Prenatal Stress Exposure Amplifies Effect of Maternal Suicidal Ideation on Offspring Early Childhood Behavioral Trajectories

Abstract

The in-utero environment influences fetal development, and may predispose to disease later in life. This study examines whether maternal suicidal ideation during pregnancy is associated with children’s behavioral trajectories across early childhood, and whether prenatal maternal traumatic stress accelerates the trajectories. The study included mother-child dyads (N=331, 51.1% boys) from the longitudinal Stress In Pregnancy study; 31.1% (n=103) mothers were Exposed to Superstorm Sandy. During their second trimester, 12.4% (n=41) women reported suicidal ideation during pregnancy. Mothers completed the Behavior Assessment Scale for Children-2 annually from ages 2- to 6-years-old to assess multiple behavioral domains. Hierarchical linear modeling estimated within-person longitudinal trajectories of clinical behaviors, and between-person effects of maternal suicidal ideation and disaster-related stress in-utero on changes in child behavior. For children exposed to both risks, Atypical behaviors (i.e., unusual behaviors, social disconnection) increased linearly across early childhood. Exposure to Superstorm Sandy and maternal suicidal ideation were independently associated with non-linear increases in Anxiety severity and maternal suicidal ideation during pregnancy was associated with a linear increase in Attention problems across early childhood. Maternal suicidal ideation during pregnancy is associated with increased risk for a range of behavioral and emotional difficulties in early childhood and the trajectory of atypical behaviors was amplified by disaster-related traumatic stress. Findings highlight the need for health professionals to screen for suicidal ideation among their pregnant patients. Pregnant women who experience severe stress may require additional monitoring and support to reduce risk for poorer early childhood outcomes.

The prenatal period represents a critical period for development for all bodily systems. As such, the embryo/fetus is highly sensitive to its surrounding environment (Barker, 2004). Exposure to environmental insults during this developmental period can lead to structural, metabolic, and physiological changes, which may ultimately predispose to disease later in life (Barker, 2004). Maternal mental health is associated with fetal and child health outcomes, possibly through changes to how the body regulates its stress response (van den Bergh et al., 2020). This study examines whether the interaction of two increasingly common stressors affecting pregnant women – suicidal ideation and natural disaster stress – is associated with trajectories of young children’s behavioral and emotional development.

Suicide is a leading (and likely under reported) cause of maternal mortality (Gelaye et al., 2016; Mangla et al., 2019). Suicidal ideation, a major risk factor for attempted suicide, is common during pregnancy, with prevalence rates as high as 33% of pregnant women (Newport et al., 2007). Risk of suicidal behaviors, including ideation, is further amplified for pregnant women who are experiencing significant stress. Among Gavin et al.’s (2011) community sample of pregnant women, those who reported experiencing high stress from an array of stressors (e.g., financial difficulties, employment problems, interpersonal conflicts, and sexual/physical/emotional abuse) were over three times as likely to report suicidal ideation during the prenatal period. This finding held after controlling for depression. In Tabb et al.’s (2013) study, 14 non-hospitalized pregnant women who were receiving prenatal care at a hospital-based maternity clinic completed qualitative interviews about their experiences of suicidal ideation. Nearly half (42.8%) of the women reported high levels of perceived stress. Thematic network analysis of the interviews revealed that suicidal ideation was associated with coping with multiple stressors.

Despite how common the experience of suicidal ideation is for pregnant women, surprisingly little research has examined its effects on children’s outcomes (see Gelaye et al., 2016, for a review). Instead, studies have tended to focus on how parental death by suicide or parental suicide attempt impacts children’s development. It cannot be expected, however, that parental suicidal ideation is linked to similar outcomes. Of the limited work conducted so far on the effect of suicidal ideation on children’s development, most has focused on pregnancy/perinatal complications or infant physical outcomes (Gelaye et al., 2019; Zhong et al., 2019). Maternal suicidal ideation has been associated with increased risk for small gestational age of infant (Gelaye et al., 2019), pre-term labor, placental abruption and hemorrhage (Zhong et al., 2019) over and above maternal mental health difficulties. To the knowledge of these authors, no such studies have examined whether maternal suicidal ideation during pregnancy is related to emotional and behavioral outcomes later in children’s development; yet, it is likely a potent risk factor. Vulnerability to suicidal behaviors includes abnormal serotonin transmission and hypothalamic-pituitary-adrenal axis functioning (van Heeringen & Mann, 2014). In pregnant women, maternal glucocorticoids readily pass the placenta and ultimately stimulate production of fetal cortisol, programing the fetal HPA axis (Van den Bergh et al., 2020). It may be, therefore, that suicidal ideation confers dysfunctional psychobiological influences to children from gestation throughout early childhood.

Children’s risk for poorer outcomes increases as exposure to additional risk factors also increases (Evans et al., 2013; Sameroff, 1998). An abundance of research shows that maternal stress during the prenatal period is associated with greater risk for adverse child outcomes across physical, behavioral and emotional domains (Huizink et al., 2003; Moss et al., 2017; Van den Bergh et al., 2020). The experience of stress during pregnancy is common, with 70% of women reporting that they had experienced at least one stressful life event in the 12 months preceding their child’s birth (CDC, 2015).

Disaster-related stress is a severe form of stress affecting a large geographical area that has a toxic effect on an array of children’s outcomes. Our knowledge in this domain has been informed by cohorts of children born to pregnant women who experienced disasters in different regions around the world, including floods, ice storms, terrorist attacks and hurricanes. Children have been assessed across key periods of development, including infancy (e.g., Buthmann et al., 2019), childhood (e.g., McLean et al., 2020) and early adolescence (e.g., Cao-Lei et al., 2021). Most cohorts have fairly good representation of boys and girls, although racial and ethnic diversity of the study populations vary depending on where the sample was recruited. Most of the studies have employed multiple methods to obtain comprehensive assessments, including psychiatric interviews of caregivers about their children (Finik & Nomura, 2017), questionnaires about children’s behavior (e.g., Finik & Nomura, 2017; Simcock et al., 2019), standardized assessments of cognition (Laplante et al., 2008; Moss et al., 2017), physiological measures of arousal (McLean et al., 2020), and neuroimaging (e.g., Cao-Lei et al., 2021). Taken together, the literature supports a significant negative effect of prenatal exposure to disaster stress on children’s physiological reactivity (Brand et al., 2006; McLean et al., 2020), anxiety (McLean et al., 2020), sleep (Simcock et al., 2019), temperament (Zhang et al., 2018), motor development (Moss et al., 2017), cognition (Laplante et al., 2008), and behavior (Jones et al., 2019).

One purported path through which these effects occur is programing of the fetal HPA axis in response to maternal glucocortoid secretion (van den Bergh et al., 2020). Collectively, these findings are especially concerning given the increasing frequency and intensity of natural disasters (van Aalst, 2006).

As both maternal suicidal ideation and response to disaster-related stress can impact maternal physiological reactivity, which in turn may influence fetal development, it is possible that presence of both risks will amplify risk for negative behavioral and emotional outcomes for children. That is, not only will maternal suicidal ideation during pregnancy directly influence children’s clinical behaviors, but Sandy exposure will augment the impact of maternal suicidal thoughts during pregnancy on early childhood outcomes.

The current study examines whether maternal suicidal ideation during pregnancy is associated with the trajectories of children’s behaviors across early childhood, and whether prenatal maternal traumatic stress differentially accelerates the suboptimal trajectories. A racially and ethnically diverse sample of pregnant women exposed or unexposed to Superstorm Sandy in NYC were recruited into this longitudinal study of stress in pregnancy. Maternal suicidal ideation was assessed during pregnancy and children were evaluated approximately annually from 2 through 6-years-old. It was hypothesized that the presence of both maternal suicidal ideation and stress during pregnancy would have a synergistic relation to the development of children’s behavioral and emotional difficulties and contribute to suboptimal trajectories across early childhood.

Methods

Participants

The current study included mothers and their children from the longitudinal Stress In Pregnancy – SIP – study. Women were recruited from prenatal obstetrics and gynecological clinics in metropolitan New York. They were followed throughout their pregnancy in the first phase of the study. Subsequently, mothers of children born between June 2011 and July 2015 rejoined the second phase of the study at different times and completed questionnaires about their child’s development at 2, 3, 4, 5 and 6 years of age. Children who had at least 1 assessment of behavior outcomes were included, giving a subsample of 331 mother-child dyads (51% boys). Exclusion criteria for participation included HIV infection, maternal psychosis, maternal age <15 years, life-threatening maternal medical complications, and congenital or chromosomal abnormalities in the fetus (Finik & Nomura, 2017, for further details). All mothers provided written informed consent on behalf of their minor children. There was no formal assenting process as children were younger than 8 years of age at all time points. The protocol was approved by the Institutional Review Board at Queens College of the City University of New York (IRB Approval #339130-QC). Participants were reimbursed for their time and efforts.

Our of 361 participants followed during childhood, this study uses the 331 participants with at least one child behavioral assessment (clinical and composite summary scores). Thirty participants were excluded from the analysis due to missing assessment data. Data was collected for 183 children at 2 years (wave 1), 164 at 3 years (wave 2), 195 at 4 years (wave 3), 171 at 5 years (wave 4), and 144 at 6 years (wave 5). Participants had, on average (SD), 2.6 (1.4) assessments: (32% had one, 22% two, 20% had three, 15% had 4, and 19% had all).

Participants reported their demographic information during the second trimester (Table 1). Women were, on average (SD), 28.06 (6.12) years old. Just under half (n=153, 46%) of women had an Associate’s degree or higher educational qualification and 51% (n=169) were married. Approximately half of the children (n=169, 51%) were male, and overall, were a racially/ethnically diverse group: 18% (n=60) White, non-Hispanic, 22% (n=72) Black, 47% (n=157) Hispanic, 10% (n=33) Asian, and 3% (n=9) “Other” race/ethnicity.

Table 1.

Characteristics of participants by the status of prenatal exposure to maternal suicidal ideation and Superstorm Sandy (N=331)

		Overall Sample	No Prenatal Suicidal Ideation		Prenatal Suicidal Ideation
Demographic characteristics		N=331	SSa (−) (n=196)	SSa (+) (n=94)	SSa (−) (n=32)	SSa (+) (n=9)	Statistics, p-value
Child Sex	Girls, N (%)	162 (49)	91 (46)	51 (47)	15 (54)	5 (57)
	Boys, N (%)	169 (51)	105 (54)	43 (53)	17 (46)	4 (44)	X2(1)=1.77, p =.62
Birthweight (g)	Mean (SD)	3228 (590)	3181 (589)	3342 (627)	3143 (486)	3346 (365)	F(3,323)=1.65, p =.18
Gestational Age (weeks)	Mean (SD)	38.92 (2.06)	38.78 (2.17)	38.98 (2.05)	39.31 (1.42)	39.63 (1.40)	F(3,324)=.81, p =.49
Maternal Age	Mean (SD)	28.06 (6.12)	28.31 (6.26)	27.39 (5.90)	28.61 (6.06)	27.89 (5.67)	F(3,327)=.56, p =.64
Mother’s Education	Primary/ Some High School, N (%)	31 (9)	22 (11)	5 (5)	3 (9)	1 (11.1)
	High School/GED, N (%)	63 (19)	37 (19)	17 (18)	7 (22)	2 (22)
	Some College, N (%)	84 (25)	50 (26)	20 (21)	9 (28)	5 (56)
	Associate Degree, N (%)	62 (19)	36 (18)	17 (18)	8 (25)	1 (11)
	Bachelor’s Degree, N (%)	88 (27)	49 (25)	34 (36)	5 (16)	0 (0)
	Graduate Degree, N (%)	3 (1)	2 (1)	1 (1)	0 (0)	0 (0)	X2(15)=15.26, p =.43
Marital Status	Married, N (%)	169 (51)	87 (44)	57 (61)	17 (53)	5 (56)
	Common Law, N (%)	11 (3)	8 (4)	3 (3)	0 (0)	0 (0)
	Single, N (%)	149 (45)	98 (50)	33 (35)	15 (47)	3 (33)
	Divorced/Separated, N (%)	5 (2)	3 (2)	1 (1)	0 (0)	1 (11)	X2(9)=14.70, p =.10
Child’s Race	White, N (%)	60 (18)	29 (15)	25 (27)	5 (16)	1 (11)
	Black, N (%)	72 (22)	53 (27)	11 (12)	7 (22)	1 (11)
	Hispanic, N (%)	157 (47)	84 (43)	50 (53)	18 (56)	5 (56)
	Asian, N (%)	33 (10)	23 (12)	6 (6)	2 (6)	2 (22)
	Other, N (%)	9 (3)	7 (4)	2 (2)	0 (0)	0 (0)	X2(12)=20.18, p =.06
Prenatal substance use	Mean (SD)	0.27 (0.25)	0.25 (0.50)	0.38 (0.66)	0.27 (0.59)	0.22 (0.67)	F(3,327)=.48, p =.70
Maternal stress and psychological problems during pregnancyb	Low, N (%)	135 (41)	93 (47)	3 (9)	37 (40)	2 (22)
	Medium, N (%)	143 (44)	80 (40)	14 (44)	43 (47)	6 (67)
	High, N (%)	53 (16)	25 (13)	15 (45)	12 (13)	1 (11)	X2(6)=33.23, p <.001
Objective Sandy distress	Mean (SD)	2.70 (2.83)	2.39 (2.63)	3.16 (3.15)	3.41 (3.39)	2.22 (2.17)	F(3,323)=2.32, p =.08

^aSS = Superstorm Sandy

^bLatent Profile Analysis extracted the index using maternal depressive symptomatology during pregnancy (measured using 9 items of the Edinburgh Postnatal Depression Scale (Cox et al., 1987), the State- and Trait- Anxiety Inventory (STAI) (Spielberger, 1989), the Pregnancy-related anxiety questionnaire-revised (PRAQ-R), the 14-item Perceived Stress Scale (PSS-14) (Cohen & Williamson, 1988). The three-class solution (low, medium, and high) was determined to be most optimal.

Measures

Child Behaviors

The Behavior Assessment System for Children – 2^nd Edition, Parent Rating Scale (BASC2-P) (Reynolds & Kamphaus, 2004), a well-standardized, multidimensional evaluation of the behavior of young children, was used to measure clinical dimensions of behavior. Clinical dimensions included measures of Hyperactivity (i.e., overactivity), Aggression (i.e., hostile, threatening behavior), Anxiety (e.g., fear and worry), Depression (e.g., sadness, stress), Somatization (i.e., complaints of physical problems), Atypical Behaviors (i.e., “odd” behaviors), Withdrawal (i.e., avoidance of social contact), and Attention Problems (i.e., distractibility and poor concentration). Composite scales, which represent broad clusters of problematic behaviors were also calculated. These included Externalizing Problems (composite of Hyperactivity and Aggression scales), Internalizing Problems (composite of Anxiety, Depression and Somatization scales), and the Behavioral Symptom Index (BSI), which reflects the overall level of problem behavior. As such, the BSI is a composite of the Hyperactivity, Aggression, Depression, Attention Problems, Atypicality and Withdrawal scales. For all scales, we used age-normed T-scores, with a mean (SD) of 50 (10), where higher scores indicate more severe behavioral disturbance. In the normative sample, the BASC-2 PRS demonstrated its reliability, with good to excellent internal consistency and good test-retest reliability over 1 to 8 weeks, as well as validity, with strong correlations with other measures of child behavioral difficulties (Reynolds & Kamphaus, 2004).

Prenatal Sandy Exposure

Sandy status was defined as whether mothers were pregnant during Hurricane Sandy. Mothers were categorized as Exposed to Sandy (n=103, 31%) or Not Exposed control group (n=228, 69%).

Maternal Suicidal Ideation

Presence and absence of suicidal ideation was measured using Question 10 of the Edinburgh Postnatal Depression Scale (EPDS) (Cox et al., 1987), ascertained during the second trimester of the pregnancy. Participants were asked if the thought of harming themselves had occurred to them. Responses of “Yes Quite Often,” “Sometimes” and “Hardly Ever” were coded as “Yes,” while a response of “Never” was coded “No.” N=41 (12%) women endorsed experiencing suicidal ideation during their pregnancy, of whom 9 were also pregnant when Superstorm Sandy hit NYC.

Covariates

Child and Maternal Demographic Variables.

Selected demographic variables including child sex and race, maternal age, and education were determined as potential confounders a priori. Maternal age at birth was calculated based on the mother’s and child’s dates of birth. Maternal education was self-reported, which is considered the best indicator of socioeconomic status (SES) among pregnant women, as other indicators (e.g., employment, income) may introduce systematic bias if mothers chose to stay at home during pregnancy.

Maternal psychological functioning was extracted using latent profile analysis (LPA, Tein, Coxe, & Cham, 2013) with the Pregnancy-Related Anxiety Questionnaire-Revised (PRAQ-R; Huizink et al., 2004), and the 14-item Perceived Stress Scale (PSS-14) (Cohen & Williamson, 1988). Mother’s depression symptoms were measured by the Edinburgh Postnatal Depression Scale (EPDS) (Cox et al., 1987) and anxiety symptoms by the State- and Trait- Anxiety Inventory (STAI) (Spielberger, 1989). All measures show adequate reliability and validity (Cohen & Williamson, 1988; Cox et al., 1987; Huizink et al., 2004; McDowell, 2006). In the current sample, internal consistency for each of the PRAQ-R, PSS-14, EPDS, and STAI was α=.86, α=.91, α=.84, and α=.89, respectively. LPA tested 2- and 3-class solutions, which both showed good model fit with entropy of 0.8 or greater and significant Lo-Mendell-Rubin (L-M-R) test scores. Both the BIC and ABIC values decreased from the 2-class model to the 3-class model. When comparing the model fits between 2-class and 3-class models, there was a significant improvement (p=.002). Taken together, the 3-class model (low, medium, and high) was selected.

Maternal Substance Use during Pregnancy:

The absence or presence of tobacco, cannabis, and alcohol use during pregnancy were ascertained via mother’s self-report during pregnancy. The “substance use” variable reflected the number of substances used by the mother.

Objective Severity of Sandy Exposure was measured using Storm32 (King & Laplante, 2005), a 20-item questionnaire that assesses salient aspects of disaster exposure. For example: did your residence suffer damage as a result of Hurricane Sandy; did you experience a loss of personal income; did your family stay together for the duration of the storm; were you in danger as a result of downed electrical power lines; did you experience lack of potable water? Reliability and validity of Storm32 was demonstrated among 380 mother-child dyads (Buthmann et al., 2019). For the current sample, mean (SD) was 2.90 (2.98); range was 17; internal consistency was α=.90.

Statistical Analyses

Hierarchical linear modeling (HLM) was used to estimate both within-person longitudinal effects, from 2 to 6 years of age, and between-person effects (Raudenbush & Byrk, 2002) using Hierarchical Linear Modeling (HLM) software. The within-person model mapped the developmental trajectory of clinical behaviors among children at the five data points in this study. The between-person model estimated how maternal suicidal ideation and Sandy exposure in utero influenced changes in child behavior. All models in the analysis were corrected for non-normal distributions of level 2 residuals by applying the full maximum likelihood estimation with robust standard errors that incorporated the missing data imputation (Maas & Hox, 2004). Model 1 evaluated the trajectories of child behaviors. The main effects of maternal suicidal ideation and Sandy exposure in utero on child behaviors were tested in Model 2. Finally, the interaction term between maternal suicidal ideation and Sandy exposure was added in Model 3.

Model 1: Changes in Child Behaviors Over Time

The baseline model was designed to characterize trajectories of child behaviors from 2 to 6 years of age, free of a predictor and covariates. For each outcome, we tested the model for linear change and then curvilinear change, as behaviors might not display linear change, especially in early development. If a significant or marginally significant quadratic effect was not observed, the quadratic term was removed from the model. Further, tests of relative model fit were computed by comparing the deviance statistics between the quadratic and linear models. If the Chi-Square test of difference demonstrated a significant reduction in deviance scores, the quadratic model was maintained (see Supplementary Table 1). Note that random effects were included in the intercept and change coefficients (all linear and quadratic terms were retained). Age was centered at 2 years (i.e., the intercept represented the average behavior score at age 2).

Unadjusted Linear Models (a).

In the linear model, child behavior is a function of an intercept plus a linear effect for Time (i.e., assessment wave). The linear model equation is as follows:

$$\begin{array}{c}\text{Level}1\\ {\text{Behavior}}_{ij}={\beta }_{0j}+{\beta }_{1j}*\left({\text{Time}}_{ij}\right)+r_{ij}\\ \text{Level}2\\ {\beta }_{0j}={\gamma }_{00}+u_{0j}\\ {\beta }_{1j}={\gamma }_{10}+u_{1j}\end{array}$$

Unadjusted Quadratic Models (b).

The quadratic model equation is as follows:

$$\begin{array}{c}\text{Level}1\\ {\text{Behavior}}_{ij}={\beta }_{0j}+{\beta }_{1j}*\left({\text{Time}}_{ij}\right)+{\beta }_{2j}*{\left({\text{Time}}_{ij}\right)}^2+r_{ij}\\ \text{Level}2:\\ {\beta }_{0j}={\gamma }_{00}+u_{0j}\\ {\beta }_{1j}={\gamma }_{10}+u_{1j}\\ {\beta }_{2j}={\gamma }_{20}+u_{2j}\end{array}$$

Model 2: Models with a Predictor of Intercepts and Slopes – Main Effects

After choosing the linear (a) or quadratic (b) model, as appropriate, we explored whether maternal suicidal ideation and Sandy exposure explained significant variance in mean intercept or slope of child behavior simultaneously in the same models. Covariates were added in the adjusted models, including child’s sex, child’s race, maternal education, marital status, maternal age, maternal prenatal substance use (tobacco use, cannabis use, and alcohol use) during pregnancy, maternal psychological functioning (trait anxiety, depression during pregnancy, perceived stress and pregnancy-related anxiety) and objective severity of Sandy exposure. If any aspect of child behavior displayed neither linear nor quadratic change over time, predictors were added to calculate the intercept only (2-year-old behaviors).

Linear Models (a).

$$\begin{array}{c}\text{Level}1\\ {\text{Behavior}}_{ij}={\beta }_{0j}+{\beta }_{1j}*\left({\text{Time}}_{ij}\right)+r_{ij}\\ \text{Level}2\\ {\beta }_{0j}={\gamma }_{00}+{\gamma }_{01}*\left(\text{suicidal}\text{ideation}\right)+{\gamma }_{02}*\left(\text{Superstorm}\text{Sandy}\text{exposure}\right)+{\gamma }_{03}*\left(\text{child}\text{sex}\right)+{\gamma }_{04}*\left(\text{child}\text{race}\right)+{\gamma }_{05}*\left(\text{maternal}\text{age}\right)+{\gamma }_{06}*\left(\text{maternal}\text{education}\right)+{\gamma }_{07}*\left(\text{marital}\text{status}\right)+{\gamma }_{08}*\left(\text{maternal}\text{psychological}\text{functioning}\right)+{\gamma }_{09}*\left(\text{maternal}\text{substance}\text{use}\right)+{\gamma }_{010}*\left(\text{objective}\text{severity}\text{of}\text{Sandy}\text{exposure}\right)+u_{0j}\\ {\beta }_{1j}={\gamma }_{10}+{\gamma }_{11}*\left(\text{suicidal}\text{ideation}\right)+{\gamma }_{12}*\left(\text{Superstorm}\text{Sandy}\text{exposure}\right)+{\gamma }_{13}*\left(\text{child}\text{sex}\right)+{\gamma }_{14}*\left(\text{child}\text{race}\right)+{\gamma }_{15}*\left(\text{maternal}\text{age}\right)+{\gamma }_{16}*\left(\text{maternal}\text{education}\right)+{\gamma }_{17}*\left(\text{marital}\text{status}\right)+{\gamma }_{18}*\left(\text{maternal}\text{psychological}\text{functioning}\right)+{\gamma }_{19}*\left(\text{maternal}\text{substance}\text{use}\right)+{\gamma }_{110}*\left(\text{objective}\text{severity}\text{of}\text{Sandy}\text{exposure}\right)+u_{0j}\end{array}$$

Quadratic Models (b).

The quadratic model equation with covariates is as follows:

$$\begin{array}{c}\text{Level}1\\ {\text{Behavior}}_{ij}={\beta }_{0j}+{\beta }_{1j}*\left({\text{child}\text{Time}}_{ij}\right)+{\beta }_{2j}*{\left({\text{Time}}_{ij}\right)}^2+r_{ij}\\ \text{Level}2\\ {\beta }_{0j}={\gamma }_{00}+{\gamma }_{01}*\left(\text{suicidal}\text{ideation}\right)+{\gamma }_{02}*\left(\text{Superstorm}\text{Sandy}\text{exposure}\right)+{\gamma }_{03}*\left(\text{child}\text{sex}\right)+{\gamma }_{04}*\left(\text{child}\text{race}\right)+{\gamma }_{05}*\left(\text{maternal}\text{age}\right)+{\gamma }_{06}*\left(\text{maternal}\text{education}\right)+{\gamma }_{07}*\left(\text{marital}\text{status}\right)+{\gamma }_{08}*\left(\text{maternal}\text{psychological}\text{functioning}\right)+{\gamma }_{09}*\left(\text{maternal}\text{substance}\text{use}\right)+{\gamma }_{010}*\left(\text{objective}\text{severity}\text{of}\text{Sandy}\text{exposure}\right)+u_{0j}\\ {\beta }_{1j}={\gamma }_{10}+{\gamma }_{11}*\left(\text{suicidal}\text{ideation}\right)+{\gamma }_{12}*\left(\text{Superstorm}\text{Sandy}\text{exposure}\right)+{\gamma }_{13}*\left(\text{child}\text{sex}\right)+{\gamma }_{14}*\left(\text{child}\text{race}\right)+{\gamma }_{15}*\left(\text{maternal}\text{age}\right)+{\gamma }_{16}*\left(\text{maternal}\text{education}\right)+{\gamma }_{17}*\left(\text{marital}\text{status}\right)+{\gamma }_{18}*(\text{maternal}\text{psychological}\text{functioning}+{\gamma }_{19}*\left(\text{maternal}\text{substance}\text{use}\right)+{\gamma }_{110}*\left(\text{objective}\text{severity}\text{of}\text{Sandy}\text{exposure}\right)+u_{0j}\end{array}$$

Model 3: Models with a Predictor of Intercepts and Slopes – Interactions

After the main effects of suicidal ideation and Sandy exposure in utero were evaluated, the interaction between suicidal ideation and Superstorm Sandy exposure was added in Model 2.

Effect Size

Variance explained for a multilevel model is calculated based on a formula for R² specific to multilevel models (Snijders & Bosker, 2012), which represents proportional reduction in prediction error at the individual level.

$$R^2=1-\frac{{\sigma }_F^2+{\tau }_F^2}{{\sigma }_E^2+{\tau }_E^2}$$

σ² represents the level 1 random error variance and τ represents the level 2 random error variance. F represents for the full model (i.e., the model of interest) and E for the empty model with only covariates.

Variance components based on the empty model and the full model were estimated. The effect size measure related to variance explained for the overall model is computed as below.

$$f^2=\frac{R^2}{1-R^2}$$

0.02 is a small effect, 0.15 is a medium effect, and 0.35 is a large effect (Cohen, 1992).

Missing data

To compensate for missing data, common in many longitudinal cohort studies, HLM was used to yield parameter estimates for the missing time points for dependent variable data (i.e., child behaviors) at level 1 (i.e., within subject variability), but not for predictor variables at level 2 (i.e., between subject variability) (Raudenbush & Byrk, 2002). Given that there was a considerable follow-up time (from 2–6 years of age) and we have missing behavior data for at least one time point, this approach was optimal for examining differences in developmental trajectories. There is no missing data at Level 2 with the exception of the severity of Superstorm Sandy score, where there were 1% missing data. Following Schafer and Graham (2002), HLM was used to compute average estimates that reflected the uncertainty of the missing level 2 data.

Results

Demographics differences in children whose mothers had and did not have suicidal ideation

Major demographic characteristics between those who were included (N=331) and those who were excluded (N=30) in this study did not differ significantly; that is, child sex (p=.06), race (p=.16), and maternal age (p=.81), with the exception of maternal education, where those included had a higher maternal education than those excluded.

Model 1: Change in Behavior Scores Over Time Without a Predictor

We first modeled child behavior as a function of the intercept plus the linear/quadratic effect of age. Supplementary Figure 1 depicts the developmental trajectories of each of the eight clinical behaviors and each composite indicator. In the clinical domain, Somatization increased, while Atypical Behavior and Attention Problems decreased linearly with age. Hyperactivity, Aggression, Anxiety, and Depression, as well as all the composite scores, increased in the earlier age range (2–4/5 years of age) but then gradually decreased up to age 6.

Model 2: A Predictor of Intercepts and Slopes (Main effects)

After retaining better-fitting linear or quadratic models, we examined whether maternal suicidal ideation and Sandy exposure predicted mean intercepts (i.e., behaviors at 2 years) or slopes (i.e., rate/direction of change between ages 2 and 6). Only significant results related to either suicidal ideation or Sandy exposure are reported. Adjusted models included all covariates.

Figure 1 shows the trajectory of the 8 clinical domains and 3 composite scales by maternal suicidal ideation during pregnancy. Supplementary Table 2 shows the t-ratio, associated p-value and effect size for prenatal exposure to maternal suicidal ideation and Superstorm Sandy exposure.

Figure 1.

Developmental trajectories of child behaviors for the overall sample as a function of prenatal exposure to maternal suicidal ideation. Linear models are selected for 5. Somatization and 6. Atypical Behavior, 7. Withdrawal, and 8. Attention Problems.

NB: Exact t-ratio and p values can be seen in Supplementary Table 2.

For children of mothers who experienced suicidal ideation in pregnancy, Attention Problems showed a linear increase in severity from 2- to 6-years-old (t-ratio=3.75, df=324, p <.001, f²=.05). Children of mothers who were pregnant at the time that Superstorm Sandy made landfall in NY showed a significant linear slope for Anxiety (t-ratio=2.13, df=322, p=.03, f²=.01).

Model 3: A Predictor of Intercepts and Slopes (Interaction between Suicidal Ideation and Superstorm Sandy In Utero)

Figure 2 shows significantly different developmental trajectories of child behaviors by prenatal exposures to maternal suicidal ideation and its interaction with Superstorm Sandy. Table 2 presents the t-ratios for intercept, linear change, and curvilinear change attributed to prenatal exposure to maternal suicidal ideation, Superstorm Sandy exposure in utero, and the interaction of the two. The most notable findings were significant suicidal ideation x Sandy exposure interactions for Withdrawal (intercept: t-ratio=3.96, df=327, p<.001, f²=.10), Atypical behaviors (linear t-ratio = −2.51, df=324, p =.01, f²=.10) and BSI (curvilinear t-ratio = −1.62, df=322, p=.05, f²=.40). Children exposed to both risks had lower Withdrawal behaviors at age 2 years. For Atypical Behaviors, a significant linear slope for the suicidal ideation x Sandy exposure interaction showed the upward trajectory of Atypical Behaviors by suicidal ideation was amplified by the effect of Superstorm Sandy. Specifically, children whose mothers experienced suicidal ideation during pregnancy, or were exposed to both maternal suicidal ideation and Superstorm Sandy, showed an increase in severity of atypical behaviors, while those children whose mothers did not report suicidal ideation or whose mothers were solely exposed to Superstorm Sandy showed a decrease in severity of Atypicality from ages 2–6.

Figure 2.

Significantly different developmental trajectories of child behaviors by prenatal exposures to maternal suicidal ideation and its interaction with Superstorm Sandy.

Table 2.

Developmental trajectories of child behaviors from age 2 to 6 by prenatal exposures to maternal suicidal ideation and Superstorm Sandy, and the interaction of the two. (Adjusted, centered at age 2)

	Maternal Suicidal Ideation									Superstorm Sandy Exposure									Interaction
	Intercept			Linear			Curvilinear			Intercept			Linear			Curvilinear			Intercept			Linear			Curvilinear
	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f	t-ratio	p	f
Clinical Scales
HYP	1.50	.14	.02	−0.74	.45	.002	−0.11	.92	.07	0.69	.45	.01	−0.16	.87	.001	0.42	.68	.06	1.19	.23	.007	−0.77	.42	.03	0.06	.95	.07
AGG	2.12	.03	.05	−1.00	.32	.005	0.64	.52	.05	0.28	.78	.001	0.51	.61	.02	−1.59	.11	.02	−0.89	.37	.005	1.32	.19	.04	−1.59	.11	.05
ANX	1.38	.16	.03	−0.51	.61	.02	−2.33	.02	.06	−0.97	.33	.04	2.44	.01	.12	−2.14	.03	.13	0.32	.75	.005	−1.00	.32	.09	1.17	.25	.12
DEP	1.01	.32	.02	−0.10	.92	.02	−1.25	.21	.04	−0.73	.49	.007	0.43	.67	.04	0.70	.48	.09	0.01	.99	<.001	−0.27	.79	.01	0.53	.59	<.001
SOM	3.75	<.001	.18	−1.74	.08	.67	--	--	--	0.98	.32	.05	0.08	.94	.06	--	--	--	−1.32	.19	.02	0.61	.54	.02	--	--	--
ATP	−0.21	.84	<.001	2.61	.009	.84	--	--	--	0.14	.89	.11	−0.86	.39	.09	--	--	--	1.17	.24	.03	−2.51	.01	.10	--	--	--
WDL	−2.50	.01	.05	−0.54	.59	.08	--	--	--	−0.54	.59	.35	0.34	.73	.09	--	--	--	3.96	<.001	.10	−1.42	.16	.01	--	--	--
ATN	−0.06	.95	.06	2.99	.003	.87	--	--	--	0.11	.92	.04	0.56	.57	.09	--	--	--	0.51	.61	.003	−0.46	.65	.001	--	--	--
Composite Scales
EXT	2.11	.04	.04	−0.83	.41	.06	−0.18	.18	.16	0.69	.49	.03	0.31	.76	.003	−0.18	.86	.03	−0.03	.98	.005	0.34	.73	.06	−0.91	.36	.12
INT	2.12	.04	.09	1.33	.32	.15	0.85	.39	.16	−1.38	.17	.01	1.98	.05	.11	−0.40	.68	.12	−0.10	.92	.002	−0.21	.84	.03	0.13	.90	.02
BSI	1.05	.29	.02	1.10	.12	.03	−1.62	.10	.17	0.53	.60	.006	−1.02	.31	.001	1.55	.12	.15	0.90	.37	.007	0.67	.50	.01	−1.62	.05	.40

NB: Time was centered at age 2; the intercept stands for age 2.

All models were adjusted for sex and race of the child, maternal education, marital status, maternal age, prenatal maternal stress and psychological functioning, objective severity of Sandy exposure, and maternal substance use.

HYP = Hyperactivity; AGG = Aggression; ANX = Anxiety; DEP = Depression; SOM = Somatization; ATP = Atypical Behaviors; WDL = Withdrawal; ATN = Attention Problems; EXT = Externalizing Problems composite scale; INT = Internalizing Problems composite scale; and BSI = Behavioral Symptoms Index.

Degree of freedoms for each model are: HYP 326; AGG 324; ANX 323; DEP 317; SOM 325; ATP 324; WDL 327; ATN 327; EXT 325; INT 324; BSI 322

For Attention Problems, a significant linear effect of exposure to maternal suicidal ideation was observed (linear t-ratio=2.99, df=327, p=.003, f²=.87), such that Attention Problems increased in severity from 2 to 6 years of age for exposed children. There was no significant main effect of exposure to Superstorm Sandy and no maternal suicidal ideation x stress interaction.

Main effects of maternal ideation (curvilinear t-ratio=−2.33, df=323, p =.02, f²=.06) and Superstorm Sandy exposure (curvilinear t-ratio= −2.14, df=323, p=.03, f²=.13) were found in Anxiety, but not the interaction of the two. Children of mothers who experienced suicidal ideation showed an increase in Anxiety severity before peaking at 4–4.5 years and declining through age 6. Children who were exposed to Sandy in utero also saw a rise in Anxiety severity in early childhood with only a small decline in Anxiety severity seen in children whose mothers were pregnant when Superstorm Sandy landed. For these children, Anxiety levels remained relatively high at age 6.

Discussion

We investigated whether maternal suicidal ideation during pregnancy is associated with the trajectories of children’s behaviors across early childhood, and further evaluated whether prenatal maternal traumatic stress accelerates the trajectories. In this underrepresented sample of pregnant women, suicidal ideation was common. 12% of women reported experiencing suicidal thoughts, which is within the prevalence range reported in the literature (Gelaye et al., 2016).

It was hypothesized that prenatal exposure to both maternal suicidal ideation and natural disaster-related stress would have a synergistic effect on developmental outcomes. Consistent with this, children exposed to maternal suicidal ideation and traumatic stress in utero showed worsening of “atypical” behaviors across early childhood, whereas these behaviors declined in severity for children exposed to neither risk or to Superstorm Sandy only. The Atypicality subscale assesses usual behaviors, as well as a child’s disconnection from their environment (Reynolds & Kamphaus, 2004). Elevated scores on this scale have been seen in children with Autism Spectrum Disorder (Volker et al., 2010) and children with severe behavioral/emotional disturbance (Reynolds & Kamphaus, 2004). We know of one study that examined children’s temperament and emotion regulation as a function of maternal lifetime history of suicidal behavior (Sheftall et al., 2020). Exposed children were rated by their mothers as having higher negative affect, including more Sadness and Discomfort (negative emotional response to sensory stimulation), and lower Soothability (speed of recovery from peak emotion/arousal) (Sheftall et al., 2020). A temperament profile characterized by high negative affect/low emotion regulation may increase children’s risk for more severe emotional and behavioral difficulties.

Attentional development was influenced by exposure to maternal suicidal ideation. Children exposed to neither risk and only to Superstorm Sandy showed a linear decline in attention problems, which is the expected trajectory of attention difficulties in childhood (Rueda & Posner, 2013). In contrast, children exposed to both risks saw the steepest significant linear increase in attention difficulties, followed by children exposed to maternal suicidal ideation only.

Suicidal behaviors broadly, which includes suicidal ideation, have been associated with HPA dysregulation, with evidence for both hyper- and hypo-reactivity, depending on factors such as age (O’Connor et al., 2016), type and recency of suicidal behavior (e.g., O’Connor et al., 2017; 2020), and history of childhood trauma (O’Connor et al., 2020). HPA dysregulation among pregnant women who experience suicidal ideation may increase fetal psychobiological risk for attentional problems. To our knowledge this mechanism has not yet been studied directly, but HPA dysregulation in children has been associated with greater attentional difficulties. Pinto and colleagues (2016) showed that cortisol levels lowered at a faster rate for adolescent boys with high ADHD symptom scores, while Isaksson and colleagues (2012) showed that for children older than age 10 years, those with ADHD had lower cortisol levels at waking and at bedtime. Recently, a longitudinal study showed that preschoolers with lower concentration of cortisol in their hair were at greater risk of developing more severe ADHD symptoms and diagnosis at school age (Pauli-Pott et al., 2019). These findings may signify under-arousal in children, which accounts for their attentional difficulties (Isaksson et al., 2012; Pinto et al., 2016). Establishing mechanism, particularly in very young children, is an area of future research priority.

Anxiety showed a non-linear trajectory over early childhood. Single exposures (suicidal ideation or stress) were associated with more severe Anxiety than children exposed to neither insult. Children whose mothers experienced suicidal ideation in pregnancy showed a decline in Anxiety severity by age 6, however, children whose mothers were pregnant during Superstorm Sandy had only a modest decrease so that their Anxiety remained high at school age.

The Anxiety subscale of the BASC-2 reflects a child’s tendency to feel nervous, worried or fearful (Reynolds & Kamphaus, 2004). The differing trajectories of Anxiety among children in this sample by differing risk exposures adds to the growing literature on the development of Anxiety in very young children (e.g., Broeren et al., 2013; de Lijster et al., 2019; Kertz et al., 2019). Differing patterns of Anxiety trajectories have been observed depending, in part, on the sample (population-based vs. at-risk), symptom type (overall anxiety vs. specific symptom clusters) and age of the sample at baseline assessment (toddlerhood vs. preschool) (Broeren et al., 2013; de Lijster et al., 2019; Kertz et al., 2019). Utilizing data from the Generation R study, de Lijster et al. (2019) showed that children with Increasing symptom severity and Preschool-Limited patterns of Anxiety across childhood had poorer school outcomes and lower self-esteem at age 10 (de Lijster et al., 2019), suggesting that even when anxiety symptoms were limited to early childhood, risk for poorer outcomes was established. Kertz and colleagues (2019) also demonstrated the high risk posed by elevated anxiety during the preschool period. In their children, who were oversampled for depression, higher levels of anxiety during preschool years were associated with greater likelihood of an anxiety diagnosis at preschool, rates of new anxiety disorders in early childhood, and anxiety prevalence at age 11. These findings highlight the potential risk for poorer outcomes for children in our study who were exposed to either of the prenatal risks, as on average, these children showed high levels of anxiety during the preschool period. Although not tested in this study, possible underlying mechanisms include children’s HPA axis dysfunction (as measured through cortisol reactivity and diurnal cortisol levels) and altered autonomic nervous system functioning, and structural and functional brain changes to fronto-limbic regions that modulate emotional and behavioral functioning (see Lautarescu et al., 2020, and van den Bergh et al., 2020 for reviews).

Notably, the findings in this study held even after adjusting for maternal stress and psychopathology, which included prenatal depression, pregnancy-specific anxiety, general trait anxiety, and perceived stress, suggesting a unique risk posed by suicidal ideation, disaster-related stress, or their interaction, on children’s affective and behavioral development. This is consistent with findings showing that for 1/3 to 1/2 of pregnant women, suicidal behaviors occur in the absence of depression or other mental illness (Gavin et al., 2011; Zhong et al., 2016).

The findings also suggest that prenatal exposure to maternal suicidal ideation predisposes children to an increased risk for a range of behavioral and emotional difficulties, which may be amplified by additional exposure to severe stress. Although anxiety and attentional problems can be considered as specific pathological phenotypes, fear/worry and attentional difficulties are also cross-cutting symptoms that are seen over an array of psychopathological conditions (see Lupien et al., 2017, for a review). Children who were exposed to maternal suicidal ideation and stress in utero, may therefore experience a broad set of behavioral and emotional difficulties, which may increase risk for any number of psychiatric presentations as the get older.

Taken together, findings highlight the importance of health professionals explicitly screening for and documenting presence or absence of suicidal behaviors among pregnant women, even in the absence of depression. Although prevalence of suicidal behaviors (including ideation) during pregnancy differs depending on the sample (Gelaye et al., 2016), pregnant women appear to experience suicidal behaviors at least as often as those in the general population (Gavin et al., 2011) or at much higher rates (Newport et al., 2007). Of concern, recent data suggest that prevalence rates are increasing (Admon et al., 2021), with the rate more than doubling over a 6-year period (Zhong et al., 2016). High-risk groups include unmarried women, stressed women, younger women, those residing in urban areas and women of color (Gelaye et al., 2016). The latter two factors may be of particular relevance to the current sample of women, who all reside in a major urban center, and approximately half of whose children identify as Latinx. In addition, despite the risk conferred by suicidal behavior, it is often not documented in a woman’s chart (Zhong et al., 2019). Screening for suicidal ideation and behavior should become a routine component of prenatal health care, and where present, safety plans formulated.

Findings also highlight the importance of supporting pregnant women who experience a natural disaster given their vulnerability to adverse outcomes. For example, they may experience food or housing insecurity, infrastructure damage may compromise sanitation or access to health care; and pregnant women’s compromised immune systems may make them more vulnerable to infection (Harville et al., 2010). Worry about the perceived challenges may be a marked stressor, especially during a time of great vulnerability such as pregnancy. These are all stressors that may compromise fetal development and increase risk for poorer outcomes later in childhood.

In addition, results suggest it will be important for health care providers to monitor young children for the emergence of early signs of psychopathology and/or behavioral difficulties, particularly inattention, anxiety and atypical behaviors. Furthermore, if these behaviors are observed, then early intervention is likely essential. In recent years, interventions have been developed for very young children, with promising results for improving attention (e.g., Halperin et al., 2020), anxiety (e.g., Ooi et al., 2022), and behaviors associated with Autism Spectrum Disorder (e.g., Rollins & De Froy, 2022). The racially/ethnically diverse sample comprising this study emphasizes the need for prenatal and post-natal screening, monitoring and care of families to be culturally-informed and collaborative (Aquino et al., 2015; Jones et al., 2017).

The present study has several strengths. Its quasi-experimental design reduces the influence of familial confounding on results (Rice et al., 2010). Children’s behavior was repeatedly evaluated during a developmental period when problematic behaviors are emerging. The analytic strategy allowed us to examine the trajectories of the child behaviors instead of cross-sectional snap shots of one early age, optimizing the longitudinal study design. Last, the high risk, underrepresented sample was retained over many years with limited missing data.

This study also has several limitations. First, we were not able to model the influence of postnatal environmental factors. Significant effects of prenatal factors on children’s developmental outcomes may (e.g., Jones et al., 2019; O’Connor et al., 2003) or may not remain (e.g., Jones et al., 2019) after adding postnatal factors to models. It is conceivable that mothers who experienced suicidal ideation while pregnant continued to experience behavioral and/or emotional difficulties – possibly including suicidal ideation - in the postnatal environment. Women with a history of suicidal ideation and trauma are at greater risk for depression, which is associated with less sensitive caregiving and greater child psychopathology (Goodman et al., 2011). Future work should examine these complex, but testable, mediational models.

Another limitation is that mom was the single informant of children’s behavior over time, raising concerns that a mother’s affective state may influence her perception of her child’s behaviors. Recent encouraging findings by Olino and colleagues (2021) utilizing a large sample from the ABCD study and sophisticated statistical analyses suggest that mothers’ reports of their children’s emotional and behavioral difficulties were not biased by maternal psychopathology.

The present study asked women about their suicidal ideation during pregnancy, but we did not ask about other suicidal behaviors, including formulation of plans to end their lives and/or suicide attempt. Women who exhibit such suicidal behaviors may represent a particularly high-risk group. In addition, it would be useful to know how women coped with suicidal thoughts to understand protective factors or factors that may amplify risk for harm.

Last, the present study tested for a SS x SI interaction for 11 outcomes (8 clinical scales and 3 composite scores), but did make any adjustments for multiple testing. Although multiple testing increases risk for Type 1 error, the three significant findings obtained for the interaction of SS x SI were of moderate effect size. Nevertheless, it would be important to replicate current findings among other samples of children and to apply a correction for multiple comparisons.

Future research should seek to understand the mechanism by which these prenatal factors affect children’s emotional and behavioral development. Consistent with Barker’s (2004) hypothesis that exposure to environmental insults during the prenatal period sets up the fetus for greater risk of disease later in life, HPA axis dysregulation may be a potential target for investigation. Future work may also examine which specific stressors related to Sandy exposure were most harmful for children’s outcomes (e.g., homelessness, bereavement).

In summary, the current study showed that in utero exposure to maternal suicidal ideation was associated with greater anxiety, attention and atypical behaviors in early childhood, and how these trajectories were accelerated by natural disaster related stress during pregnancy. The study highlights the significant needs of pregnant mothers and those of their children in early development. In light of the increasing number of natural disasters each year as well as increasing rates of maternal suicidal ideation during pregnancy, this study’s findings are a call for action to support women and their children during these crucial periods of development.

Availability of data and material:

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request