It is well understood that early emotional and behavioural pathologies are indicative of a wide range of maladaptive outcomes. For example, early internalising problems such as emotional problems and anxious/withdrawn behaviours have been commonly associated with future major depressive disorder, social and specific phobias, body dissatisfaction, alcohol-related problems and suicidality, amongst others (e.g., Caspi et al., 1996
; Goodwin et al., 2004
; Orri et al., 2018
; Patalay et al., 2015
). Early externalising problems (e.g., conduct problems, hyperactivity) have also been associated with subsequent depressive disorder and alcohol dependence, along with higher rates of school dropout, unemployment, homelessness, teenage parenthood, relationship problems, poor health, criminality, and substance abuse (e.g., Bevilacqua et al., 2018
; Collishaw et al., 2004
; Colman et al., 2009
; Shaw et al., 2012
). Moreover, early peer problems have been associated with a wealth of similar negative outcomes, including but not limited to, lower academic achievement, higher rates of truancy and school drop-out, subsequent relationship problems, externalising, and internalising problems (Woodward & Fergusson, 2000
: Laird et al., 2001
; Reijntjes et al., 2010
). Given this vast range of maladaptive outcomes, it is important to better understand potential heterogeneity within longitudinal presentations of these difficulties across childhood. More specifically, whether there are subgroups of children following distinct trajectories of concomitant internalising, externalising, and peer problems, as the pathways to poorer outcomes, along with their etiology, may differ across distinct groups. In the current study, concomitant trajectories of internalising (i.e., emotional problems), two externalising (i.e., conduct problems and hyperactivity/inattention), and peer problems were modelled from early to late childhood.
Developmental theories, such as the early childhood perspective of aggression (Tremblay et al., 2018
) and the dual taxonomy of antisocial behaviour (Moffitt, 1993
), have shown that children with the most elevated levels of early and sustained behavioural problems for example, are those with the poorest outcomes in adulthood, particularly if no remediation or interventions are sought. However, developmental research suggests that for a large proportion of children, early displays of externalising behaviours such as aggression and conduct problems are normative in the toddler years and typically decline for the majority around the time when children enter formal schooling, when higher-order skills (e.g., language, impulse control, emotional regulation) are better developed (Cole et al., 2011
; Tremblay, 2010
; Tremblay et al., 2004
). Only a small percentage of children, between 5–10%, within population-based studies (e.g., Girard et al., 2019
; Nagin & Tremblay, 1999
; Odgers et al., 2007
; Shaw et al., 2003
) have been found to continue with elevated chronic levels of aggression and conduct problems into and across adolescence. Population-based studies have suggested higher levels, between ~ 15–20% of children, with elevated chronic levels of hyperactivity/inattention from childhood to adolescence (e.g., Galéra et al., 2011
; Pingault et al., 2013
; Shaw et al., 2005
), with up to 40% having continuing and elevated levels into adulthood (e.g., Daley, 2006
). On the other hand, internalising problems (e.g., emotional problems, anxious/depressed and withdrawn) are more likely to increase from early childhood onwards, particularly for girls (e.g., Gilliom & Shaw, 2004
; Leve et al., 2005
). Similar mechanisms (i.e., higher-order skill development such as cognitive and emotional regulation) have been suggested for increases found in internalising problems across childhood and adolescence, given that these skills bring with them a better capacity for reflection, rumination, and self-blame (e.g., Garnefski et al., 2005
). Population-based studies have found that between 5–15% of children experience high stable and/or high rising trajectories of internalising problems across childhood/adolescence, and between 30–55% experience moderate stable or moderate rising trajectories (e.g., Veldman et al., 2015
; Côté et al., 2009
; Toumbourou et al., 2011
). Studies examining developmental trajectories of peer problems have suggested between 10–25% of children are experiencing increasing trajectories of peer victimisation, with close to 5% of children experiencing chronically high levels across time (e.g., Barker et al., 2008
; Boivin et al., 2010
). This is of particular concern as taken together, up to 30% of children may be experiencing high increasing and/or chronic levels of peer problems starting in early childhood. Given the bidirectional association between peer problems and both internalising and externalising problems across childhood and adolescence (e.g., Vaillancourt et al., 2013
), there is a strong need to simultaneously examine developmental trajectories of internalising, externalising, and peer problems starting in early childhood, to better understand potential sensitive periods when comorbidity or multimorbidity are most likely to emerge.
Comorbidity, (and by extension multimorbidity), a common occurrence with internalising, externalising, and peer problems, places children at even greater risk for subsequent maladaptive outcomes (Newman et al., 1998
). Studies of comorbidity have found higher odds ratios for concurrent homotypic comorbidity (i.e., comorbidity within behavioural domains such as conduct disorder and hyperactivity), as compared to concurrent heterotypic comorbidity (i.e., across domains – internalising and externalising). However, certain heterotypic comorbidity (e.g., conduct disorder and depression) has been found to present with similarly high odds (Angold et al., 1999
), and particularly when the comorbidity is successive. Thus, consideration of comorbidity (and multimorbidity where possible), should also be of high priority when examining developmental models of psychopathology. Yet few studies to date have modelled concomitant trajectories of both internalising and externalising problems, with some exceptions (e.g., Nivard et al., 2017
; Murray et al., 2020
; Patalay et al., 2017
), in particular starting as early as late toddlerhood (e.g., Fanti & Henrich, 2010
), whilst using person-centered approaches. Moreover, no study to the best of my knowledge, has also included peer problems when modelling concomitant internalising and externalising trajectories starting in early childhood.
Joint and Multi-trajectories of Internalising and Externalising Problems
Nivard et al. (2017
) modelled internalising (major depression, generalised anxiety disorder, specific and social phobia) and externalising (oppositional defiance disorder, conduct disorder and ADHD) problems in a population-based cohort from the UK, from seven to 15 years of age. Trajectories of internalising and externalising problems were first modelled individually and then estimates of conditional probabilities were used to examine membership in internalising groups based on membership in externalising groups and vice versa. A five-group model was identified and largely suggested associated internalising and externalising trajectories (e.g., decreasing internalising membership was associated with decreasing externalising membership, increasing internalising membership was associated with increasing externalising membership). Only one group, the adolescent onset of internalising problems group, was found to be independent of having any externalising problems, supporting a ‘pure’ internalising group in this study. Murray et al. (2020
), examined multi-trajectories of internalising (anxiety, depression), externalising problems (oppositional defiance disorder, conduct disorder, aggression), and ADHD in the same age range (i.e., seven to 15 years), using a cohort from Switzerland. They used the multi group-based trajectory approach, which allows for the modelling of multiple behaviours simultaneously. Six distinct groups were identified, with a majority of groups similarly displaying comorbidity across domains. That is, two groups showed low to non-existent problems across domains, one group had low initial levels of internalising, externalising and ADHD but increased across all domains across time, one group had higher initial levels of internalising, externalising and ADHD but decreased across all domains across time, and one group had chronically elevated ADHD and internalising with decreasing externalising problems. Only one group evidenced a ‘pure’ trajectory of internalising problems, similarly to Nivard et al. (2017
A further two studies examined joint trajectories starting in earlier ages (Fanti & Henrich, 2010
; Patalay et al., 2017
). Fanti and Henrich (2010
), examined joint trajectories of internalising and externalising problems from two to 12 years old in a US cohort. Similar to Nivard et al. (2017
), individual trajectories of internalising and externalising problems were modelled first, followed by joint conditional probabilities of internalising and externalising problems. Eleven joint trajectory groups best fit the data and revealed groups with no-low problems, co-occurring internalising and externalising problems, high internalising and high decreasing externalising problems, and ‘pure’ internalising problems; similar to results found in studies with older children. Conversely, with this younger sample, evidence of ‘pure’ externalising trajectories, both moderate and chronic, were also identified. Again using a younger sample, Patalay et al. (2017
), examined joint trajectories of internalising (emotional symptoms) and externalising problems (conduct problems), in a UK population-based cohort from three to 11 years. Five trajectory groups were identified with similar patterns (albeit with fewer groups) to Fanti and Henrich (2010
), including support for a ‘pure’ moderate externalising trajectory group. Taken together, these studies suggest comorbidity is common when modelling joint trajectories of internalising and externalising problems in childhood and adolescence. ‘Pure’ internalising trajectories are also consistently identified across childhood and adolescence, whereas in contrast, support for ‘pure’ externalising trajectories appear to only be found in studies modelling trajectories starting from earlier ages.
Within a developmental psychopathology framework, it is understood that not all children with emotional or behavioural pathologies will present with comorbidity or even on the same pathways to poorer outcomes, as the above studies demonstrate, and so it is important to better understand individual variation in the presentation of longitudinal (and potentially comorbid/multimorbid) emotional and behavioural problems. The commonly used variable-centered approaches to the study of developmental emotional and behavioural problems have greatly advanced our knowledge, particularly regarding a better understanding of normative
developmental change (Laursen & Hoff, 2006
) and factors implicated in variation around mean-level trends in behaviour (Nagin, 2005
). Despite this, the common assumptions of homogeneity and linearity in these models (Bergman et al., 2006
) have their limitations as they suggest that the developmental course of behaviour x
, will either increase or decrease over time, and in a similar fashion for the majority of individuals. We know however, that internalising, externalising, and peer problems are non-normative rather than normative behaviours, and consequently likely to have more nuanced patterns of continuity and change over time, for different clusters of individuals within the population. Thus, applying assumptions of heterogeneity and non-linearity when modelling trajectories of concomitant internalising, externalising, and peer problems (i.e., a person-centered approach), offers a complimentary view to understanding any individual patterns of change in behaviours across development.
In line with the assumption of heterogeneity when modelling concomitant trajectories of internalising, externalising, and peer problems, this assumption is carried forward to the question of whether there are differences in etiology for the subgroups of children identified. In particular, the question of ‘common’ versus ‘specific’ risk factors in the etiology of trajectories (e.g., pure, homotypic, heterotypic comorbid subgroups) needs investigating (Cohen et al., 1990
). Through the use of longitudinal cohorts, it has become better understood that there is intergenerational continuity of early and sustained emotional and behavioural problems, resulting from two paths of transmission (i.e., genetic and environmental; Kim et al., 2009
; Capaldi et al., 2017
). Thus, the early identification of subgroups of children with emotional, behavioural, and peer problems, along with the identification of early risk factors for specific group-membership is of high importance to ‘breaking the cycle’. Some commonly examined environmental factors previously associated with internalising and externalising problems include pre- and postnatal risks (e.g., maternal prenatal smoking, premature delivery), parenting characteristics and behaviours (e.g., maternal depression, young mothers, harsh parenting) and indicators of socioeconomic status (Button et al., 2005
; Spittle et al., 2009
; Goodman et al., 2011
; Girard et al., 2014
; Girard et al., 2016
; Fergusson & Lynskey, 1993
In the current study, 13 factors were investigated and included children’s sex, preterm birth, low birth weight, stay in the neonatal intensive care unit, prenatal exposure to smoking, maternal age, maternal education, maternal depression, marital status, indicators of SES, quality of parent–child attachment, and maternal stress. While these factors are not exhaustive, they cover risk and protective factors from multiple levels of the ecological systems theory of development (Bronfenbrenner, 1994
). Given the novel approach of modelling concurrent internalising, externalising, and peer problems using a person-centred approach across almost a decade, starting in very early childhood, it was important to identify whether these previously established risk factors would 1) predict group membership and 2) reveal potential heterogeneity in etiology across groups.
Aims & Hypotheses
More valid classification of subgroups of children following pure and/or comorbid trajectories of internalising, externalising, and peer problems is needed to further our understanding of non-normative emotional and behavioural development from early to late childhood. The aims of this study are twofold. First, to examine trajectories of concomitant internalising, externalising, and peer problems from early to late childhood using a person-centered approach, to better understand prevalence and subgroups with possible concurrent and/or successive homotypic and/or heterotypic comorbidity. Second, to identify ‘common’ versus ‘specific’ risk factors for group-membership.
Grounded in the few emerging studies modelling comorbidity of internalising and externalising behavioural trajectories, it was hypothesised that a 6-group model would best fit the data. For example, it was expected that a low to non-engagers group would be identified. Both a comorbid increasing group and a comorbid decreasing group would be identified. A normative comorbid group (i.e., early moderate but decreasing externalising problems and low early but moderately increasing internalising problems) would be identified. And finally, two groups, an elevated chronic ‘pure’ externalising group and a ‘pure’ increasing internalising group, would be identified (Fanti & Henrich, 2010
; Murray et al., 2020
; Nivard et al., 2017
; Patalay et al., 2017
). Moreover, it was expected that some common risk factors would emerge across groups with higher behavioural problems (e.g., being male, exposure to maternal depression, lower maternal education, poorer quality of attachment), although it was similarly expected that the combination of risk factors across groups would likely vary. As a result, no specific hypotheses were made regarding risk or protective factors associated with differing trajectory group membership.
To the best of my knowledge, this study is the first to have modelled concomitant trajectories of internalising, externalising, and peer problems from early to late childhood, using a group-based multi-trajectory approach, whilst examining antecedent risk factors associated with group membership. A common method for investigating developmental psychopathologies is to examine individuals who deviate from the majority (e.g., Cicchetti & Cohen, 2006
; Laursen & Hoff, 2006
; Bergman et al., 2006
). Whilst this approach is well suited in many instances, it risks potential dilution of unique clusters of individuals, in particular when examining trajectories of homotypic and heterotypic comorbidity, within population-based samples. An advantage of the multi-group-based approach is the identification of numerous clusters of children exhibiting distinct patterns of simultaneous behaviours over time. In using the GUI Infant cohort, the results revealed that a 6-group model best fit the data. Complex and comorbid patterns of internalising, externalising, and peer problems were found in groups presenting with higher problems. Both continuity and change over time were observed within trajectory groups, along with common and specific risk factors predicting group membership. Given the extensively documented maladaptive outcomes associated with internalising, externalising, and peer problems when examined individually (e.g., Bevilacqua et al., 2018
; Rivenbark et al., 2018
; Fairchild et al., 2019
; Gutman & McMaster, 2020
; Galera et al., 2021
; Orri et al., 2018
), identification of comorbid and multimorbid trajectories using a population-based sample, along with a better understanding of antecedent risk factors associated with group membership, sheds several important new insights.
First, within the 6-group model, there was little evidence to support the existence of a ‘pure’ trajectory group in early to late childhood, for internalising, externalising, or peer problems, contrary to the hypothesis and previous findings (Fanti & Henrich, 2010
; Patalay et al., 2017
). Instead, difficulties in one domain appeared to indicate the presence of difficulty in another domain (both homo- and heterotypic comorbidity). Whilst distinct, groups 1–3 all exhibited low levels of internalising, externalising, and peer problems. That is, 76.1% of the sample fell within a normal range of behaviours, albeit with variation (i.e., increases/decreases/stability) over time. This is a higher proportion of children exhibiting no to low internalising, externalising, and peer problems as compared to the two previous studies modelling joint trajectories of internalising and externalising problems with similar age groups (Fanti & Henrich, 2010
; Patalay et al., 2017
). However, a notable difference is the inclusion of peer problems modelled in the trajectories within the current study.
With the exception of hyperactivity/inattention, groups 4 and 5 displayed similar patterns of behaviours across time and comprised a combined 20.9% of the sample. In both groups, emotional symptoms were found to increase linearly over time, although at higher levels for group 4. Whilst conduct problems were quadratic in shape for group 4 and linear for group 5, decreases from moderate levels at age three, to low levels by age nine were observed in both groups, following an expected developmental pattern (Shaw et al., 2003
). The experience of peer problems in both groups was moderate but stable across time. In contrast, hyperactivity/inattention for group 4 was moderate but stable across time, whereas for group 5 it was quadratic in shape, starting at moderate levels and increasing before tapering off at high levels thereafter. Whilst the analytic approach can only infer multimorbidity and not directionality between behaviours, it could be hypothesised that the early moderate levels of conduct problems and moderate-high levels of hyperactivity/inattention resulted in the observed stability of moderate peer problems in these groups across time. For instance, notable links have been found to support the early presence of homotypic comorbidity of conduct problems and hyperactivity/inattention with future peer problems in childhood (Andrade & Tannock, 2014
; Becker et al., 2012
; Gresham et al., 1998
), suggestive of successive heterotypic comorbidity. Moreover, stable levels of peer problems may contribute to increasing emotional symptoms over time (Bond et al., 2001
; O'Brennan et al., 2009
Group 6 included an estimated 3% of the sample and presented with chronic-increasing levels of elevated problems across domains. Levels of conduct problems started high and remained stable. In contrast, all other behaviours increased from moderate to high between three and nine years of age. The elevated levels across all behaviours highlight both concurrent and successive homotypic and heterotypic comorbidity in this group, another notable finding of the study. Much attention has been paid to homotypic comorbidity with somewhat less attention focusing on heterotypic comorbidity. This is particularly the case for mapping trajectories of possible concomitant internalising and externalising problems starting in early childhood. The increasing level of emotional symptoms, hyperactivity/inattention, and peer problems against the backdrop of stably elevated conduct problems suggests dependency across behaviours in a small group of children. High comorbidity between conduct problems and hyperactivity/inattention in developmental studies is not uncommon (Angold et al., 1999
; Beauchaine et al., 2010
). Theoretically, the combined presence of elevated conduct problems and hyperactivity/inattention may result in increased difficulties and rejection by peers over time, cascading into increased emotional symptoms. On the other hand, early emotional symptoms may result in withdrawal from peers and subsequent difficulty with attention during daily tasks. However, this latter hypothesis would not account for the early elevated and stable conduct problems found in this group. The identification of a smaller chronic/elevating group is line with previous studies examining developmental psychopathologies, which in individual and joint trajectories have suggested < 10% of children following a high-chronic developmental course (e.g., Barker et al., 2008
; Nagin & Tremblay, 1999
; Patalay et al., 2017
; Shaw et al., 2003
). Given the modelling of concomitant internalising, externalising, and peer problems, the prevalence rate of the chronic group in this study was on the lower end, with approximately 3% of children.
Knowledge of potential common versus specific risk factors associated with group-membership is critical to furthering our understanding of differing patterns of concomitant behavioural problems across childhood. Only three risk factors were uniformly found to predict membership in all elevated groups as compared to the combined reference group. These included prenatal exposure to smoking, maternal education, and maternal stress. There remains debate as to whether the link between prenatal exposure to smoking and consequent behavioural difficulties is direct or an artefact of characteristics associated with mothers who smoke (e.g., Roza et al., 2009
). In the current study, prenatal exposure to smoking was measured by the number of household members smoking during pregnancy, which may or may not have included the mothers, rather than maternal engagement alone. It is possible then that this finding may support a direct link between in-utero exposure to toxins and future emotional and behavioural problems. More work in this area is however first needed before conclusions can be drawn. Similarly to previous studies, lower maternal education was also a common predictor (e.g., Nagin & Tremblay, 2001
). Children with mothers who had primary/no or second level education as compared to third level were approximately 1.5 times at greater risk for belonging to groups 4 and 5. The relative risk ratio doubled for children in group 6, whereby having primary/no education increased the risk of membership by 3.1 and second level education by 1.8. A higher level of maternal stress was also a common risk factor for group membership in elevated groups. For each point increase in maternal stress, there was a 6%, 2%, and 5% increase in membership in groups 4, 5, and 6, respectively.
Three factors, low birth weight, maternal depression, and medical card status, were commonly associated with increased risk for membership in groups 4 and 6. Notably, the relative risk ratios for all risk factors were consistently larger for membership in group 6. The relative risk ratio for low birth weight in group 6 was almost two-fold as compared to the reference group, possibly driven by the high levels of both internalising problems and hyperactivity/inattention in this group (Aarnoudse-Moens et al., 2009
; Nigg & Breslau, 2007
). In group 4, the relative risk ratio was 1.4. Maternal depression slightly increased risk of membership in group 4 (RRR: 1.3), but more than doubled the risk of membership in group 6 (RRR: 2.1). This finding was not surprising given the characteristically poorer quality interactions between children and mothers suffering from depression, in addition to increased genetic risk of mental health difficulties (Kim-Cohen et al., 2005
). Medical card status, a proxy of income, was also found to increase the risk slightly for group 4, but more substantially for group 6 (RRR: 2.4 and 3.0 for full and partial coverage respectively).
Boys were at a two-fold higher risk for membership in group 5, and almost four-fold the risk for membership in group 6. Whilst boys are routinely found at risk for elevated trajectories of externalising problems (e.g., Girard et al., 2019
), the evidence is mixed when examining internalising problems (Dekker et al., 2007
; Mesman et al., 2001
). These results suggest that heterotypic comorbidity of internalising and externalising problems increases the risk of membership in elevated groups for boys in particular. Of interest, boys were not at higher risk for membership in group 4, which presented similarly to the low increasing internalising/mixed (moderate-decreasing, high-increasing) externalising/stable peer problems group (group 5) on emotional symptoms (albeit at lower levels), conduct problems and peer problems, but not on hyperactivity/inattention. In the latter group, hyperactivity/inattention increased over time whereas in the former group hyperactivity/inattention remained stable. This would suggest that the presence of either increasing difficulties in hyperactivity/inattention, lower levels of emotional symptoms, or both, resulted in this increased risk in group membership for boys. This result is notable in that it suggests that complex and concomitant internalising, externalising, and peer problems, particularly when hyperactivity/inattention is high, is more common amongst boys.
Quality of attachment was a common factor for membership in both groups 4 and 5. More specifically, risk of group membership decreased by 6% and 9%, respectively, for each point increase, suggesting that early positive attachment played a protective role against moderate levels of internalising, externalising, and peer problems, but not for the highest levels of problems.
Three factors were specific to group membership and included young maternal age (group 4), single parents and lower social class (group 5). Perhaps most surprising is young maternal age not increasing the risk of membership in group 6 (Lee et al., 2020
). Given the low number of mothers under 21 years of age in this sample, replication is however needed. Taken together, this study suggests some overlap of antecedent risk factors common to a couple of the elevated groups, although only three factors uniformly increased the risk for all three elevated groups. It would be of great interest to see future studies that extend the current work by examining these trajectories into adolescence, a transition period marked by great challenges for many youngsters. Moreover, given that the current study focuses uniquely on predictors of group membership at 9 months, future work would do well to build upon these findings by examining time-varying predictors (e.g., cognitive abilities, academic outcomes, parenting practices), which may provide additional understanding around continuity and change within trajectory groups, along with informing protective and risk factors across development.
Despite the noteworthy strengths of this study, including being the first to use a person-centered approach in modelling concomitant trajectories of internalising, externalising, and
peer problems, from early to late childhood, whilst using a large and representative population-based cohort, consequently allowing unbiased prevalence rates, some limitations need mention. First, parent reports were used to collect information on both risk factors and children’s outcomes. Thus, shared method variance is a possible concern and future studies would be well placed in using multiple informants. Second, attrition across waves resulted in the loss of almost a third of the initial cohort, with higher attrition amongst families at greater social disadvantage. However, to circumvent the disproportionate distribution, sampling weights were used which resulted in a 0.5% difference between the included sample and the population with respect to participant characteristics. Additional group-based multi-trajectory analysis using all participants with data on the SDQ for at least one time (n = 10,170) was also explored given the high attrition rates. Results were largely consistent (i.e., a 6-group model best fit the data, the shape of trajectories remained unchanged), although the proportion of group membership was slightly increased for the elevated groups when using the entire sample. Third, data was only collected up to age nine. Extending trajectories into adolescence would provide additional opportunities to better understand whether groups with elevated problems continue on the same path and whether potential adolescent-onset groups would emerge. Fourth, whilst the SDQ is one of the most widely used behavioural screening measures (Stone et al., 2010
), Cronbach’s alpha, which represents the ‘lower bound’ of the true reliability (Cronbach, 1951
), was below the desired threshold of ≥ 0.70 for three of the four scales in the current study. A potential consequence is the underestimation of behavioural trajectories and their associated risk factors given that attenuation, via the reduction of maximum observable associations between variables, is more likely (Schmitt, 1996
). Fifth, given the interest in understanding risk/protective factors specific to trajectory groups with elevated behavioural problems, the primary analysis was conducted in a two-stage process. That is, trajectory groups were extracted following the modelling of groups to collapse the first three (no-low) trajectory groups, as there is no way to collapse groups within the model itself. Consequently, groups were treated as observed. Finally, whilst risk factors were identified, inference of risk mechanisms could not be asserted given the observational nature of the study. Future work should build on this study by examining risk mechanism.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.