Compared to their full-term born peers (FT; 38–42 weeks’ gestation), children who were born very preterm (VPT; <32 weeks’ gestation) display greater behavioral difficulties, such as inattention, emotional dysregulation, socio-communication problems, anxiety, and internalizing behaviors (Rothbart et al., 2007; Johnson and Marlow, 2011; Arpi and Ferrari, 2013; Burnett et al., 2013; Brydges et al., 2018). VPT children also exhibit increased autism spectrum condition (ASC)-like traits (Treyvaud et al., 2013) and are more likely to receive an autism diagnosis compared to their term-born peers (7% vs. 1.5%, respectively) (Agrawal et al., 2018).

Studies focusing on ASC traits in VPT and extremely preterm (EPT; <28 weeks’ gestation) samples have revealed a distinct phenotypic expression from that observed in FT children, suggesting a preterm-specific ASC traits aetiology (Verhaeghe et al., 2016; Bröring et al., 2018; Chen et al., 2019). ASC trait aetiology in VPT/EPT children appear to be primarily rooted in poor socio-emotional abilities and difficulties with social communication and interaction (Johnson and Marlow, 2011; Jaekel et al., 2013; Johnson et al., 2018), rather than a combination of social communication and interactions problems, and rigid and repetitive behaviors and interests, as observed in FT samples. Moreover, social communication and interactions impairments appear to be more homogenously distributed across VPT compared to FT populations, who display greater behavioral heterogeneity (Chen et al., 2019), thus supporting the idea of a preterm-specific ASC trait aetiology.

VPT children are not only more likely to display elevated ASC traits compared to FT-born controls, but also exhibit broader developmental difficulties. A within-group stratification study conducted by Johnson et al. (2018) explored distinct patterns of cognitive and behavioral development in 2 year-old preterm children, and identified a “non-optimal” subgroup characterized by ASC symptomatology, poorer cognitive and language skills, greater social, emotional and attention difficulties, and heightened risk of developing behavioral problems. Indeed, VPT-born children differ on measures of temperament, displaying shorter attention span, poorer ability to focus, and higher rates of activity compared to their FT-born peers (Cassiano et al., 2020), which underlie externalizing behavior (Bora et al., 2014). VPT-birth has also been associated with greater cognitive difficulties, with VPT children showing lower intellectual quotient, working memory, verbal abilities and processing speed scores than their FT-born peers (Allotey et al., 2017). However, it is unclear how these difficulties identified in VPT children interact or influence one another, highlighting the need to investigate multiple cognitive and behavioral outcomes as a set of interconnected, dynamically interacting facets of development. Indeed, such a paradigm could allow for the identification of core cognitive-behavioral features to better understand preterm-specific ASC symptomatology (Borsboom and Cramer, 2013; Eadeh et al., 2021), in alignment with a (multi-)dimensional and trait-based approach to ASC (Constantino, 2009; Volkmar and McPartland, 2015; Constantino and Charman, 2016). Exploring a broad range of developmental outcomes impacted by very preterm birth in a single model, may allow us to delineate preterm-specific developmental correlates of ASC traits.

Previous studies have applied network analysis to ASC populations, studying the network topology of ASC and other behaviors that tend to be comorbid with ASC (Anderson et al., 2015; Ruzzano et al., 2015; Montazeri et al., 2019). These studies have increased our understanding of the dynamic interactions between ASC features and of pathways to outcome phenotypes. Network analysis is a statistical approach which uses nodes (i.e., variables indexing psychological/behavioral constructs) and edges (i.e., partial correlations between nodes) to map the interaction between psychological constructs, and evaluate how variables correlate with one another. Contrary to a correlation matrix, network analyses allow for a graphical representation of the degree and pathways of interaction, allowing for the most important variables underlying specific phenotypes to be identified.

To our knowledge, no prior study has used network analysis to explore multiple facets of development following VPT birth in mid-childhood. This developmental stage is particularly important, as evidence suggests that during early life it is complex to differentiate between ASC and ADHD traits and behavioral and temperamental difficulties (Shephard et al., 2019), hence a network analysis approach may be particularly useful in delineating complex phenotypes. In the present study we use network analysis to investigate ASC traits in the context of other behavioral and cognitive measures, in order to understand whether structural relationships between these outcome phenotypes differ between VPT and FT children. One of the advantages of this approach is that it allows us to explore how different developmental domains, previously analysed independently in VPT children, interact with one another in a single model. We focus on bridge symptoms, or nodes connecting different disorders or constructs (Jones et al., 2021), to understand comorbidity and shared characteristics between psychopathology, behavioral, emotional or temperamental difficulties.

The current study places a particular focus on ASC traits when comparing FT and VPT networks. While this study is mostly exploratory, we hypothesized that, in addition to VPT participants displaying greater ASC traits, behavioral, cognitive and temperamental difficulties compared to their FT peers, they would show a topologically different cognitive-behavioral network architecture. Specifically, we hypothesized that, similar to findings pertaining to VPT toddlers (Johnson et al., 2018), ASC traits would be more strongly associated with emotional and temperamental difficulties in VPT compared to FT children (Vlaeminck et al., 2020).

In this study we used network analysis to explore how ASC traits, cognitive, behavioral and temperament outcomes dynamically interact in VPT and FT born children. VPT participants displayed elevated ASC traits compared to their FT peers, with group differences in the SRS-2 subscales showing the largest effect sizes of all behavioral measures. Subsequent network analyses allowed us to further investigate the underlying mechanisms that may be driving these behavioral differences. The exploratory nature of this approach allowed us to uncover a number of novel associations that have not previously been observed in the published literature. Compared to the FT group, VPT participants’ ASC traits, cognitive, behavioral and temperament outcomes seemed to be predominantly centered around and driven by conduct and executive function difficulties, in particular shifting attention and organizing materials. In comparison, the FT participants’ behavioral, cognitive and temperamental profiles appeared to be associated with difficulties in initiating behaviors, emotional regulation, and to a lesser degree, executive function abilities compared to the VPT children.

While the main focus of this study was to explore how ASC traits and cognitive, behavioral and temperament outcome measures relate to one-another in VPT and FT groups, the findings regarding increased conduct problems in VPT compared to FT children are unexpected. This is because previous research has suggested that conduct problems may not be elevated in VPT compared to FT children (Wolke, 1998; Johnson et al., 2010). A direct network comparison indicated that in VPT children, conduct problems were more closely linked to constructs from other domains other than ASC traits, acting as a bridge between constructs, such as emotional control and organizing their environment. These findings suggest that where conduct problems do arise in VPT children, these may be more tightly linked with problems across different cognitive and temperamental domains than is the case in FT children. The absence of group differences in emotional problems is also at odds with previous findings. This could be due to differences in measures used to assess these. Indeed, while a group difference was observed on emotional problems, as measured by the BRIEF-2, with a low to moderate effect size, there was no such difference on with SDQ-rated emeotional problems. A principal component analysis conducted on an overlapping sample of the current participants aged 4 to 7 (Vanes et al., 2021) suggests that, while the BRIEF-2 emotional problem sub-scale score is mostly associated with cognitive abilities, the SDQ adopts a more emotional, affect driven conceptualization of emotional problems. This may explain the discrepant findings, given that children in our VPT sample have significantly poorer cognitive abilities, in turn affecting their socio-emotional and behavioral development (Montagna and Nosarti, 2016).

The observations in the VPT group reflect the presence of a constellation of symptoms commonly reported in VPT populations: poorer socializing abilities, increased internalizing behaviors and poorer attention, which are referred to as the “preterm behavioral phenotype” (Johnson and Marlow, 2011). At first glance, the co-occurrence of executive function and socio-communication difficulties seen in the VPT group may be erroneously interpreted as ASC-like social cognition deficits (Alduncin et al., 2014). However, VPT children may display a unique ASC phenotype, characterized by more predominant difficulties in cognition, attention and socio-emotional processing. Furthermore, despite VPT children exhibiting elevated autistic traits, these do not always reach clinical thresholds (Johnson and Marlow, 2011). Given the isolated and highly clustered SRS-2 subscale scores in the FT network, we tentatively speculate that the SRS-2 in VPT children captures more general behavioral, cognitive, temperamental and socio-emotional difficulties, with a differing etiological pathway to ASC traits or preterm-specific ASC aetiology and symptomatology, rather than more “pure” autistic traits.

The specific network dynamics of individual SRS-2 subscales, the primary variables of interest in our investigation, highlight the way in which ASC-like traits may interact uniquely with other developmental measures in otherwise healthy VPT children. For example, social motivation and social cognition in VPT participants were related to executive functions, in particular working memory, planning and organizing. Notably, the VPT network revealed that the link between working memory and social cognition was not direct, but mediated by verbal comprehension. Social cognition deficits seemed to impact verbal comprehension, which was in turn associated with working memory. Although the direction of effect cannot be ascertained from our analysis, bidirectional mediating effects might be possible. In a large sample of healthy individuals aged 16–91, Froiland and Davison (2020) found that when controlling for other cognitive measures, working memory had no effect on social perception, or the ability to make inferences and form impressions from social interactions. In their study, verbal comprehension had the largest effect on social perception out of all cognitive measures, with increased verbal comprehension difficulties associated with worse social perception. This finding is in line with ASC research in FT adolescents, which highlights poorer verbal comprehension in ASC participants compared to controls, and associated difficulties in social perception (Holdnack et al., 2011). With respect to planning and organizing, Vogan et al. (2018) showed that an improvement in planning and organizing, self-monitoring and initiating behaviors was associated with reduced social deficits in children with autism. However, improvements in emotional control, shift and inhibition were not associated with better social development. These findings are also aligned with our VPT network results, suggesting that VPT children display interactions between ASC traits and other developmental measures that are similar to those observed in clinical ASC groups.

The findings of co-occurring executive function difficulties underlying ASC phenotype and symptomatology in VPT participants is also in line with ASC research in FT, cognitively able children. Pellicano (2010) showed that early executive function skills predicted future abilities in Theory of Mind tasks in children aged four to seven, even when controlling for age and (non)verbal comprehension, thus highlighting the link between socio-cognitive skills and domain-general processes. These findings could be also interpreted in the context of literature revealing associations between cognitive impairments and socio-emotional difficulties in VPT children (Mansson et al., 2014; Rogers et al., 2016). Neuroimaging studies could offer a possible explanation for the overlap between VPT-specific executive function and broader cognitive difficulties and ASC traits, highlighting associations between preterm-related neonatal brain alterations and a later ASC diagnosis (Ure et al., 2016; Eklöf et al., 2019).

Taken together, our findings highlight the underlying role of executive function and temperamental difficulties in the VPT presentation of ASC traits. In particular, our findings suggest that ASC traits, as measured by the SRS-2, may reflect more general developmental difficulties specifically in VPT children, rather than a segregated or “pure” presentation of autistic traits, highlighting the importance of understanding the way in which these traits interact with other developmental measures. This is consistent with recent efforts to develop a cognitive understanding of ASC (Frith, 2021; Happé and Frith, 2021), as a selective focus on behavioral phenotypes could lead to a too broad a definition of ASC, capturing behaviors that are not intrinsically inherent to ASC. Studies focusing on a cognitive understanding of ASC allow for core differences and specificities to be revealed, which in turn reduces the risk of misattributing VPT phenotype-related behaviors to ASC.

Our findings demonstrate the theoretical and practical relevance of departing from diagnostic-driven labels, and focusing on dimensional and transdiagnostic traits, such as executive function deficits, in order to increase our current understanding of psychopathology (Krakowski et al., 2020; Siugzdaite et al., 2020; Vaidya et al., 2020). Our findings further support the concept of equifinality (Cicchetti and Rogosch, 1996), recognizing the possible existence of multiple pathways associated with specific outcomes. We have in fact recently shown that only a subgroup of VPT children exhibiting high ASC traits displayed neonatal cerebellar alterations, suggesting distinct aetiological trajectories associated with ASC outcomes (Hadaya et al., 2022). This study, and in particular our results highlighting the centrality of executive function and behavioral problems in driving other difficulties in VPT children, further strengthens the argument in favor of person-based treatment approaches targeting underlying behaviors and difficulties rather than surface level symptoms (Kasari et al., 2014). This is particularly important when recognizing the potential for these underlying mechanisms, such as poor executive control, to predict later ASC-related difficulties (Pellicano, 2010) and the importance of targeting these mechanisms early in development.

Our study has several limitations. First, since we only considered VPT children, our findings are not generalizable to moderate-to-late preterm children (32–37 weeks’ gestation). Second, the higher prevalence of VPT children reaching a clinical threshold cut-off on the SRS-2, which is parent-rated, compared to other studies, could introduce a rater bias when compared to studies using clinical observations (Aldridge et al., 2012; Treyvaud et al., 2013). This limitation extends beyond the answers on the SRS-2, as all our measures used parental reports for evaluating child behavior, except IQ, which may lead to common method variance bias (Podsakoff et al., 2003). In addition, all measures included in this study were collected at a single time point, making inferences about directionality challenging. Future research can usefully investigate longitudinal network changes over time, in order to fully understand the potential causal dynamics between the behavioral measures under investigation. Finally, the unequal sample sizes of the VPT and FT groups may have contributed to qualitative and quantitative differences in network associations between groups, given that smaller sample sizes are more likely to result in empty networks with non-significant edges (Epskamp, 2016). However, even our smaller (FT) sample was sufficiently powered to establish associations in a comparable range to the VPT group, suggesting that both groups were similarly sensitive to detect associations between nodes. In addition, an explicit comparison of networks was enabled using a bootstrapping procedure, which mitigates the impact of unequal sample sizes.

While this study shows that VPT children are likely at greater risk of displaying ASC traits compared to their FT counterparts, it also highlights that simple group comparisons and focusing on symptom thresholds in isolation, without considering other developmental markers, are not sufficient to fully understand the complexity of the interplay between VPT birth and ASC traits. Our network analysis highlighted underlying constructs leading to ASC behaviors, which may have practical implications, offering a perspective for person-based interventions for VPT children. Indeed, rather than targeting perceived symptoms belonging to specific diagnostic categories, it might be beneficial to focus on their underlying root traits. Future studies can build on these findings by investigating the influence of protective factors that may attenuate the sequelae of VPT birth. We and others have previously shown that supporting parenting and a stimulating home environment promoted resilience against behavioral difficulties in VPT samples (Feldman, 2007; Treyvaud et al., 2009; Ranger et al., 2014; Vinall and Grunau, 2014; Vanes et al., 2021), and may offer useful avenues of further exploration.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by London South East Research Ethics Committee (REC: 19/LO/1940) and London Stanmore Ethics Committee (REC: 18/LO/0048). Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.

ADE, CN, ML, and LDV: conceptualization. LH, DK, ML, and LDV: methodology. ML and LDV: formal analysis and investigation. ML, LH, LDV, and CN: writing—original draft preparation. ML, LDV, LH, PD, ES, SJC, FH, ADE, and CN: writing—review and editing. CN: supervision. ADE, SJC, PD, ES, and CN: funding acquisition. All authors contributed to the article and approved the submitted version.

This work was financially supported by the Medical Research Council (UK) (grant number: MR/S026460/1). The perinatal data analyzed in this study were acquired during independent research funded by the National Institute for Health and Research (NIHR) Program Grants for Applied Research Program (RP-PG-0707-10154).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpsyg.2023.1119196/full#supplementary-material