PhD Student: Matthew Hutchinson; Partner: BERRI; Supervisors: Prof. Peter Tino and Prof. Stephane De Brito; School: School of Computer Sciences.
Looked-after children suffer disproportionately from mental health struggles later in life as well as other poor life outcomes, due to adversities experienced during childhood affecting their development. To address this issue and improve mental health outcomes and life outcomes for children in care, more needs to be understood about the early indicators of adverse outcomes so that interventions can support at-risk children pre-emptively. Machine learning techniques have been widely applied in recent years to medical diagnosis and the prediction of psychopathology and neurodegenerative disease. The overarching aim of this project is to better understand the key markers indicative of risk or resilience for neurodiversity and later mental health needs of children in care, as well as predicting poor life outcomes by combining machine learning methods and both cross-sectional and longitudinal data.
This project will work alongside BERRI, a clinical tool set up to help support the needs of looked-after children. BERRI tracks the wellbeing of children in care using a questionnaire based around five key themes (Behavior, Emotional wellbeing, Relationships, Risk and Indicators (of neurodevelopment or psychiatric conditions) that is filled in at repeated intervals, so that children can be assessed accurately, and support workers are enabled to appropriately address the needs of these young people. Using data collected by BERRI alongside other existing longitudinal datasets and data collected over the course of the project, I will develop new machine learning techniques to classify trajectories and identify the key indicators of risk and resilience in looked after children, as well as to validate whether the BERRI tool is effective at identifying at-risk children. I will first study impulsivity in children who have experienced maltreatment, with the aim of identifying which facets of state and trait impulsivity are associated with different types of maltreatment, and how impulsive behaviors mediate the relationship between early maltreatment and later mental health. This will be followed by a number of subsequent studies aimed at identifying trajectories and key indicators in an understandable and interpretable way using data from BERRI and understanding how risk and resilience present within the BERRI tool. I will look at wider samples of vulnerable children and children from schools and the general community, as well as children in care, to test the models of risk and resilience and identify any differences across groups.
As well as significant academic impact, contributing a number of papers that will improve understanding of neurodiversity, mental health and machine learning methods, the project will be practically useful and will enable parents and professionals to be able to better recognize early signs of risk or resilience in children in care, which will allow for future interventions to address any concerns pre-emptively and improve the outcomes of at-risk children. The project also has potential to improve screening for CAMHS services, identifying those most at risk, in turn improving the efficiency of the service, which currently is struggling to meet high levels of demand.