Multi-Trajectories of Conduct Problems, Hyperactivity/Inattention, and Peer Problems Across Childhood: Results from the Growing Up in Scotland Birth Cohort

This study used data from families enrolled in the Growing Up in Scotland (GUS) cohort study. The GUS dataset is a population-based cohort from Scotland launched in 2005 with children identified through the child benefit tax register. The initial sample included two cohorts, a birth cohort (N = 5,217) comprising infants born between 2004 and 2005 and a child cohort (N = 2,858) involving children born between 2002 and 2003. Additional details of recruitment and data collection can be found in Anderson et al. (2007). Data from sweeps 1 to 8 for the birth cohort were used, corresponding to when children were 10 months, two, three, four, five, six, eight and 10 years old. Inclusion criteria in the current study were (i) children who were enrolled in the birth cohort sample to ensure information on antecedents of early risk factors were available, and (ii) children having at least three assessments of conduct, hyperactivity/inattention and peer problems from sweeps 4 to 8, to model quadratic growth. Thus, the final sample in this study includes 3,578 families (68.6% of the total initially recruited sample). Sample characteristics of the included cohort as compared to those excluded from the current study can be found in Table 1. Comparison between samples revealed a greater degree of advantage across birth and demographic characteristics for children included in this study. Ethical approval was granted from Scotland’s Scottish “A” Multicentre Research Ethics Committee (MREC). Informed written parental consent was collected prior to each wave of data collection.

To examine concomitant trajectories of conduct problems, hyperactivity/inattention, and peer problems, a group-based multi-trajectory modelling analysis (Nagin, 2005; Nagin et al., 2016) was conducted. The group-based multi-trajectory model is an extension to the original group-based modelling analysis (Nagin, 2005) and adheres to a person-centred approach (Magnusson, 2003; Nagin, 2005). This analysis is an application of finite mixture modelling for identifying trajectory groups following diverse developmental patterns and considering multiple variables at once (e.g., conduct problems, hyperactivity/inattention, peer problems). According to Nagin’s recommendations (Nagin, 2005), two steps were followed for group selection: defining the number of groups that best fit the data and selecting the shape of each trajectory by determining polynomial terms for each group. For the first step, two to eight groups were assessed to compare model fit according to the Bayesian Information Criteria (BIC), with a larger BIC (in contrast to alternative modelling approaches) indicating a better fitting model (see Nagin, 2005). The main fit improvement occurred with the six-group model since the following groups were characterised by a more minimal BIC change from group 6 to group 7 and from group 7 to group 8 (Table 2). Further inspection of each model’s graph showed no distinguishing group difference when adding more groups (i.e., 7 and 8 groups). Therefore, considering the main fit improvement of the BIC statistic and the principle of parsimony, a six-group multi-trajectory model was chosen as best describing the distinct characteristics of the data.

For the second step, polynomial terms (i.e., quadratic, linear, and constant) were tested in the six-group multi-trajectory model until each parameter estimate was statistically significant. After selecting the number of groups that best fit the data and selecting the polynomial terms for each group, further criteria, such as the average posterior probabilities and the odds of correct classification were applied to assess the fit of the selected model (Nagin, 2005). The average posterior probabilities denote the chances that an individual displaying a particular pattern of behaviours will be assigned to a specific trajectory group. Odds of correct classification quantifies how much better the model is at allocating participants to groups than randomly assigning participants based on group size. Nagin (2005) suggested that posterior probabilities greater than 0.7 (or 70%) and odds of correct classification higher than 5 indicate good model fit. In this study, both indices offered further support for the six-group multi-trajectory model (Table 3). After model selection, risk factors for group membership were assessed at the bivariate level using analysis of variance and chi-square tests. Subsequently, only statistically significant factors at the bivariate level were entered into the multi-trajectory model estimation. Analyses were conducted using Stata v17.0. The statistical threshold was set to p = 0.05. The term significance is used in place of statistical significance henceforward.

Six trajectory groups were identified, exhibiting distinctive patterns of conduct problems, hyperactivity/inattention, and peer problems from four to 10 years old. Group 1 represented 21.1% of the estimated cohort and was the largest group. This group was labelled ‘non-engagers’ and was characterised by low decreasing conduct problems and hyperactivity/inattention, and low stable peer problems. Group 2 included an estimated 17.1% of children and was labelled ‘normative’. Compared to Group 1, this group also showed low decreasing conduct problems and hyperactivity/inattention, and low stable peer problems. However, the level of both conduct problems and hyperactivity/inattention were more elevated in Group 2 as compared to Group 1. Group 3 comprised an estimated 18.2% of children and was labelled ‘decreasing externalising/low peer problems’. In this group, both conduct and peer problems followed a quadratic trajectory, similar to the first and second groups, showing low decreasing conduct problems and low stable peer problems. Hyperactivity/inattention problems was more elevated in this group as compared to the previous two groups and followed a linear decreasing trajectory between four to ten years old. Group 4 included 14.9% of the estimated cohort and was labelled ‘low externalising/moderate peer problems’. Similar to group 3, conduct problems and hyperactivity/inattention were initially low and followed a decreasing trajectory. In contrast, peer problems were moderate at four years old, remaining constant thereafter. Group 5 included an estimated 20.2% of children and was labelled ‘moderate externalising/increasing peer problems’. Conduct problems were initially moderate and followed a slightly decreasing trajectory, while hyperactivity/inattention started at a moderate level, following a peak at age seven and then marginally decreasing at age 10. Peer problems had a linear increasing trajectory over time, although at a lower initial level and approaching moderate levels at age 10. Finally, Group 6 included an estimated 8.6% of the cohort, labelled as ‘multimorbid moderate-high chronic’. This group was characterised by moderate conduct problems, following a peak at age seven and then slightly decreasing when children were 10 years old. Hyperactivity/inattention problems also followed a quadratic trajectory with a peak at age seven and a marginal decrease by age 10, albeit at a high level across time. Peer problems started at a moderate level and followed an increasing linear trajectory, with high levels of problems from 6 to 10 years old. Multi-trajectory groups are displayed in Fig. 1 and parameters estimates in Supplementary File 1.

Bivariate analyses, including analysis of variance and chi-square tests, showed that low birth weight and area of the family dwelling were not significant factors in discriminating between group membership. All the other variables were significant for group membership (see Table 4). Subsequently, all significant factors at the bivariate level were modelled within the multi-trajectory estimation, using the non-engagers trajectory group (Group 1) as the reference group. At the perinatal level, having spent time in the NICU after childbirth increased the likelihood of membership only for the multimorbid moderate-high chronic group (Group 6). At the child level, male sex increased the likelihood of membership in all groups relative to the non-engagers reference group, with the exception of group 2, the normative group. Conversely, children with higher expressive language skills had a decreased likelihood of membership in groups 4, 5 and 6 as compared to the non-engagers group. At the family level, maternal age (i.e., younger mothers, under 20 years of age at childbirth) was not significant in discriminating between group membership. Single mother status increased children’s likelihood of membership in groups 4, 5 and 6. Low maternal educational level (i.e., leaving school with no qualifications) increased the likelihood of group membership only in the moderate externalising/increasing peer problems group (Group 5). Better than average caregiver mental health and higher maternal-infant attachment scores decreased the likelihood of membership in all groups as compared to the non-engagers group. Finally, having a family-of-origin categorised in the most socially deprived quintile (i.e., quintile five) increased the likelihood of membership in the multimorbid moderate-high chronic group (Group 6). Table 5 shows risk factors associated with group membership at the multivariable level.

The study’s first aim was to better understand subgroup prevalence and patterns of change within trajectories of conduct problems, hyperactivity/inattention, and peer problems across childhood. Accordingly, a four-group model was hypothesised to best fit the data, including a group with low problems, a normative group with low-moderate decreasing externalising problems and low peer problems, a high-decreasing group, and a high stable group. Findings partially supported the study hypotheses of group selection. Six groups emerged instead of four, with three groups matching our hypotheses (i.e., Groups 1, 2, and 6). The presence of the additional groups, ‘decreasing externalising/low peer problems’ (Group 3), ‘low externalising/moderate peer problems’ (Group 4), and ‘moderate externalising/increasing peer problems’ (Group 5) demonstrated greater developmental heterogeneity than anticipated. Group 3 displayed similar patterns to the first two groups’ conduct and peer problems. However, levels of hyperactivity/inattention in this group, while decreasing, were consistently more elevated across time compared to trajectories for Groups 1 and 2. Group 4 showed similar decreasing trajectories in the externalising problems domains to group 3, whereas peer problems were moderate but remained stable over time. This may reflect the influence of other factors associated with the maintenance of peer problems over time not included in this study, such as emotional problems (Forrest et al., 2020). Moreover, Girard (2021) reported two similar groups to Group 3 and 4 in the current study with respect to the level and progression of both conduct and peer problems over time, albeit with differing progressions of hyperactivity/inattention. In the current study, hyperactivity/inattention decreased over time in both groups whereas they remained stable across time in the former study. These findings suggest that externalising difficulties and peer problems are not occurring in isolation during childhood, and would support previous work demonstrating that behavioural and peer problems feed into the other across time (Hodges & Perry, 1999; Reijntjes et al., 2011; Schwartz et al., 2001).

The presence of multimorbidity in this study has practical and theoretical implications. In a practical sense, children exhibiting multimorbid problems (e.g., groups 5 and 6, combined estimated 28.8%) have been described at a greater risk for later maladaptive outcomes (Newman et al., 1998). Hence, children in the sampled population displaying elevated conduct problems, hyperactivity/inattention and increasing levels of peer problems may develop problematic outcomes in the future, which can potentially carry a higher disease burden and economic costs to societies (Pelham et al., 2007). Therefore, longitudinal identification of multimorbid groups is important for targeting children that may need additional support. From a theoretical perspective, the presence of multimorbidity challenges the classic classification systems of psychopathology, which are characterised by discrete and exclusive problems categories. Thus, multimorbidity findings from this study add evidence for further elaboration of classification systems of behavioural problems (e.g., the hierarchical taxonomy of psychopathology) and developmental psychopathology theories.

As anticipated, this study identified a non-engager (Group 1), normative (Group 2) and high-chronic (multimorbid moderate-high chronic group, Group 6) group. The non-engager and the normative group, a combined estimated 38.2% of children in the sample, exhibited low or declining symptoms in conduct problems, hyperactivity/inattention, and peer problems. These two low risk groups are in line with previous findings in the developmental literature (e.g., Fanti & Henrich, 2010; Girard, 2021; Patalay et al., 2017), suggesting that behavioural problems in preschool children may be normative, particularly as children enter into the classroom setting and are learning to navigate social relationships with peers. For most children, these early difficulties are expected to decline as higher-order skills are better developed, resulting in low-risk trajectories of externalising behaviours and peer problems after entry into formal schooling. Group 3 also followed a low-risk trajectory. Combined, groups 1–3 included 56.4% of children that presented within a normal range of problems in this sample. Finally, as expected, we identified a smaller group of children following a high-risk multimorbid trajectory across all externalising behaviours and peer problems (i.e., the multimorbid moderate-high chronic group), which included an estimated 8.6% of children. Previous developmental studies examining individual and joint trajectories of externalising behaviours and peer problems have also identified smaller groups of children at high-risk for elevated levels of chronically stable and/or increasing problems, however identified groups in these studies have typically consisted of a smaller proportion of the sampled population (i.e., between 3–6%; Shaw et al., 2005; Barker et al., 2008; Girard, 2021; Girard et al., 2019; Nagin & Tremblay, 1999; Patalay et al., 2017). As compared to these other studies, a higher proportion of children in the GUS cohort were identified as belonging to the high-risk group with increasing-chronic problems across multiple domains in childhood. According to previous research in Scotland using the GUS cohort (Marryat et al., 2015; Parkes et at., 2016) and a review of the Glasgow effect (Cowley et al., 2016), the higher proportion of children belonging to group 6 may reflect the cumulative impact of multiple life stressors on children’s wellbeing, including deprivation factors (i.e., socioeconomic inequalities, family distress), urban effects (i.e., remote location with lack of social connectedness among neighbours, urban stress), and cultural stressors (i.e., social attitudes).

The second aim of this study was to identify early risk and protective factors for group membership at multiple ecological levels, defining protective factors as influences that change a child’s reaction to environmental hazards that increase the risk of developing problematic outcomes (Rutter, 1985). Two protective factors at the family level were associated with membership in all groups relative to the non-engager group, which included primary caregiver mental health and maternal-infant attachment. Children with mothers reporting better than average mental health were less likely to belong to all groups compared to the non-engagers group, with odds ratio indicating a linear trend across higher problems groups. For example, children were 47%, 34%, 69%, 75% and 82% less likely to belong to groups 2, 3, 4,5 and 6, respectively. These findings are consistent with previous literature indicating that maternal mental health is an important factor in the development of children’s externalising behaviours and peer problems over time (Maruyama et al., 2019). Potential intergenerational transmission of risk mechanisms have been described, including genetic vulnerability, adverse socioeconomic factors, and parenting practices (Elgar et al., 2004). Moreover, maternal-infant attachment was also a protective factor across groups, with an increasing linear trend for groups display higher problems. For each point increase in attachment quality, children were 10%, 10%, 11%, 15% and 22% less likely to belong to groups 2, 3, 4, 5 and 6, respectively. These findings were in line with our expectations given the literature suggesting that caregiver mental health and attachment quality are associated factors that affect mother-infant interactions, and which impact upon children’s later behavioural outcomes (Kim-Cohen et al., 2005). Findings suggest that both protective factors are suitable targets for early prevention efforts, which can be especially beneficial for groups displaying greater problems (i.e., moderate externalising/increasing peer problems and multimorbid moderate-high chronic groups).

Three factors were common for elevated groups and included child sex, receptive language skills, and belonging to single parent households. In line with the literature (e.g., Girard, 2021; Girard et al., 2019; Teymoori et al., 2018), boys were more likely to belong to elevated groups with the largest odds (i.e., 3.1 higher odds) of belonging to the multimorbid moderate-high chronic group. It has been suggested that social determinants for males and females, such as parental and environmental socialisation, impact gender roles behaviours and, in turn, sex differences in childhood problems (Tremblay & Côté, 2019). Consequently, boys have been associated with greater conduct problems and hyperactivity/inattention (Loeber & Keenan, 1994). Genetic influences have also been acknowledged as important for sex differences. For example, evidence from twin studies exhibited a greater genetic influence on hyperactivity/inattention problems in boys but not girls, with girls needing greater liability to manifest symptoms (Rhee & Waldman, 2004; Rhee et al., 1999). Moreover, the gender-environment interaction has also been inspected to explain sex differences in externalising problems, with evidence showing that the social context plays an important role in the gender-specific phenotypic expression of genes (Dmitrieva et al., 2011). Therefore, our study findings add that when conduct problems, hyperactivity/inattention, and peer problems occur together, boys are more likely to belong to groups with greater problems than the non-engagers and normative groups. This result may reflect that boys are more vulnerable to multimorbid childhood problems, as similarly stated in studies modelling joint externalising problems and internalising problems (Fanti & Henrich, 2010; Girard, 2021; Murray et al., 2021). However, further research inspecting the role of social and genetic determinants is needed to better comprehend aetiology and sex differences in multimorbid problems.

In addition, better expressive language abilities at age three were identified as a protective factor for groups 4, 5 and 6. In line with previous research examining associations between children’s early language skills and externalising problems (e.g., Girard et al., 2014, 2016), our results highlight the importance of children’s early language ability to guard against subsequent externalising behaviours and peer problems. When children are better able to express themselves in social situations, the likelihood of using alternative forms of communication such as aggressive or acting out behaviours should decrease. This may consequently help to further reduce the likelihood of peer problems. Furthermore, children with single parents at 10 months old had greater odds (i.e., 1.8, 2.14, and 3.54 higher odds) of belonging to groups 4, 5 and 6, respectively. This finding may reflect the importance of early family support for mothers (Balaji et al., 2007) and its impact on later children’s outcomes. Single mothers may experience greater stress levels for facing parenthood, impacting mental health and mother-infant interactions, and subsequent offspring behavioural problems. However, further work inspecting this factor is necessary before asserting conclusions.

Finally, three factors were specific for group membership in groups 5 and 6. First, children whose mothers had no educational qualifications had an increased odds more than two-fold for group membership in group 5. Previous studies have reported the importance of maternal education in children’s behavioural outcomes independent of their socioeconomic status (Kalff et al., 2001). This association may be explained via shaping parenting practices related to having more knowledge and experience in the educational system. Surprisingly, maternal education was not similarly a risk factor for the multimorbid moderate-high chronic group (group 6), warranting further investigation. Two factors were significantly associated with increased likelihood of membership in group 6: NICU admission and social deprivation. Children who spent time in the NICU after birth had an increased odds of 1.9 for membership in the multimorbid moderate-high chronic group. Potential mechanisms suggested for the association between having stayed in the NICU and later externalising problems have included the stressful environment of the NICU, early experiences to invasive interventions, and the limited early interactions with caregivers, which may consequently reduce early stimulation and impact upon later social competencies (Linsell et al., 2019). Regarding social deprivation, children from families belonging to the most deprived quintile had two-fold odds of membership in the multimorbid moderate-high chronic group. Potential pathways to explain this association include the family stress model, which suggests that socioeconomic deprivation impacts parental mental health through economic stress, affecting parent–child interactions and later children’s outcomes (Elder, 1977).

Overall, the multimorbid moderate-high chronic group (Group 6) had more associated risk factors for group membership, followed by group 5 with some overlaps. This finding was expected since, according to the cumulative model (Atzaba-Poria et al., 2004; Rutter & Sroufe, 2000), groups with higher problems (i.e., groups 5 and 6) would have more associated risk factors than children within the no-engagers or normative groups. For these two groups, being a boy, staying in the NICU, displaying less developed language skills, having a single parent with no educational qualifications and with socioeconomic deprivation, cumulatively increases the risk for greater and more stable externalising problems and peer problems. In contrast, children with mothers reporting better than average mental health and attachment quality are less likely to develop these difficulties across childhood. All these risks and protective factors are already evident early in infancy and toddlerhood, suggesting clear implications for public health via social determinants of mental health and informing targets for prevention. For example, selective prevention strategies suggested by the National Institute for Health and Care Excellence (NICE) included the following risk factors for externalising problems: low socioeconomic status, low school achievement, child abuse or parental conflict, separated parents, parental mental health or substance abuse, and parental contact with the criminal justice system (NICE, 2017). However, important antecedent factors identified in this study could complement the ones suggested by NICE, including postnatal health antecedents (i.e., NICU stay), child sex (i.e., boys), and children with lower expressive language skills. Regarding early intervention strategies, NICE guidelines recommend interventions based on social learning theory for conduct problems in children aged three to 11 years old (NICE, 2017) and parenting training programmes for hyperactivity/inattention difficulties in children aged five years or younger (NICE, 2018). Considering the importance of attachment quality and parents’ mental health as protective factors within the context of the current findings, we also recommend exploring hybrid parenting interventions based on cognitive and attachment theory such as Mellow Parenting (Levi et al., 2019) or social learning theory and attachment theory like the Incredible Years (Webster-Stratton, 2001). Furthermore, in light of the multimorbidity between externalising behaviours and peer problems in the elevated symptoms groups, prevention and intervention efforts should also evaluate and incorporate strategies for addressing negative peer relationships that may precede or emerge alongside or as a consequence of externalising problems.

This study has several strengths, including being the first one using a group-based multi-trajectory approach to model conduct problems, hyperactivity/inattention and peer problems simultaneously across childhood, whilst using a large and representative population-based cohort from Scotland. In addition, we inspected several variables at multiple ecological levels as predictors for group membership, while using a multi-informant approach. Nonetheless, our findings are subject to some limitations to consider. First, the parent-report version of the SDQ was utilised to assess children’s outcomes, relying only on maternal reports. The combination of both maternal and teacher-reported SDQ would have been optimal considering that peer problems are generally detected at school. Nevertheless, maternal reports are frequently used in developmental research, and the parent-report of the SDQ has robust validity (Stone et al., 2010). Second, even when psychometric properties of the SDQ have been well validated (Goodman, 2001; Stone et al., 2010), for this study, Cronbach’s alpha was lower than the desirable threshold of ≥ 0.70 (Nunnally, 1978) in the conduct and peer problems scales. Moreover, the Condon Maternal Postnatal Attachment Scale also presented a Cronbach’s alpha below this threshold. A potential effect of this is the underestimation of the conduct and peer problems trajectories and their associated predictors – including maternal-infant attachment – because attenuation is more likely due to the diminution of maximal observable associations (Schmitt, 1996). Third, in order to properly apply quadratic growth, we only included children with at least three assessments in the period under investigation, reducing the final sample size used in this study. Attrition between the GUS birth cohort and the included sample exhibited several demographic disparities that may impact population representation. Comparison between samples revealed an underrepresentation of children with postnatal medical antecedents, younger mothers with lower educational levels, and socioeconomically deprived families, among other variables. Therefore, considering the underrepresentation of families facing more risk factors at the perinatal, child and family levels, it is possible that our findings may be specific to the subsample used in this study, requiring future replication to generalise results. Fourth, even though the inspection of risk and protective factors at the perinatal, child, and family levels provided further insights about each trajectory groups antecedent factors that were associated with group-membership, there are still ambiguities about the causal aetiology of co- and multi-comorbid trajectories of conduct problems, hyperactivity/inattention, and peer problems. Consequently, future research inspecting underlying causal mechanisms is needed to better understand the aetiology and developmental pathways of these problems. Finally, the GUS birth cohort data was only available until children were 10 years old. Thus, it was not possible to examine the progression of these developmental trajectories into adolescence, which is an important developmental period when changes in externalising behaviours and peer problems may become apparent.