ethnic or cultural backgrounds. The moderating role of family SES on the associations between PPD and mental health outcomes of children should be investigated using professional diagnostic exposure and outcome measures in ethnically and socioeconomically diverse samples. Our findings may be limited to the social and economic context of Quebec, Canada, during the late 1990s to 2010s. As the impact of SES is highly dependent on context, further research is needed to examine its moderating effect on the relations between parental depression and child socioemotional development in other jurisdictions where universal health care is unavailable or where parental leave programs differ, for example. The absence of same-sex parents in our sample and the fact that the gender of parents was not assessed limit the generalizability of our results for LGBTQIA2+ (lesbian, gay, bisexual, transgender, queer [or questioning], intersexual, asexual, and 2-spirited) families. Further research should be conducted with more recent cohorts designed to be more inclusive in the recruitment process and assessing the parents’ sex and gender to ensure the generalizability of the results to LGBTQIA2+ families and reduce health iniquities for this population.

The presence of PPD in both parents was associated with worse mental health symptoms in children and adolescents, and these associations strengthened as family SES decreased. To address these issues, more research should investigate the moderating role of SES and its various dimensions in the intergenerational transmission of mental health problems. More research is also needed to investigate the mechanisms related specifically to associations of maternal or paternal PPD with mental health outcomes in children. Exploring mechanisms distal to the child and more on the familial contextual level might help to identify good targets for intervention on a more structural level, which is known to reduce health inequalities more effectively than individual-level interventions.⁶⁵

CRediT authorship contribution statement

Myriam Clément: Writing – review & editing, Writing – original draft, Visualization, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Massimiliano Orri: Writing – review & editing, Methodology, Investigation, Formal analysis, Conceptualization. Marilyn N. Ahun: Writing – review & editing, Methodology, Investigation, Conceptualization. Pascale Domond: Writing – review & editing, Investigation. Gregory Moullec: Writing – review & editing, Supervision, Methodology, Investigation, Conceptualization. Sylvana M. Côté: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization.