Maternal postnatal depression and anxiety and their association with child emotional negativity and behavior problems at two years.

Procedure

At 9 weeks postpartum, the EPDS and GAD-Q were administered to mothers to screen for symptoms of MDD and GAD. At 3 months postpartum the Structured Clinical Interview for DSM–IV Disorders was administered to mothers in their homes to assess the presence or absence of MDD and GAD as well as associated Clinician Severity Ratings (described below in Measures). At subsequent assessments (6, 10, 14, and 24 months postpartum), mothers completed the EPDS and GAD-Q and were reinterviewed using the Structured Clinical Interview for DSM–IV Disorders. The following number of mothers provided data: 296 at 6 months postpartum, 253 at 10 months postpartum, 233 at14 months postpartum, and 234 at 24 months postpartum.

At 24 months postpartum, in addition to the maternal assessment described above, mothers also completed the Child Behavior Checklist for Ages 1.5–5 and Early Childhood Behavior Questionnaire (described below in Measures). Also at 24 months postpartum, mothers and children participated in a standardized semistructured mother–child interaction play task for 10 min (e.g., NICHD Early Child Care Research Network, 1999 ). The mother–child interaction play task involved mother and child playing together with a toy farm set. The farm roof and door were given to the mother and child to assemble as they wished, and several rubber farm animals, a plastic tractor, four sets of wooden fences, and a wooden ladder were also provided. Upon completion of the mother–child interaction play task, the child completed a standardized semistructured individual play task: the child continuing to play with the farm set on his or her own for 5 min. while his or her mother completed questionnaires nearby.

Also at 24 months postpartum, children participated in a frustration inducing toy removal procedure, called the barrier paradigm, that originated from the Laboratory Temperament Assessment Battery ( Goldsmith & Rothbart, 1991 ). Such toy removal procedures have been successfully used to elicit frustration in 24-month-old children (e.g., Blair et al., 2015 ; Diener & Mangelsdorf, 1999 ). Mother and child sat side by side at a table with the researcher opposite the mother. The mother was asked not to interact with or respond to the child. The child was given a toy to explore for 15 seconds; the toy was then placed behind a transparent screen located at arm’s length from the child for 30 seconds. The child could see but not reach the toy. The toy was then removed and this process repeated for two additional toys. From the child’s perspective, each play session terminated unexpectedly and the child had to cope with the toy being taken away while still visible.

Maternal GAD

Interview assessed GAD clinical severity ratings served as one indicator of the latent variable representing maternal GAD symptom severity. GAD clinical severity ratings were measured using the same 0- to 8-point scale discussed above for maternal MDD, with ratings of 4 or greater indicating clinical severity and ratings of 2 or 3 indicating subclinical symptom severity. The GAD-Q, a well-validated 9-item questionnaire that assesses GAD symptoms, was divided into two subscales that also served as maternal GAD latent variable indicators symptom severity. As above, two subscales were used to decrease the number of parameters estimated and improve model convergence rates; as with maternal MDD, these subscales are referred to as Sx 1 and Sx 2 (for GAD symptoms) in Figure 1 and the Data Analysis section. These subscales had alpha reliability estimates of .85 and .88 at 3 months postpartum.

Child Behavior Checklist for Ages 1.5–5 (CBCL)

The CBCL ( Achenbach, Edelbrock, & Howell, 1987 ), a well-validated 103-item questionnaire, was completed by the mother. The CBCL assessed children’s emotional and behavioral problems and competencies in the two months prior to the 24 months postpartum assessment. In addition to the CBCL total score (total problems), three subscales were used for the current analyses: attention, internalizing, and externalizing problems. The internalizing subscale measures behaviors related to anxiety and depression, whereas the externalizing subscale measures behaviors related to aggression, hyperactivity, and noncompliance. The attention subscale assesses behaviors related to attention problems such as difficulty maintaining attention.

Early Childhood Behavior Questionnaire (ECBQ)

The ECBQ is a well validated instrument for assessing temperament in children between the ages of 18 and 36 months. Items from four dimensions of the ECBQ ( Putnam, Gartstein, & Rothbart, 2006 ) were administered to mothers to assess key components of child temperament: child soothability, frustration, attentional focusing, and attentional shifting. Child soothability refers to the rate of recovery from distress, excitement, or general arousal. Frustration is the negative affect that is related to the interruption of ongoing tasks. Soothablity and frustration are both indicators of negative emotional reactivity. Attentional focusing refers to the ability to sustain orienting on an object of attention, or to resist distraction. Attentional shifting is the ability to transfer focus of attention from one activity to another.

Observational measures

Two cameras recorded the mother–child interaction play task, child individual play task, and barrier paradigm. The cameras had different viewing angles connected to a split-screen generator to permit precise coding. Videotaped recordings of each interaction were coded by a single rater blind to maternal diagnostic status and study hypotheses. The rater was trained to reliability with a gold-standard rater prior to rating study videotapes. Checks were made periodically to ensure that there was no drift. A set of behaviors reflecting the constructs being examined were rated on a predefined ordinal scale; scores derived in this manner for the constructs assessed below have demonstrated adequate reliability and validity (e.g., Murray, Fiori-Cowley, Hooper, & Cooper, 1996 ; Stein et al., 2012 ; Stein, Woolley, Cooper, & Fairburn, 1994 ). Cases were randomly selected and coded by a second rater for reliability purposes. Square-weighted Kappa (κ) values (provided below) demonstrated good interrater reliability for all of the observational measures.

Child negative emotional reactivity was coded in response to each play termination during the barrier paradigm on a scale ranging from 1 ( very distressed ) to 5 ( calm ); these three measurements were averaged as the index of child negative emotional reactivity. Ratings were based on the extent to which children remained calm or displayed distress in response to the frustrating task (play termination). There is evidence that child responses to such toy removal procedures index negative emotional reactivity ( Blair et al., 2015 ). Interrater reliability for child negative emotional reactivity was κ = .786.

Child negative emotional tone

Child negative emotional tone was coded on a scale from 1 ( very unhappy ) to 5 ( very happy ) with ratings made on the basis of vocalizations, facial expressions, and behavioral responses during the first two play tasks and after each play termination during the barrier paradigm. These five measurements were averaged as the index of child negative emotional tone. Such scores have evidenced reliability and validity at assessing child negative emotional tone (e.g., Stein et al., 1994 ). Interrater reliability for child negative emotional tone was κ = .775.

Maternal sensitivity

Maternal sensitivity toward their child was coded on a scale ranging from 1 ( not at all sensitive to child’s needs ) to 5 ( highly sensitive ). Ratings were based on the extent to which the mother responded to her child’s signals and communications, taking account of whether her responses were appropriate, prompt, and warm in manner. Maternal sensitivity was coded once every 5 min. during the first play task and these two measurements were averaged as the index of maternal sensitivity. Such scores have evidenced reliability and validity at assessing maternal sensitivity (e.g., Halligan et al., 2013 ; Murray et al., 1996 ). Interrater reliability for maternal sensitivity was κ = .716.

Data analysis

Mplus version 5.0 statistical software ( Muthén & Muthén, 1998 ) was used for structural equation modeling. Participant attrition was not significantly predicted from 3 months postpartum GAD or MDD symptom severity. Thus, missing data were accommodated using full information maximum-likelihood under the assumption of missing at random. Model goodness of fit was evaluated using the root mean square error of approximation (RMSEA) and the comparative fit index (CFI). To conclude good fit between the observed data and hypothesized model, RMSEA should be less than .06 and CFI should be greater than .95 ( Hu & Bentler, 1998 ; Yu, 2002 ).

First, descriptive statistics for all study variables will be examined, including variable means, standard deviations, and correlations among all study variables. Next, longitudinal measurement models will be examined separately for maternal MDD and GAD symptoms to ensure that manifest variables are significant indicators of their latent constructs at each of the five time points. The bottom half of Figure 1 illustrates the state latent variables (S 3M-P to S 24M-P ) and their corresponding manifest variable indicators (clinical severity ratings, Sx 1 , and Sx 2 ) at each time point. For maternal MDD symptom severity, S t represents state standing on the MDD symptom severity latent variable at time-point t , indicated by MDD clinical severity ratings and the EPDS subscales (described above) at time t . Similarly, for maternal GAD symptom severity, S t represents state standing on the GAD symptom severity latent variable at time-point t , indicated by GAD clinical severity ratings and GAD-Q subscales (described above) at time t . Removal of the top half of Figure 1 (the O and T latent variables), and the addition of pathways allowing correlations among all of the state latent variables at each time-point results in longitudinal measurement models for MDD and GAD symptom severity. If longitudinal measurement models for MDD and GAD symptom severity fit the data well, then metric invariance with time will be examined to determine if latent construct indicators function the same way at each point. This is done to ensure that actual change in the latent construct with time is not confounded by change in the measurement of the construct with time.

Next, full TSO models (see Figure 1 ) will be examined by removing correlational pathways among the state latent variables from the longitudinal measurement models and including autoregressive occasion factors at each time point (O t ) and a stable trait factor (T). TSO model fit will be examined separately for maternal MDD and GAD symptoms. If TSO models for MDD and GAD symptom severity fit the data well, information on the chronicity of maternal MDD and GAD symptoms will be provided. Specifically, the TSO model enables variance in state standing (S t ) of symptom severity at each time point to be completely partitioned into that explained by standing on the stable trait factor (symptom chronicity), and that which is not chronic throughout the period under consideration: that explained through the prior occasion factor through the autoregressive pathway and that unexplained by knowing state standing at other time points.

Next, for descriptive purposes, individuals will be grouped based on their continuous MDD and GAD symptom severity trait factor scores. Specifically, four groups will be created for each disorder by dividing individuals into quartiles based on their trait factor scores. Then, for each of these groups, information will be provided about average clinical severity ratings across all time points and number of time points with MDD and GAD. This information is purely descriptive in nature and is meant to provide insight into how the TSO model represents chronicity by providing descriptive information about those who score at different levels on factor representing chronicity in the model, the trait factor.

Key study hypotheses will be tested next using regression analyses conducted within structural equation modeling and using the TSO framework. Specifically, each child outcome will be examined independently of the other outcomes and separate models will be tested for maternal MDD and GAD symptom severity; all models will first be examined to determine if they fit the data well. To examine the hypothesis that that the chronic component of maternal symptomatology would be related to child outcomes, each child outcome at 24 months postpartum will be regressed onto the trait factor (β T in Figure 2 ). This will be done while simultaneously regressing each child outcome at 24 months postpartum on all of the occasion factors (β O 3 M - P , β O 6 M - P , β O 10 M - P , β O 14 M - P , β O 24 M - P , Figure 2 ) to examine whether or not there are sensitive periods during development regarding child effects of exposure to maternal symptomatology (it is not hypothesized that there will be). Thus, each Figure 2 regression pathway represents the unique relation between a TSO model component and child outcome, when accounting for the relations between the other TSO model components and the child outcome.

As seen in Table 1 , marital status and infant birth order differed significantly between diagnostic status groups. Given that these variables are potentially confounded with measures of MDD and GAD symptom severity, they will be included as predictors of child outcomes in all models. However, they have not been included in Figure 2 for clarity purposes.

For child outcomes that are found to be significantly predicted by both MDD and GAD TSO components, MDD and GAD TSO models will be combined in a single model to determine if MDD symptom severity independently predicts these child outcomes when accounting for GAD symptom severity, and if GAD symptom severity independently predicts these child outcomes when accounting for MDD symptom severity. Given the known strong relation between MDD and GAD, trait factors will be all