Introduction
More than half of adolescents with a mental health problem meet diagnostic criteria for more than one disorder (Kessler et al.,
2012), highlighting problems with classifying psychopathology using narrow diagnostic categories. This issue is recognised in the DSM-5 (American Psychiatric Association,
2013), which calls for further research into empirically-supported frameworks that allow a conceptualisation of psychopathology along broader dimensions. Dimensions such as internalising (a propensity to experience anxious, depressive and somatic symptoms) and externalising (a propensity to experience aggressive, impulsive and disruptive behaviour; Achenbach
1966; Achenbach et al.,
2016) problems provide alternative ways to understand, diagnose and manage psychopathological difficulties. While dimensional approaches are not new, there is a need for greater knowledge of how such dimensions manifest in the population, whether they truly reflect the way symptoms cluster in individuals, and whether they form distinct or overlapping profiles of psychopathology. Large accessible datasets, such as the Adolescent Brain Cognitive Development (ABCD) study, provide new opportunities to identify subgroups of individuals with shared psychopathological profiles, and to explore ‘behavioural phenotypes’ thought to be associated with known risk factors. The comprehensive cognitive and neural data collected in the ABCD study also allows exploration of the neurocognitive factors that are associated with different psychopathological profiles (Dick et al.,
2021).
One such behavioural phenotype is the ‘preterm behavioural phenotype’ (PBP) which has been associated with Very Preterm (≤ 32 weeks of gestation) birth. Among Very Preterm cohorts, there is an increased risk for inattention, anxiety and depression, and peer relationship difficulties relative to birth at term, with the risk for conduct problems or aggressive or delinquent behaviours remaining similar to term-born peers (Fitzallen et al.,
2020; Hille et al.,
2001; Johnson & Marlow,
2011; Mathewson et al.,
2017; Wolke et al.,
2019). This pattern of difficulties is echoed in diagnostic studies in which attention-deficit/hyperactivity disorder (ADHD), autism spectrum disorder (ASD) and anxiety and depressive disorders are the most prevalent psychiatric disorders after Very Preterm birth (Wolke et al.,
2019). It is proposed that risk for this profile of symptoms results from interruptions to maturational processes in brain development or brain injury following Very Preterm or Extremely Preterm birth (≤ 28 weeks gestation; Volpe
2009), the risk of which increases as gestational age at birth decreases (Rogers et al.,
2018). The PBP was proposed on the basis of diagnoses observed at the group level, and much of the evidence focuses on group-level analysis of symptoms common in cohorts born Very Preterm or Extremely Preterm. More research is required to better understand how behavioural symptoms cluster in individuals born preterm, whether those born Moderate-Late Preterm ( 32 to 36 weeks gestation) may be at risk for the PBP, and the extent to which this phenotype is unique to preterm birth.
Investigations of how dimensions of psychopathology are observed in the ABCD dataset more generally have, to date, only been conducted using forms of factor analysis (Michelini et al.,
2019; Moore et al.,
2020); a variable-centred technique. Variable-centred approaches can identify how symptoms align along dimensions in a dataset, but assume they align in the same way across all individuals in the population. Conversely, person-centred approaches to psychopathology assume that associations between symptoms can differ across individuals, and examine this heterogeneity to define subgroups for whom symptoms cluster in ways that are maximally similar within the group, and are different to individuals in other groups. Thus, person-centred approaches are a valuable way to investigate phenotypes and characterise psychopathological profiles.
In other datasets, researchers have demonstrated how applying person-centred approaches such as latent class analysis (LCA; Bianchi et al.,
2017; Essau & de la Torre-Luque,
2019) and latent profile analysis (LPA; Basten et al.,
2013; Bonadio et al.,
2016; Webb et al.,
2021) to dimensional measures in child and adolescent samples can help us better understand how symptoms cluster. Measures of psychopathology have included parent-report questionnaires such as the child behaviour checklist (CBCL; Basten et al., 2012; Bianchi et al.,
2017), interviews such as the Composite International Diagnostic Instrument (CIDI; Essau & de la Torre-Luque,
2019), or multi-source measures (Bonadio et al.,
2016; Webb et al.,
2021). Samples have differed in (i) size, from smaller samples of 1,206 (Bonadio et al.,
2016) to large samples of 10,123 (Essau & de la Torre-Luque,
2019); (ii) age, with some recruiting children only (Basten et al., 2012), adolescents only (Webb et al.,
2021; Essau & de la Torre-Luque,
2019), or spanning childhood and adolescence (Bianchi et al.,
2017; Bonadio et al.,
2016); and (iii) source, with recruitment from the community (Basten et al., 2012; Webb et al.,
2021; Essau & de la Torre-Luque,
2019) and clinically referred populations (Bianchi et al.,
2017; Bonadio et al.,
2016). Despite diversity in approaches, there is much consistency in the findings across studies.
Along with a ‘normative’ profile of individuals who display low or no risk for psychopathology, studies find a profile consistent with the internalising dimension of psychopathology, predominantly characterised by self-directed negative emotions such as anxiety and depression (Basten et al.,
2013; Bianchi et al.,
2017; Bonadio et al.,
2016; Essau & de la Torre-Luque,
2019; Webb et al.,
2021). Profiles aligned with the externalising dimension are more variable. Basten et al. (
2013) and Essau & de la Torre-Luque (
2019) identified a profile consistent with externalising problems, while Bonadio et al. (
2016) identified
two profiles characterised by externalising behaviour; one in which aggressive and oppositional behaviours were moderate, and the other distinguished by additional severe problems with delinquency and for which aggressive and oppositional behaviours were also more severe. On the other hand, rather than an externalising profile, Bianchi et al. (
2017) identified a group with higher risk of inattention and hyperactivity, while the risk for symptoms of aggression and delinquency remained moderate and the risk for internalising problems was low. Finally, with the exception of Essau & de la Torre-Luque (
2019), who found their sample was best described by only 3 profiles (normative, internalising, externalising), most studies have also identified a profile characterised by difficulties in most, if not all, domains (Basten et al.,
2013; Bianchi et al.,
2017; Bonadio et al.,
2016; Webb et al.,
2021). This is often termed the ‘dysregulation’ profile. Importantly, these profiles often do not map directly on to traditional diagnoses. For example, probabilities of diagnoses of ADHD are elevated to a similar degree in both internalising and externalising profiles (Essau & de la Torre-Luque,
2019). The use of person-centred approaches may therefore provide an important adjunct to more traditional diagnostic approaches and opportunities to enhance our mechanistic understanding.
Only a small number of studies have used person-centred approaches to examine the PBP. These studies include samples at a range of ages and born at a range of gestations; 8-year-olds born at < 28 weeks (Burnett et al.,
2019), 5-year-olds born at < 30 weeks (Lean et al.,
2019), 6-year-olds born at < 36 weeks (Poehlmann-Tynan et al.,
2015), and 2-year-olds born at 32 to 36 weeks (Johnson et al.,
2018) of gestation. Children were classified on the basis of the profile of behavioural and emotional difficulties they demonstrated (Burnett et al.,
2019), but a number of studies also incorporated measures of cognition (Johnson et al.,
2018; Lean et al.,
2019; Poehlmann-Tynan et al.,
2015) into the indicators used to classify subgroups. These studies showed that in infancy and childhood those born at preterm gestations were either over-represented compared with term-born children in profiles reflecting sub-optimal outcomes (Burnett et al.,
2019; Johnson et al.,
2018; Lean et al.,
2019), or in the case of Poelhmann and colleagues (2015) which did not include a term-born group, the majority of the sample (70%) were allocated to sub-optimal classes. Consistent with the conception of the PBP informed by cohort studies, the sub-optimal profiles in which preterm-born children were over-represented emphasised the risk for elevated, but often sub-clinical difficulties (Lean et al.,
2019), and with a tendency for higher risk of inattention and hyperactivity (Burnett et al.,
2019; Lean et al.,
2019), socio-emotional difficulties (Burnett et al.,
2019; Johnson et al.,
2018; Lean et al.,
2019), and a lower risk of conduct problems (Burnett et al.,
2019; Lean et al.,
2019; Johnson et al.,
2018). However, there is limited evidence of a profile that reflects a single set of difficulties specific to preterm-born individuals. Lean et al. (
2019) identified a ‘school-based hyperactive-inattentive profile’ to which only 3% of term-born children, relative to 15% of Very Preterm children, were allocated, that they considered may reflect the PBP. Yet only Johnson et al. (
2018), which recruited a large sample of children born Moderate-Late Preterm, identified a profile that was uniquely observed in their preterm sample, of which 7% were allocated to this class. Indeed, some studies found that Very Preterm (Lean et al.,
2019) or Extremely Preterm (Burnett et al.,
2019) children were over-represented in multiple sub-optimal classes, rather than a single class representing the PBP.
However, person-centred approaches such as LCA require large sample sizes for good class recovery, with an evidence-based heuristic indicating that at least 500 cases are required for most models (Nylund-Gibson & Choi,
2018). This requirement becomes more important still when subgroups of interest may comprise a small proportion of the overall sample (Nylund et al.,
2007), particularly for researchers who wish to explore correlates of class membership. With the exception of Johnson et al. (
2018; 638 Moderate-Late Preterm and 765 term), in previous studies using these approaches to examine the PBP fewer than 500 cases were included, with as few as 125 (85 Very Preterm) children included in Lean et al. (
2019). Moreover, studies of the PBP have, to date, focussed on profiles observed in children aged 8 years or younger and there is a relative paucity of research into outcomes in adolescence.
Beyond preterm populations, person-centred studies of psychopathology in adolescence have also been limited in the extent to which they examine risk factors associated with sub-optimal profiles. This likely stems from the reliance on survey-based data collection to recruit large samples. Risk factors previously examined have included those that can be easily measured by self-report, such as demographics (e.g. Essau & de la Torre-Luque,
2019) or exposure to life experiences (e.g. Webb et al.,
2021). Yet neurocognitive markers, which have been a common focus in studies of diagnostic groups, have not to our knowledge been investigated in relation to transdiagnostic profiles derived from person-centred analyses. Datasets such as those created by the ABCD study provide new opportunities to examine such associations.
Indeed, emerging work has begun to examine cognitive and neural correlates of latent dimensions of psychopathology identified via variable-centred approaches. For example, in the ABCD sample, general and specific dimensions of psychopathology have been linked to measures of broad cognitive function such as crystallised and fluid intelligence (Michelini et al.,
2019), as well as more specific areas of cognition such as executive function (Cardenas-Iniguez et al.,
2020; Romer & Pizzagalli,
2021) and indirectly (via executive functioning) to white matter microstructure (Cardenas-Iniguez et al.,
2020). Executive function refers to the set of processes responsible for planning actions and regulating behaviour, including working memory and inhibitory control, which continue to mature through adolescence and have been linked to a variety of psychological disorders (Snyder et al.,
2015). White matter microstructure, which reflects the structural integrity of white matter connections in the brain, has also been linked directly to general psychopathological risk in other samples (Neumann et al.,
2020; Riem et al.,
2019; Vanes et al.,
2020).
Not only are executive functioning and white matter microstructure of particular interest when it comes to psychopathological risk, but both have been considered of mechanistic importance in relation to increased risk of psychopathology in preterm samples. A composite measure of executive function has been found to mediate the relationship between Very Preterm birth and total behavioural difficulties at school age (Schnider et al.,
2020), while studies have also examined the interplay between specific executive functions and symptom domains (e.g. Retzler et al.,
2019), showing that similar executive processes are implicated in both Very Preterm and term-born children. Similarly, white matter microstructure has been both pinpointed as a valuable transdiagnostic marker of psychopathology across the lifespan (Alnæs et al.,
2018), but also specifically, white matter development is commonly adversely affected following preterm birth and has been associated with the PBP (Brenner et al.,
2021; Loe et al.,
2013). Research into the risk factors associated not with individual dimensions of psychopathology, but with the transdiagnostic profiles of symptoms that are actually experienced, is needed to further understand the neurocognitive correlates of behavioural difficulties, and the extent to which preterm birth may confer risk for specific psychopathological difficulties.
The Current Study
The ABCD study recruited more than 11,000 pre-adolescents aged 9 to 10 years (Barch et al.,
2018; Volkow et al.,
2018). The data obtained at baseline included psychopathological, cognitive and MRI data, as well as retrospective parent-report measures of developmental history, including gestational age at birth. Although the sample was not recruited as a representative cohort of pre-adolescents born preterm, the numbers recruited provide a large sample in which to use person-centred approaches to examine profiles of psychopathology among pre-adolescents born preterm, and consider whether these reflect a PBP.
In the current analysis we made use of this comprehensive dataset to achieve three key aims. Firstly, we used LCA to identify separable classes based on psychopathology and allocate individuals to their most likely class (or psychopathological profile). Secondly, to ascertain whether a class that could reflect the PBP was observable, we tested whether preterm birth was associated with class membership. Finally, to build on evidence of relationships between neurocognitive factors and dimensions of psychopathology, we examined neurocognitive factors associated with class membership. Because cognitive factors beyond executive functioning may relate to dimensions and profiles of psychopathology (Blanken et al.,
2017), measures of language, memory and processing speed included in the ABCD cognition test battery were analysed in addition to measures of executive function. From the range of neural markers available in the ABCD dataset, white matter microstructure was selected for this analysis based on evidence of its relevance to psychopathology.