Differentiating Early-Onset Persistent Versus Childhood-Limited Conduct Problem Youth

The Avon Longitudinal Study of Parents and Children is an ongoing population-based study designed to investigate the effects of a wide range of influences on the health and development of children. Pregnant women residing in south-west England who had an estimated date of delivery between April 1, 1991, and December 31, 1992, were invited to participate in the study. The study cohort consisted of 14,541 pregnancies and 13,971 singletons/twins who were still alive at 12 months of age. When compared with the 1991 U.K. National Census Data, the Avon Longitudinal Study of Parents and Children sample was found to be similar to the U.K. population as a whole, except the study sample showed a slightly higher proportion of married or cohabiting mothers and families who were owner-occupiers of their residences and (consistent with the area where the study is based) a smaller proportion of mothers from ethnic minorities (4.1% versus 7.6%) (16) . Ethical approval for the study was obtained from the Avon Longitudinal Study of Parents and Children Law and Ethics Committee and the local Research Ethics Committees. Additional, more detailed information about the Avon Longitudinal Study of Parents and Children is available online at http://www.bris.ac.uk/alspac/.

Mothers completed questionnaires at multiple time points during their pregnancy and their child’s infancy and childhood. Data on childhood conduct problems were collected from ages 4 to 13 years. The early childhood predictors examined in the present study were drawn from questionnaires completed between 32 weeks antenatal and 4 years of age.

Repeated assessments of child conduct problems during the past year were measured at ages 4, 7, 8, 10, 12, and 13 years by maternal reports of conduct problems using the Strengths and Difficulties Questionnaire (17) . The Strengths and Difficulties Questionnaire is a widely used screening measure with extensive psychometric support (www.sdqinfo.com). The questionnaire assesses conduct problems with the following five items: 1) “often has temper tantrums or hot tempers”; 2) “generally obedient, usually does what adults request”; 3) “often fights with other children or bullies them”; 4) “often lies or cheats”; and 5) “steals from home, school or elsewhere.” We created binary indicators (0=not high risk; 1=high risk) at each age based on national norms established for 5- to 10-year-old boys and girls in England and Wales (18) . The cutoffs used in the present study are strong predictors of conduct disorder (17) .

Socioeconomic status, marital status/cohabitation, and age of the mother at the birth of the child were each reported at 18 weeks postnatal. Socioeconomic status was coded via the Registrar General’s social class scale (19) . We compared mothers in classes IV and V (low socioeconomic status) with those in classes I, II, and III. Marital status reflected 1) no partner/no cohabitation, 2) not married but cohabiting, and 3) married/cohabiting. The age of the mother (mean=24.34 [SD=4.99] years) was dichotomized to contrast mothers who gave birth during their teens (e.g., age 19 and younger, code=1) with older mothers (code=0).

Maternal education was coded according to the mother’s highest level of education (at 32 weeks antenatal) as follows: 1) Certificate of Secondary Education or vocational qualifications only (basic school unfinished/some vocational training); 2) “O” level (minimum qualifications typically required for entry into further education at the time the study began); or 3) “A” level or higher (higher school-based academic qualifications [typically awarded at age 18], along with degrees and other postschool qualifications).

Drinking during pregnancy (18 weeks antenatal [e.g., risk was considered at least 2 units per day]), smoking during pregnancy (18 weeks antenatal [e.g., “yes” smoked cigarettes]), any maternal contact with the police (at child ages 0 to 2 years and/or 2 to 4 years), and maternal family history of alcohol use (17 weeks postnatal [e.g., “yes” mother’s mother or father had an alcohol abuse problem]) were all obtained between pregnancy and when the child was approximately 4 years of age. Each variable was dichotomized to indicate risk (coded 1) or nonrisk (coded 0).

Child birth weight, gestational age, and birth complications (e.g., abruption, preterm rupture, cervical suture) were recorded at the time of birth. Birth complications were dichotomized to contrast mothers with any complications (coded 1) versus those without complications (coded 0). Parity was obtained at 18 weeks gestation from a series of questions about previous pregnancies. Multiparous mothers were coded 1 and primiparous mothers were coded 0.

Language development was assessed using the McCarthy Scales of Children’s Abilities (20) at 24 months postpartum.

Child temperament was assessed with the Carey Infant Temperament Scale (21) at 24 months postpartum. The following four scales, thought to be most representative of undercontrolled temperament, were included: activity, adaptability, intensity, and mood.

Maternal depression and anxiety were assessed using the Edinburgh Postnatal Depression Scale (22, 23) and the Crown-Crisp Experiential Index (24, 25), respectively, at 32 weeks antenatal and 8 weeks postnatal. A cutoff of 13 has been found to be a good predictor of clinical depression (23), which was used in the present study. A cutoff of the top 15% was used for maternal anxiety, since it has been shown to predict both emotional and behavioral difficulties in children at 4 years of age and individual differences in cortisol levels in children at 10 years of age (25) .

Harsh parenting was assessed at age 24 months by the mothers answering the question, “When you are at home with your child, how often do you do the following”: 1) shout at him/her and/or 2) slap him/her. The response scale (reversed coded) was from 1 (everyday) to 5 (rarely/never).

Data pertaining to partner cruelty to the mother (child’s age 0 to 2 years; child’s age 2 to 4 years [e.g., any indication of emotional and/or physical abuse from the mother’s partner]), low emotional support for the mother (during pregnancy; child’s age 0 to 2 years [e.g., the mother having no one with whom to discuss her feelings]), and low practical support for the mother (during pregnancy; child’s age 0 to 2 years [e.g., whether there was anyone who could give the mother financial support and/or to whom the mother could turn in times of trouble]) were collected between pregnancy and when the child was approximately 4 years of age. Each variable was dichotomized to indicate risk (coded 1) or nonrisk (coded 0).

Data pertaining to child head injury were collected at multiple ages. We examined reports obtained at age 54 months that indexed whether or not the child had become unconscious as a result of a head injury at 0 to 2 years of age or 3 to 4 years of age.

Maternal attitude toward the child was measured via the following four questions targeting variations in parental attitudes at 33 months postpartum: 1) “I really enjoy this child”; 2) “Having this child makes me feel fulfilled”; 3) “I would have preferred that we not had this child when we did” (reverse coded); and 4) “I dislike/hate the mess that surrounds this child.” Response scales (reverse coded) varied from 1 (agree) to 4 (disagree).

Participants with at least four data points for conduct problems were allowed into the analysis. This resulted in a sample size of 7,218 children (boys=51%). We tested all risk factors for child conduct problems as predictors of exclusion from the analyzed sample. Mothers who were excluded from the present analysis were higher for certain sociodemographic risk factors (teen motherhood: odds ratio [OR]=1.42; maternal education: no basic qualifications, OR=1.98; no mid-level qualifications, OR=1.50]) and maternal smoking during pregnancy (OR=1.64). No other mother or child factors were associated with exclusion from the present analysis. Inevitably, there were also some missing responses for the early childhood predictors, although the effect was minimal within the selected sample (range=0.01%–7%).

The analyses proceeded in three steps. First, models for the developmental trajectories were estimated for mother-rated conduct problems. Growth mixture models (26) were used to estimate the trajectories in Mplus, Version 4.2 (27) . A series of models was fitted beginning with a one-group trajectory model and moving to a six-group trajectory model. All models were estimated with random starting values. We also compared models that allowed boys and girls to vary in growth parameters of trajectory groups (variant) with models for which we restricted the growth parameters of each individual trajectory to be the same for boys and girls (invariant). When constraining growth parameters in Mplus (27), the statistical indices available to establish the best fit model are via the Bayesian Information Criterion (28) and entropy. The Bayesian Information Criterion is a commonly used fit index in which lower values indicate a more parsimonious model. Entropy is a measure of classification accuracy with values closer to 1 indexing greater precision (range=0–1) (29) .

In the second step, we used univariate multinomial logit regressions, in SAS PROC CATMOD (30), for boys and girls, respectively, to examine the capacity of family characteristics, child characteristics, and child experiences to distinguish membership of the persistent and childhood-limited trajectories. Cases were weighted by the posterior probabilities of trajectory membership.

In the third step, we used multivariate multinomial logit regressions, for boys and girls, respectively, to identify independent robust risks that would predict trajectory membership above and beyond the variance attributable to the other predictors. We then examined sex interactions between identified significant multivariate predictors.

Table 1 contains the fit indices for the sex variant and sex invariant models. Overall, the sex invariant models fit better than the sex variant models. The four-trajectory invariant model had the best Bayesian Information Criterion value, but entropy decreased compared with the two-trajectory and three-trajectory models. We chose to examine the four-trajectory model, since it was consistent with our hypotheses and supported previous longitudinal research (6 – 8) .

The four trajectories ( Figure 1 ) were 1) early-onset persistent conduct problems (9%), 2) childhood-limited conduct problems (15%), 3) adolescent-onset conduct problems (12%), and 4) children with low involvement in conduct problems (64%). Boys were overrepresented in the early-onset persistent, childhood-limited, and adolescent-onset trajectories.

Table 2 demonstrates the contrasts between the persistent and childhood-limited trajectories for boys. Significant predictors for family characteristics were lower levels of maternal education and higher levels of teenage motherhood, maternal trouble with the police, and smoking during pregnancy. For child characteristics, youths with persistent conduct problems had higher levels of all four indicators of undercontrolled temperament. For child experiences, persistent youths had higher levels of exposure to maternal depression and anxiety (both antenatal and postnatal), harsh parenting, partner cruelty to the mother (child age: 0 to 2 years; 2 to 4 years), low emotional support for the mother (child age: 0 to 2 years), and low practical support for the mother (during pregnancy; child age: 0 to 2 years). For maternal attitudes toward the child, boys with persistent conduct problems had mothers who reported that they did not enjoy their child.

Table 3 shows the contrasts between the persistent and childhood-limited trajectories for girls. Significant predictors for family characteristics included low socioeconomic status and not married but cohabiting mothers. For child characteristics, girls with persistent conduct problems had somewhat higher levels of undercontrolled temperament. For child experiences, girls with persistent conduct problems had mothers with higher levels of depression (postpartum) and anxiety (antenatal and postpartum), harsh parenting, partner cruelty to mother (child age: 0 to 2 years; 2 to 4 years), low emotional support of mother (during pregnancy; child age: 0 to 2 years), and low practical support of mother (child age: 0 to 2 years). For maternal attitudes toward the child, girls with persistent conduct problems had mothers who reported that they 1) did not enjoy the child and 2) hated the mess surrounding the child.

For boys, significant multivariate predictors of the persistent versus childhood-limited contrasts were 1) temperamental activity levels, 2) maternal prenatal anxiety, 3) low maternal enjoyment of the child, and 4) partner cruelty toward the mother (child age: 2 to 4 years). For girls, significant predictors were 1) harsh parenting, 2) partner cruelty toward the mother (child age: 0 to 2 years), and 3) low practical support for the mother (child age: 0 to 2 years).

Table 4 demonstrates the multivariate solution for boys and girls simultaneously. Sex interactions were not significant and are therefore not shown. All predictors remained significant, except practical support for the mother. In summary, factors that increased the risk of the persistent conduct problems trajectory relative to the childhood-limited conduct problems trajectory, for both boys and girls, were 1) temperamental activity levels, 2) anxiety during pregnancy, 3) harsh parenting, and 4) partner cruelty toward the mother.

The present results should be interpreted in the context of five main limitations. First, most measures were based on maternal reports, raising the possibility of shared method variance. Future studies should incorporate multiple informants as well as biological indices (e.g., cortisol) of prenatal maternal stress. Second, although the mothers and children of the Avon Longitudinal Study of Parents and Children represent a broad spectrum of socioeconomic status backgrounds, the sample includes relatively low rates of ethnic minorities. The present results will need replication with more ethnically diverse samples. Third, we did not support prior research that reported family history of alcohol abuse and head injuries as risks that differentiated the persistent and childhood-limited pathways. Possible reasons for these disparate findings might be that 1) the measure of family history of alcohol problems was not as detailed in the present study as it was in the prior report (9) and 2) we examined head injuries up to age 3, whereas the prior study examined injuries up to age 17 and, hence, could not differentiate whether head injury was a predictor or result of persistent conduct problems (7) . Fourth, similar to most large longitudinal cohorts, the Avon Longitudinal Study of Parents and Children has faced attrition over time. As expected, younger and more socially disadvantaged mothers were more likely to be lost to follow-up. Because these predictors of attrition are also predictors of child conduct problems, our sample almost certainly underrepresents children with the most severe difficulties. However, by the same token, we anticipate that the risk associations we detected are also likely to have been attenuated. A recent study based on the Avon Longitudinal Study of Parents and Children cohort (39) confirmed that although attrition affected the prevalence rates of antisocial behaviors and related disorders, associations between risks and outcomes remained intact (although conservative estimates of the likely true effects). Finally, we focused on prenatal and early postnatal risks assessed prior to age 4. Later child- and family-based risks may also contribute to differentiating and maintaining early behavioral trajectories and are important targets for future study.