BEACON: below average cognitive ability in childhood - outcomes and prediction
Loading...
Files
Full Text E-thesis
Date
2025
Authors
Bowe, Andrea
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
Background
Approximately 16% of children have below average cognitive ability. Around 1-2% have an intellectual developmental disorder (IDD), but a larger proportion, approximately 14%, have cognitive ability between 1 and 2 standard deviations below the mean. The outcomes for this larger group have received little attention in scientific literature. For both groups, early intervention may improve cognitive outcomes, particularly if commenced in the first 1,000 days of life. The aim of this thesis was to explore the impact of below average cognitive ability on outcomes in childhood and to examine current and alternative methods, namely the application of machine learning to clinical and epidemiological data, for the early prediction of below average cognitive ability in childhood.
Structure and methods
The thesis begins with an introductory chapter (Chapter 1), followed by a statement of aims and objectives (Chapter 2). Six original studies are then presented (Chapters 3-8). The relationship between below average cognitive ability and emotional- behavioural function in childhood is examined using generalised linear mixed models and logistic regression (Chapter 3). The causal relationship between below average cognitive ability and the early experience of school is investigated using logistic regression and directed acyclic graphs (Chapter 4). The predictive validity of a widely used developmental screening tool, the Ages and Stages Questionnaire, is then investigated (Chapter 5), before a series of papers examining the application of a range of machine learning techniques to clinical and epidemiological data for predicting below average cognitive ability in childhood among a general population (Chapters 6 & 7), and in infancy among a very preterm population (Chapter 8). The studies are based on data from the Growing up in Ireland Infant Cohort (Chapters 3, 4, & 7 ), the Cork BASELINE birth cohort (Chapters 5 & 6), and the Swedish Neonatal Quality Register (Chapter 8).
Results
Having cognitive ability between 1 and 2 standard deviations (SDs) below the mean at age 3, compared to having average cognitive ability, was independently associated with increased odds of experiencing a clinically significant emotional-behavioural problem between age 3 and 9 years (AOR 1.39, 95% CI 1.17-1.66, p<0.001). These children, on average, experienced a significant worsening in their emotional- behavioural function between age 5 and 9 years, the time coinciding with commencing formal education in Ireland (mean difference 0.48, 95% CI 0.06 to 0.90, p=0.016), compared to peers with average cognitive ability who had no significant change (mean difference -0.06, 95% CI -0.19 to 0.07, p=0.488). (Chapter 3) Compared to their peers with average cognitive ability, children whose cognitive ability was more than 1 SD below the mean at age 5 had significantly higher odds, at age 9 years, of experiencing low self-concept (AOR 1.20, 95% CI 1.02-1.42); of hardly ever or never being interested, motivated, and excited to learn (AOR 2.29, 95% CI 1.70-3.10); and of experiencing teacher-reported emotional-behavioural difficulties (AOR 1.34, 95% CI 1.10-1.63). (Chapter 4)
The Ages and Stages questionnaire, a routinely used developmental screening tool, when performed at 24 or 27 months would only detect 21% of infants who would go on to have IQ more than 1 SD below the mean at age 5. (Chapter 5) Using machine learning techniques to rebalance data and perform feature selection, on internal validation, a random forest model trained on the BASELINE cohort could correctly predict at birth, using 11 perinatal features, the cognitive outcome at age 5 of 95% of children. It achieved a sensitivity of 89% and a specificity of 98% for identifying those with an IQ less than 90 at age 5 years. External validation was not performed. (Chapter 6) On external validation using unseen data, a random forest model trained on the Growing Up in Ireland cohort, using 15 participant-reported features in the first year of infant life, could correctly predict the cognitive outcome at age 5 for 87% of infants. However, the accuracy was largely driven by a high specificity of 90%, with a sensitivity of 40%. (Chapter 7) In a high risk cohort of very preterm infants from the Swedish Neonatal Quality register, a logistic regression model containing 26 features could identify, at discharge from the neonatal unit, 93% of infants who experienced cognitive delay at 2-year follow up, with a specificity of 46%. (Chapter 8)
Conclusion
Children with below average cognitive ability, including those whose cognitive ability lies between 1 and 2 standard deviations below the mean, are more likely to experience emotional-behavioural difficulties, low self-concept, and poor classroom engagement in childhood. Early intervention could improve outcomes for these children. Among a general paediatric population, applying machine learning to readily available clinical and epidemiological data, may be a step toward an early, scalable, risk stratification tool, to identify the children who would benefit most. The predictive performance of the model developed in this thesis is not yet good enough for implementation. Among a very preterm population, the model developed in this thesis shows great promise for early identification of those at risk of later cognitive delay.
Description
Keywords
Cognitive ability , Childhood , Prediction , Machine learning , Emotional-behavioural function
Citation
Bowe, A. 2025. BEACON: below average cognitive ability in childhood - outcomes and prediction. PhD Thesis, University College Cork.