Integrating multiple data types to identify novel brain-behaviour subgroups in very preterm born children

• In Event: Poster Session 11
  In Poster Session: PS 11 Topic - Developmental Psychopathology

Abstract

Preterm birth has been associated with altered brain development and poorer cognitive and behavioural outcomes. However, the heterogeneity in outcomes among children who were born very preterm makes it challenging to identify the most vulnerable subgroups. In this study we use a data-integration and clustering approach to stratify individuals into brain-behavioural profiles. We studied 198 very preterm children (born at <33 weeks of gestation) who received multi-modal Magnetic Resonance Imaging (MRI) at term equivalent age and were followed-up for neuropsychological assessments at age 4.2 to 7.2 years (median: 4.6), as part of the Evaluation of Preterm Imaging Study (Edwards et al, 2018). Using Similarity Network Fusion (Wang et al, 2015), a network based data integration approach, we integrated the following data types: 1) environmental and clinical factors (Index of Multiple Deprivation, gestational age and clinical risk factors), 2) childhood executive function scores (Behavioural Rating Inventory of Executive Function subscale scores: Inhibit, Shift, Emotional Control, Working Memory and Plan/Organise) and 3) childhood socio-emotional outcomes (scores from the Empathy Questionnaire and Social Responsiveness Scale). We then aimed to evaluate whether resultant data-driven subgroups differed on out-of-model temperament scales (Early Childhood Behavioural Questionnaire), psychiatric internalizing and externalizing symptoms (Strengths and Difficulties Questionnaire), IQ, cognitive stimulation at home (Homes questionnaire), neonatal structural brain volumes (voxel-wise Tensor-Based-Morphometry), and neonatal white matter tract indices from the corticospinal tracts, the corpus callosum, the superior and inferior longitudinal, uncinate and inferior fronto-occipital fasciculi (Tract-Specific-Analysis of diffusion-MRI). Eigen-Gap and Rotation Cost estimated an optimal cluster number of two and the Silhouette width average score was 0.79. The two resultant subgroups included 1) a high-risk subgroup characterized by poorer executive function, IQ, socio-emotional and psychopathology outcomes and higher clinical risk, and 2) a high functioning subgroup, characterized by better cognitive and psychopathology outcomes and lower clinical risk. Upon comparing our two resultant subgroups on out-of-model brain measures, we found significantly larger cerebellar regions in the high functioning subgroup compared to the high-risk subgroup (p<0.025; correcting for gestational age, post-menstrual age at scan and sex; Threshold-free Cluster Enhancement and Family Wise Error correction applied). We found no significant differences on any of the white matter tract measures (p>0.05). In conclusion, applying data-driven integration and clustering approaches can help identify at-risk subgroups which can be characterised by specific neurobiological markers as early as term-equivalent age. This can, in turn, guide personalised behavioural interventions aimed at improving psychiatric and cognitive outcomes in vulnerable children.

Authors

• Laila Hadaya, King’s College London
  Presenting Author

• Konstantina Dimitrakopoulou, NIHR BRC, Guy's and St Thomas' NHS Trust, and King's College London
  Non-Presenting Author

• Diliana Pecheva
  Non-Presenting Author

• Dana Kanel, National Institute of Mental Health; University of Maryland, College Park
  Non-Presenting Author

• David Edwards, King's College London
  Non-Presenting Author

• Serena Counsell, King's College London
  Non-Presenting Author

• Mansoor Saqi, NIHR BRC, Guy's and St Thomas' Trust, and King's College London
  Non-Presenting Author

• Dafnis Batalle, King's College London
  Non-Presenting Author

• Chiara Nosarti, King's College London
  Non-Presenting Author