Introduction
Prenatal drug exposure remains a daunting public health problem that can affect child development and later educational and economic opportunities. The 2021 National Survey on Drug Use and Health (Substance Abuse and Mental Health Services Administration [SAMHSA], 2023) estimated that 7.7% of pregnant women used illicit drugs, with 0.2% using cocaine during the past month. When illicit drug use was combined with prior month's use of tobacco or alcohol, the percentage was an alarming 19.6%, a rate that raises concerns about adverse outcomes in offspring across developmental stages.
Prenatal cocaine exposure (PCE) crosses the placental fetal blood–brain barrier and creates a cascade of teratologic effects via direct and indirect pathways (Mayes, Reference Mayes2002; Ross, Graham, Money, & Stanwood, Reference Ross, Graham, Money and Stanwood2015). These adverse effects on developmental outcomes have been demonstrated using brain imaging (Dow-Edwards et al., Reference Dow-Edwards, Benveniste, Behnke, Bandstra, Singer, Hurd and Stanford2006) and measures of cognition across human and non-humans (Martin, Graham, McCarthy, Bhide, & Stanwood, Reference Martin, Graham, McCarthy, Bhide and Stanwood2016). Cocaine's direct effect on the monoamine-rich frontal areas of the brain (McCarthy, Kabir, Bhide, & Kosofsky, Reference McCarthy, Kabir, Bhide, Kosofsky, Diana, Chiara and Spano2014; Thompson, Levitt, & Stanwood, Reference Thompson, Levitt and Stanwood2009) has led researchers to investigate evidence of disruptions of self-regulation and behavioral adjustment.
PCE behavioral outcome studies
The association between PCE and self-regulatory behavior has been reported consistently among school-age children (for reviews, Ackerman, Riggins, & Black, Reference Ackerman, Riggins and Black2010; Lambert & Bauer, Reference Lambert and Bauer2012) and adolescents (for reviews, Buckingham-Howes, Berger, Scaletti, & Black, Reference Buckingham-Howes, Berger, Scaletti and Black2013; Min, Minnes, Yoon, & Singer, Reference Min, Minnes, Yoon, Singer and Preedy2017b). Birth-cohort prospective studies have reported the relationship of PCE with child-reported oppositional defiant disorder and attention-deficit/hyperactivity disorder at 6 years of age (Linares et al., Reference Linares, Singer, Kirchner, Short, Min, Hussey and Minnes2006), caregiver-reported aggressive behavior at 9 years (McLaughlin et al., Reference McLaughlin, Minnes, Singer, Min, Short, Scott and Satayathum2011), child-reported depressive symptoms, caregiver-rated withdrawn behaviors, and teacher-rated anxious/depressed behaviors at 10 years (Richardson, Goldschmidt, Larkby, & Day, Reference Richardson, Goldschmidt, Larkby and Day2013), teacher- and caregiver-rated externalizing (i.e. aggressive and disruptive) behaviors at age 7, 9, 11, and 13 years (Bada et al., Reference Bada, Bann, Bauer, Shankaran, Lester, LaGasse and Tan2011), adolescent-reported externalizing behaviors at 12 and 15 years (Min et al., Reference Min, Minnes, Lang, Weishampel, Short, Yoon and Singer2014a; Min, Minnes, Yoon, Short, & Singer, Reference Min, Minnes, Yoon, Short and Singer2014b), and oppositional defiance disorder at age 17 (Kim et al., Reference Kim, Minnes, Min, Kim, Lang, Weishampel and Singer2022). However, these studies typically examined externalizing and internalizing (i.e. depressive, anxious, withdrawn) behaviors separately, ignoring their well-documented co-occurrence (Bartels et al., Reference Bartels, Hendriks, Mauri, Krapohl, Whipp, Bolhuis and Boomsma2018; Vergunst et al., Reference Vergunst, Commisso, Geoffroy, Temcheff, Poirier, Park and Orri2023). Only one study to date has assessed behavioral outcomes through emerging adulthood (EA), reporting increased emotion regulation difficulties and conduct disorder at age 21 among offspring with PCE (Richardson, De Genna, Goldschmidt, Larkby, & Donovan, Reference Richardson, De Genna, Goldschmidt, Larkby and Donovan2019). Studies examining specific environmental pathways by which PCE confers greater risk for behavioral problems across adolescence and EA have also been lacking.
Environmental pathways linking PCE and behavioral adjustment in EA
In this study, we focused on three environmental pathways as mediating mechanisms of PCE that are likely to shape behavioral adjustment across adolescence and EA: (poor) family functioning, non-kinship care, and maltreatment. These pathways represent multiple underlying dimensions of adversity including deprivation, threat, and unpredictability (Ellis, Sheridan, Belsky, & McLaughlin, Reference Ellis, Sheridan, Belsky and McLaughlin2022), which are prevalent in families where caregivers misuse substances.
Family functioning comprises multiple interdependent dimensions, such as emotional bonding among family members (i.e. attachment), problem-solving and communication abilities (i.e. conflict resolution strategies), and adherence to family rules and cohesion (i.e. parental monitoring) (El-Sheikh & Buckhalt, Reference El-Sheikh and Buckhalt2003; Izzo, Baiocco, & Pistella, Reference Izzo, Baiocco and Pistella2022). Lower quality attachment, conflict resolution, and parental monitoring may collectively represent a depriving caregiving environment. PCE is a strong antecedent to caregivers’ postnatal substance use (Elkington, Bauermeister, & Zimmerman, Reference Elkington, Bauermeister and Zimmerman2011; Singer et al., Reference Singer, Min, Minnes, Short, Lewis, Lang and Wu2018), compromising their ability to promote a child's healthy behavioral adjustment. We previously found that adolescents with PCE reported lower parental attachment and higher family conflict compared to those without PCE (Min et al., Reference Min, Minnes, Yoon, Short and Singer2014b), making family functioning an important aspect of early development to evaluate in the context of PCE.
Non-kinship care, representing an aspect of environmental unpredictability, may be another critical pathway, as prenatal drug exposure is one of the most common precipitants of non-kinship caregiving placements for infants (Tung, Christian-Brandt, Langley, & Waterman, Reference Tung, Christian-Brandt, Langley and Waterman2020). In our cohort with PCE, foster/adoptive care was associated with a better-quality home environment (Singer et al., Reference Singer, Minnes, Short, Arendt, Farkas, Lewis and Kirchner2004, Reference Singer, Nelson, Short, Min, Lewis, Russ and Minnes2008) and better language and adaptive functioning (Lewis et al., Reference Lewis, Minnes, Short, Weishampel, Satayathum, Min and Singer2011; Singer, Powers, Kim, Minnes, & Min, Reference Singer, Powers, Kim, Minnes and Min2023). However, the protective effects of non-kinship foster/adoptive placement were not generalized to behavioral outcomes, with non-kinship caregivers reporting more behavioral problems across developmental stages (Bada et al., Reference Bada, Langer, Twomey, Bursi, Lagasse, Bauer and Maza2008; Linares et al., Reference Linares, Singer, Kirchner, Short, Min, Hussey and Minnes2006; McLaughlin et al., Reference McLaughlin, Minnes, Singer, Min, Short, Scott and Satayathum2011) or no difference via adolescents’ self-report (Min et al., Reference Min, Minnes, Lang, Weishampel, Short, Yoon and Singer2014a, Reference Min, Minnes, Yoon, Short and Singer2014b).
Childhood maltreatment is also a critical pathway in behavioral adjustment (Godinet, Li, & Berg, Reference Godinet, Li and Berg2014; Jonson-Reid, Kohl, & Drake, Reference Jonson-Reid, Kohl and Drake2012) and represents salient threat and distinct environmental adversity. Adolescents with PCE reported higher rates of lifetime maltreatment, compared to a cohort of similar socio-demographic status without PCE (Kim et al., Reference Kim, Minnes, Min, Kim, Lang, Weishampel and Singer2022; Singer et al., Reference Singer, Powers, Kim, Minnes and Min2023). Childhood maltreatment was related to more negative coping behaviors to stress more strongly in adolescents with PCE than in adolescents without PCE in this cohort (Min, Minnes, Kim, Yoon, & Singer, Reference Min, Minnes, Kim, Yoon and Singer2017a).
Current study
This study extends previous findings of PCE into EA by examining possible pathways in the longitudinal associations of PCE with self-reported behavioral adjustment from early adolescence to EA. Family functioning, non-kinship care, and maltreatment were specified as potential mediators of PCE. Given the established correlation between internalizing and externalizing problems and possible bidirectional relationships over time, we utilized a random-intercept cross-lagged panel model (RI-CLPM) to understand the nature of the co-development, controlling for confounders. Confounders included prenatal and post-natal environmental conditions correlated with PCE and shown to affect behavioral adjustment, such as greater exposure to alcohol, tobacco, and marijuana (Min et al., Reference Min, Albert, Lorincz-Comi, Minnes, Lester, Momotaz and Singer2021a, Reference Min, Minnes, Momotaz, Singer, Wasden and Bearer2021b; Richardson et al., Reference Richardson, De Genna, Goldschmidt, Larkby and Donovan2019; Singer et al., Reference Singer, Min, Momotaz, Powers, Minnes and Bearer2021), a caregiver's ongoing substance use (Elkington et al., Reference Elkington, Bauermeister and Zimmerman2011; Singer et al., Reference Singer, Min, Minnes, Short, Lewis, Lang and Wu2018) and psychological distress (Min et al., Reference Min, Minnes, Park, Ridenour, Kim, Yoon and Singer2018; Minnes et al., Reference Minnes, Singer, Kirchner, Satayathum, Short, Min and Mack2008), poorer quality of home environment (Lewis et al., Reference Lewis, Minnes, Short, Min, Wu, Lang and Singer2013), violence exposure (Delaney-Black et al., Reference Delaney-Black, Covington, Ondersma, Nordstrom-Klee, Templin, Ager and Sokol2002; Kobulsky, Minnes, Min, & Singer, Reference Kobulsky, Minnes, Min and Singer2016), and limited resources and support from school and neighborhood (Min, Minnes, Kim, Kim, & Singer, Reference Min, Minnes, Kim, Kim and Singer2023; Tucker, Pollard, de la Haye, Kennedy, & Green, Reference Tucker, Pollard, de la Haye, Kennedy and Green2013). We also assessed sex and IQ as covariates given the sex difference in behavioral problems (Kim et al., Reference Kim, Minnes, Min, Kim, Lang, Weishampel and Singer2022; Minnes et al., Reference Minnes, Singer, Kirchner, Short, Lewis, Satayathum and Queh2010) and a relationship between IQ and behavior problems (Pajer et al., Reference Pajer, Chung, Leininger, Wang, Gardner and Yeates2008; Van Rest et al., Reference Van Rest, Van Nieuwenhuijzen, Kupersmidt, Vriens, Schuengel and Matthys2020). We hypothesized that poorer family functioning, higher likelihood of non-kinship care, and/or maltreatment would mediate the relationship between PCE and greater internalizing and externalizing problems across adolescence and EA.
Methods
Sample and procedure
Participating offspring/emerging adults (N = 372) were recruited with their biologic mothers at birth (September 1994 to June 1996) from an urban hospital in the Midwest USA for a study of the developmental effects of PCE (Minnes et al., Reference Minnes, Min, Kim, Francis, Lang, Wu and Singer2017; Singer et al., Reference Singer, Minnes, Short, Arendt, Farkas, Lewis and Kirchner2004). Immediately before/after childbirth, drug toxicology screenings were administered to 647 pregnant women at high risk for drug use due to lack of prenatal care, self-admitted drug use, behaviors suggesting intoxication, or a previous involvement with the Department of Children and Family Services. Of the screened, 155 refused to participate and 23 did not come to the enrollment visit; 54 were excluded due to maternal HIV-positive status, chronic medical or psychiatric illness, or low IQ, infants with Down syndrome, or fetal alcohol syndrome. Enrolled were 415 newborns and their mothers.
Maternal urine samples (n = 405) were collected in hospital within 24 h before/after delivery; infant meconium (n = 341) and urine samples (n = 24) were gathered shortly after infant birth. Biologic samples were analyzed for cocaine and other drug metabolites, including amphetamines, benzodiazepines, cannabinoids, cocaethylene, and opiates. Positive screens of maternal and infant urine, infant meconium, or maternal self-report determined 218 newborns as PCE. About 62% (136/218) of newborns with PCE were determined by both biological and maternal self-report measures, with 10% (n = 22) and 28% (n = 60) determined by either biologic assays or maternal self-report (Arendt, Singer, Minnes, & Salvator, Reference Arendt, Singer, Minnes and Salvator1999). Infants negative on all PCE indicators were classified as non-cocaine exposed (NCE). Both PCE and NCE groups were exposed to alcohol, tobacco, and marijuana. Examiners blind to substance exposure status conducted follow-up assessments at 1, 6, and 12 months and 2, 4, 6, 9, 10, 11, 12, 15, 17, and 21 years postpartum at the university-based research lab.
From the enrollment, 12 (9 PCE, 3 NCE) children died by the 12-year assessment from various medical conditions (e.g. sudden infant death syndrome, respiratory distress, pneumonia, etc.). We utilized data from 372 (190 PCE, 182 NCE) offspring with outcome data at any of three time points at ages 12 (n = 352), 15 (n = 358), or 21 (n = 299), representing 92% retention of the 403 living participants at age 12. Among 372 participants, 279 (75%) had all three assessments, 79 (21%) with two assessments, and 14 (3.76%) with one assessment. Compared to 372 participants, 31 non-participants (19 drop-out, 11 lost contact, 1 low intellectual function [IQ < 50]) were more likely to be White and prenatally exposed to tobacco. No other differences were observed. The Institutional Review Board of the participating hospital approved the study. Parental written informed consent was obtained at each visit prior to offspring age 18 years, with offspring assent beginning at age 9 and consent at age 21. With a Certificate of Confidentiality (DA-98-91) from the US Department of Health and Human Services, this study was exempt from the release of drug-related information. Monetary compensation, lunch, and/or transportation costs were provided to all participants.
Measures
Prenatal cocaine and other substance exposures
At the newborn visit, birth mothers reported amount and frequency of drug use for the month prior to and for each trimester of pregnancy using a Timeline Follow Back method (Singer et al., Reference Singer, Salvator, Arendt, Minnes, Farkas and Kliegman2002). For cocaine use, the number of rocks consumed and the amount of money expended on cocaine per day were recalled and computed to a standard unit of cocaine, equivalent to $20 worth of crack cocaine. The number of drinks of beer, wine, or hard liquor per week, with each drink equivalent to 0.5 oz. of absolute alcohol, was recorded. The number of tobacco cigarettes and marijuana joints smoked per day were collected. For each drug, frequency of use was rated on a Likert-type scale (0 = not at all to 7 = daily use) to indicate the average number of days per week of drug use, except for cigarettes computed as the number smoked per day. An average use across the pregnancy was calculated based on the frequency multiplied by the amount used per day. To assess recent (prior 30-day) postpartum, caregiver drug use, current caregivers updated the drug assessment at each follow-up visit.
Internalizing and externalizing behaviors
Internalizing and externalizing behaviors were assessed using the Youth Self-Report (YSR; Achenbach & Rescorla, Reference Achenbach and Rescorla2001) at ages 12 and 15 and the Adult Self-Report (ASR; Achenbach & Rescorla, Reference Achenbach and Rescorla2003) at age 21. The YSR and ASR yield two broadband scale scores: internalizing (α = 0.86–0.90) comprise items on anxious/depressed, withdrawn/depressed, and somatic problems; externalizing (α = 0.87–0.90) comprise items on aggressive and delinquent behaviors (and intrusive behaviors only in the ASR). Standardized T scores were used, with higher scores indicating greater symptoms.
Mediators
Parental attachment, parental monitoring, and family conflict were all assessed at age 12 using the Assessment of Liability and Exposure to Substance Use and Antisocial Behavior Scale (ALEXSA; Ridenour, Clark, & Cottler, Reference Ridenour, Clark and Cottler2009), an illustration-based, audio/computer-assisted self-report for children ages 9–12. The parental attachment scale (α = 0.80) assesses relational bonding with parents. The parental monitoring scale (α = 0.74) assesses whether parent(s) usually know(s) the youth’ activities, friendship, and whereabouts. For both parental attachment and monitoring scales (five items on a 4-point Likert scale), the average scores were used, with higher scores indicating greater levels of respective constructs (potential ranges = 0–3). The family conflict index (10 items; 1 = yes) assesses family conflict tactics (e.g. yelling, throwing objects) used to resolve family dispute. The total count was reverse coded to represent lower scores to be higher levels of family conflict. The family functioning latent variable (α = 0.65) was specified with the three indicators. Childhood maltreatment was self-reported at age 17 retrospectively using the Maltreatment module of Juvenile Victimization Questionnaire – Adult Retrospective (JVQ; Hamby, Finkelhor, Ormrod, & Turner, Reference Hamby, Finkelhor, Ormrod and Turner2004). Any incidence of maltreatment, including physical abuse, psychological/emotional abuse, neglect, custodial interference/family abduction, was coded 1 (yes). Child placement was updated at each follow-up visit (1 = ever placed with non-kinship adoptive or foster care parents by age 12).
Potential confounders
Infant (sex, race, gestational age, birth weight, height, head circumference, Hobel scores) and maternal characteristics (age at delivery, marital status, education) were obtained from hospital birth records. Socioeconomic status (SES) was assessed using the Hollingshead Two-Factor Index (Hollingshead, Reference Hollingshead1971). Birth mothers’ receptive vocabulary was assessed via the Peabody Picture Vocabulary Test-Revised (Dunn & Dunn, Reference Dunn and Dunn1981) and updated at subsequent assessments via its third edition (Dunn & Dunn, Reference Dunn and Dunn1997). Maternal psychological distress was assessed using the Global Severity Index (α = 0.95), a summary scale of the Brief Symptom Inventory (Derogatis, Reference Derogatis1992). Concurrent data on current caregivers’ vocabulary and psychological distress were updated at each visit.
Offspring Full-Scale IQ was assessed at age 11 using the Wechsler Intelligence Scales for Children-Fourth Edition (Wechsler, Reference Wechsler2003). The quality of caregiving environment was assessed at age 12 via the caregiver-reported Home Observation of the Environment-Early Adolescent (α = 0.83; Caldwell & Bradley, Reference Caldwell and Bradley2003). Lifetime frequency of violence exposure (α = 0.75) was self-reported at age 12 using the ALEXSA. Ecological resources and support were assessed at age 12 using the External Assets subscale (α = 0.93) of the caregiver-reported Developmental Assets Profile (Scales & Leffert, Reference Scales and Leffert2004; Search Institute, 2005). Receipt of free school lunch (1 = yes) was recorded at age 15, and completion of high school (or equivalent) was self-reported at age 21.
Data analyses
To examine mediation of PCE on externalizing and internalizing behaviors at ages 12, 15, and 21, an extended RI-CLPM (Hamaker, Kuiper, & Grasman, Reference Hamaker, Kuiper and Grasman2015) was used using Mplus 8.3. The RI-CLPM was extended with a path model with the three potential mediators linked to the repeatedly measured outcomes. This model allows for causal mediation analysis, utilizing the RI-CLPM structure to examine interconnections between externalizing and internalizing behaviors, while distinguishing within (time-varying) and between (time-invariant, trait-like) individual effects via random effects. Variables correlated (p < 0.15) with the outcomes for any time point were included in the model as covariates (prenatal exposure to alcohol, tobacco, or marijuana, offspring sex, IQ, HOME score, violence exposure, and external assets).
Maximum likelihood, allowing use of all available endogenous data for cases with values for all exogenous variables, was used to fit the models. Coefficients were tested using two-sided z-tests. Model fit was evaluated using the χ2 goodness-of-fit test, as well as the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR) indices. Values ⩾0.95 for CFI and TLI, ⩽0.06 for RMSEA, and ⩽0.08 for SRMR indicate a good model fit (Hu & Bentler, Reference Hu and Bentler1999). Standardized estimates of coefficients were computed and interpreted.
Of the 372-participating offspring, missing covariates were found in 7% of the sample (n = 27), and the RI-CLPM analysis was based on 345 individuals with complete data on the included covariates. The analysis assumed missing completely at random for individuals and missing at random for outcomes. As the missing percentages were small, plausible violations of these assumptions were not expected to substantially affect results (Dong & Peng, Reference Dong and Peng2013). When the analysis described above was re-done using multiple imputation as a sensitivity analysis, employing a Markov Chain Monte Carlo algorithm and variance covariance model to obtain 30 completed data sets via the Bayesian estimation approach, no substantial differences were noted in all path coefficients and indirect effects using Rubin (Reference Rubin1987)'s approach. Thus, we presented findings based on 345 individuals with complete data on the included covariates.
Indirect effects of PCE were estimated using the products of coefficients approach and tested using z-tests based on delta method estimates of standard errors (VanderWeele, Reference VanderWeele2015). Proportions mediated by specific indirect paths were calculated to illustrate the relative strength of each mediator on behavioral outcomes at different ages. Statistical significance was set at p < 0.05.
Results
Sample characteristics
The birth mothers of infants with PCE were older, less educated, less likely to be married, had greater psychological distress and lower vocabulary scores, and reported greater use of tobacco, alcohol, and marijuana over their pregnancy than mothers of infants with NCE (Table 1). No difference was found in caregiver characteristics at age 12, except that caregivers of children with PCE smoked more cigarettes in the past 30 days than caregivers of NCE offspring.
Maternal and caregiver characteristics by PCE status
BSI, Brief Symptom Inventory; HOME, Home Observation for Measurement of the Environment; NCE, non-cocaine exposed; PCE, prenatal cocaine exposed; PPVT, Peabody Picture Vocabulary Test.
a No caregivers reported cocaine use in the past 30 days.
Compared to those with NCE, infants with PCE had shorter gestational ages and lower birth weight, length, and head circumference (Table 2). The PCE group reported lower parental attachment and higher family conflict; were more likely to be placed in non-kinship foster or adoptive care by age 12 and to report maltreatment. The PCE group reported greater externalizing behaviors at ages 12 and 15 than the NCE group, but no difference at age 21. No group differences were found in internalizing behaviors across all three assessments. Those with PCE were less likely to complete high school than the NCE group. Table 3 summarizes bivariate correlations among key observed variables included in the extended RI-CLPM.
Offspring characteristics by PCE status
ASR, Adult Self-Report; DAP, Developmental Asset Profile; NCE, non-cocaine exposed; PCE, prenatal cocaine exposed; WISC-IV, Wechsler Intelligence Scales for Children-Fourth Edition; YSR, Youth Self-Report.
a Adjusted for gestational age.
b Reverse coded; higher score indicates lower levels of family conflict.
c Assessed with a self-report, referring to either graduation from high school or passing the General Educational Development test.
Correlations among key observed variables in the extended random-intercept cross-lagged panel model
Note. Bold indicates significance at p < 0.05; DAP, Developmental Asset Profile; HOME, Home Observation for Measurement of the Environment; PCE, prenatal cocaine exposure; Pre. alc. exp., prenatal alcohol exposure; Pre. tob. exp., prenatal tobacco exposure; Pre. mar. exp., prenatal marijuana exposure.
a Log-transformed.
b Reverse coded.
c By age 12.
Model estimation
Figure 1 presents the final model produced from iterative model-fitting procedures. The initial, saturated model included (1) paths from PCE to all three mediators and the repeated measures of externalizing and internalizing behaviors at three ages; (2) paths from the covariates to all three mediators and the repeated measures of externalizing and internalizing behaviors; (3) correlations between the three mediators; (4) the lag-one autoregressive and cross-lagged paths; (5) correlations between internalizing and externalizing behaviors within the same assessment point; and (6) correlations between all exogenous (i.e. PCE and all covariates) variables. This initial model produced an acceptable fit, χ2 (47) = 130.94, p < 0.001, CFI = 0.931, TLI = 0.775, RMSEA = 0.072 (90% CI 0.057–0.087), SRMR = 0.059. To achieve a more parsimonious model, path coefficients with significance level of p ⩾ 0.15 were set to zero (Hosmer, Lemeshow, & Sturdivant, Reference Hosmer, Lemeshow and Sturdivant2013), χ2 (102) = 180.18, p < 0.001, CFI = 0.936, TLI = 0.910, RMSEA = 0.047 (90% CI 0.036–0.058), SRMR = 0.068, yielding insignificant difference in model fit, Δχ2 (55) = 49.24, p = 0.69. This reduced model was accepted as the final model (Fig. 1).
Extended random-intercept cross-lagged panel model of the association between prenatal cocaine exposure and externalizing and internalizing behaviors from early adolescence to emerging adulthood. χ2 (102) = 180.18, p < 0.001, CFI = 0.936, TLI = 0.910, RMSEA = 0.047 (90% CI 0.036–0.058), SRMR = 0.068. Rectangles indicate observed variables, and ovals represent latent constructs. All path coefficients are standardized; single-arrowed lines represent path coefficients, and double-arrowed lines represent correlations. Solid lines indicate statistically significant (p ⩽ 0.05) paths, and dotted lines indicate non-significant (p > 0.05) paths. Only significant path coefficients are shown; covariates and errors associated with the measurement model are not displayed for parsimony. Att, parental attachment; Ext, externalizing behaviors; Family, family functioning; Int, internalizing behaviors; Nkin, ever in non-kinship care by age 12; PCE, prenatal cocaine exposure; rFcon, family conflict, reverse coded; Pmon, parental monitoring; RExt, latent stability/trait of externalizing behaviors over-time; RInt, latent stability/trait of internalizing behaviors over-time. Estimates are adjusted for prenatal exposure to alcohol (on Nkin and Int at 21), tobacco (on Nkin and Int at 21), and marijuana (on Int at 12), biological sex (on Int at 12 and 15 and Ext at 15), violence exposure (on Int and Ext at 12), IQ (on Int at 12), and external assets (on Ext at 12, 15, and 21). *p ⩽ 0.05, **p ⩽ 0.01, ***p ⩽ 0.001.
Adjusting for relevant covariates, PCE was related to poorer family functioning at age 12 (β = −0.16, s.e. = 0.065, p = 0.01), which in turn was related to greater internalizing (β = −0.29, s.e. = 0.073, p < 0.001; indirect effect of PCE, β = 0.83, s.e. = 0.39, p = 0.036) and externalizing (β = −0.58, s.e. = 0.063, p < 0.001; indirect effect of PCE, β = 1.58, s.e. = 0.65, p = 0.016) behaviors at age 12. Greater externalizing behaviors at age 12 were related to greater externalizing behaviors at age 15 (β = 0.32, s.e. = 0.055, p < 0.001; indirect effect of PCE via family functioning, β = 0.51, s.e. = 0.23, p = 0.028). Although greater internalizing behaviors at age 12 were related to greater internalizing behaviors at age 15 (β = 0.31, s.e. = 0.069, p < 0.001), no significant indirect effect of PCE was observed via family functioning on internalizing behaviors at age 15 (β = 0.24, s.e. = 0.13, p = 0.062). PCE was also related to a higher likelihood of non-kinship placement (β = 0.28, s.e. = 0.053, p < 0.001). Non-kinship placement was related to more externalizing behaviors at age 12 (β = 0.13, s.e. = 0.049, p = 0.007; indirect effect of PCE, β = 0.63, s.e. = 0.26, p = 0.016), which contributed to greater externalizing behaviors at age 15 (indirect effect of PCE, β = 0.20, s.e. = 0.09, p = 0.027). PCE was related to a greater exposure to maltreatment (β = 0.15, s.e. = 0.054, p = 0.005), which in turn was related to greater internalizing behaviors at ages 15 (β = 0.23, s.e. = 0.056, p < 0.001; indirect effect of PCE, β = 0.64, s.e. = 0.27, p = 0.019) and 21 (β = 0.33, s.e. = 0.056, p < 0.001; indirect effect of PCE, β = 1.39, s.e. = 0.53, p = 0.011) and externalizing behavior at ages 15 (β = 0.20, s.e. = 0.055, p < 0.001; indirect effect of PCE, β = 0.50, s.e. = 0.23, p = 0.029) and 21 (β = 0.36, s.e. = 0.056, p < 0.001; indirect effect of PCE, β = 1.11, s.e. = 0.44, p = 0.012).
Direct associations of PCE with internalizing and externalizing behaviors were not observed, nor auto-regressive paths of both externalizing and internalizing behaviors from ages 15 to 21. Although internalizing and externalizing behaviors were associated within offspring (i.e. correlation within time) across all assessments (ranged r = 0.39 at age 12 to r = 0.82 at age 21; Fig. 1), no cross-lagged paths between internalizing and externalizing behaviors were significant, with time-invariant, trait-like characteristics shared by both internalizing (RInt) and externalizing (RExt) domains being partialled out. These two latent trait factors were substantially correlated (r = 0.52). Table 4 summarizes significant indirect effects of PCE via these three mediators at age 12, 15, and 21, along with mediated proportions (%) via each mediator.
Indirect effects of PCE on behavioral adjustments from ages 12 to 21
PCE, prenatal cocaine exposure.
* Calculated the specific indirect effect divided by the sum of all indirect effects. For example, for 12-year externalizing behaviors, family functioning mediated 71% [ = (1.58/(1.58 + 0.63))×100] of the indirect effect of PCE, while non-kinship placement mediated 29% [ = (0.63/(1.58 + 0.63))×100] of the indirect effect of PCE.
Regarding covariates, prenatal exposure to alcohol (β = 0.14, s.e. = 0.051, p = 0.008) and tobacco (β = 0.1, s.e. = 0.053, p = 0.045) were related to greater likelihood of non-kinship placement. Greater violence exposure was related to both internalizing (β = 0.20, s.e. = 0.057, p < 0.001) and externalizing behaviors (β = 0.24, s.e. = 0.054, p < 0.001) at age 12. Higher IQ was related to fewer internalizing behaviors at age 12 (β = −0.21, s.e. = 0.048, p < 0.001). Greater external assets were related to fewer externalizing behaviors across all three assessments (β = −0.18, s.e. = 0.048, p < 0.001 at age 12; β = −0.12, s.e. = 0.049, p = 0.013 at age 15; β = −0.13, s.e. = 0.038, p = 0.001). Males reported higher internalizing behaviors at age 12 (β = 0.12, s.e. = 0.049, p = 0.013), yet fewer externalizing behaviors at age 15 (β = −0.16, s.e. = 0.051, p = 0.002) than females.
Discussion
This study demonstrated that PCE-related behavioral problems persist into EA via multiple environmental pathways. PCE was related to poorer family functioning, a greater exposure to maltreatment, and a greater likelihood of placement in non-kinship care in childhood, each of which contributed to elevated externalizing and internalizing behaviors across adolescence and EA, supporting the long-lasting, unfolding influence of PCE channeled through environmental factors (Mayes, Reference Mayes2002). The family environment and non-kinship placement appeared to be more salient at ages 12 and 15, whereas maltreatment was salient at ages 15 and 21. Although these specific pathways have been previously reported to contribute to behavioral adjustment, they have seldom been examined simultaneously. Given the clustered nature of multiple adversities present in offspring with PCE, our simultaneous analysis of multiple environmental pathways suggests specific points of intervention that may better support healthy development.
Our study observed four noteworthy findings in the longitudinal relationship between internalizing and externalizing behaviors. First, no cross-lagged relationship was observed, suggesting that internalizing and externalizing problems are not causally related to each other. Our study supports the shared-risk hypothesis that the co-development of internalizing and externalizing problems is due to shared risk factors related to both externalizing and internalizing behaviors (Angold, Costello, & Erkanli, Reference Angold, Costello and Erkanli1999). PCE, mediated by family functioning, non-kinship care, and maltreatment, is a common correlate to explain such co-development, at least partially. Second, the time-invariant component of internalizing and externalizing problems was substantially correlated, indicating that co-development of these behavioral problems is partly attributable to stable, trait-like characteristics related to both externalizing and internalizing problems. Previous studies suggested that deficits in executive function in early childhood could be a common antecedent associated with greater problems in both internalizing and externalizing behaviors (e.g. Hughes & Ensor, Reference Hughes and Ensor2011; Oh, Greenberg, Willoughby, & The Family Life Project Key Investigators Reference Oh, Greenberg and Willoughby2020). Given the poorer executive function noted in children with PCE (Minnes et al., Reference Minnes, Singer, Min, Lang, Ben-Harush, Short and Wu2014; Noland et al., Reference Noland, Singer, Arendt, Minnes, Short and Bearer2003), future studies examining interrelationships among PCE, executive function, and behavioral adjustment, along with the three environmental pathways identified in this study, will enhance our understanding of PCE-associated problems across developmental stages. Third, there were no significant auto-regressive paths in both internalizing and externalizing behaviors from 15 to 21 years, demonstrating a low degree of stability. Although this low stability could be due to the long assessment interval, it may also be attributed to the nature of the observed period, as ages 15–21 represent a volatile developmental period of opportunities for both enrichment and risk experimentation, subject to escalating both adaptive and maladaptive adjustment (Arnett, Reference Arnett2007; Cicchetti & Rogosch, Reference Cicchetti and Rogosch2002; Min, Yoon, Minnes, Ridenour, & Singer, Reference Min, Yoon, Minnes, Ridenour and Singer2019). Lastly, the correlation between internalizing and externalizing behaviors is much higher on the ASR at 21 years than the YSR at 12 and 15 years, mirroring the lifetime prevalence of multimorbid disorder increased with age: 14.2% at ages 13–14, 29.9% at ages 15–16 (Merikangas et al., Reference Merikangas, He, Burstein, Swanson, Avenevoli, Cui and Swendsen2010), and 33.9% at ages 18–29 (Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005). It may also reflect the increasing prevalence of many mental disorders from childhood to young adulthood (Merikangas et al., Reference Merikangas, He, Burstein, Swanson, Avenevoli, Cui and Swendsen2010).
These findings should be interpreted considering limitations. First, we relied on offspring self-report on internalizing and externalizing problems, subject to offspring ability and willingness to assess and reflect on their own problems. Second, we operationalized non-kinship care as the experience of non-kinship care during the first 12 years of life, but the lack of explicit details might not fully account for the experience and its relationships with behavioral adjustment. The account of maltreatment was obtained retrospectively, prone to recall error and social desirability bias. Third, given the relatively small sample size and complexity of the model, separate models for males v. females were not tested, an important consideration for future studies. Fourth, although we controlled multiple confounders of PCE, our observational design does not dismiss that the indirect association of PCE with behavioral adjustment could be a function of unmeasured confounders of PCE. Lastly, generalizability of the findings may be limited to urban, predominantly African-American individuals of low SES whose mothers were assessed for prenatal drug use at delivery.
Strengths of the study include its longitudinal prospective design with multiple follow-ups from birth, use of multiple methods including biological markers to identify PCE, desirable retention rates, and the careful assessments of numerous environmental variables to characterize the developmental context. In addition, our novel analytical approach to separate between- and within-person effects in examining reciprocal processes between internalizing and externalizing problems over time clarified their apparent bidirectional relationships. Our study indicates that PCE increases risk for behavioral problems from ages 12 to 21 through poor family functioning, non-kinship care placement, or/and maltreatment, underscoring the critical implication of reducing substance use among pregnant women to improve next generation's mental wellbeing. Continued studies into later adulthood could expand our understanding of how PCE may contribute to behavioral, functional, and health outcomes over time.
Financial support
This research was supported by the National Institute on Drug Abuse Grants R01-07957 and R01-042747.
Competing interest
None.
Ethical standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
