Machine learning in cognitive health and disease.

Understanding human behaviour and predicting the outcomes of interventions in health and society is challenged by variability across individuals. Focusing on mental health, decline of socio-cognitive abilities with age is one of the main factors that impair quality of life. However, striking individual variability is observed in ageing: while some older adults may experience rapid decline and impairment (e.g. dementia), others retain cognitive capacities well into old age. Predicting individual socio-cognitive health is of high priority for health economies, given the translational potential for early diagnosis and personalised intervention in clinical practice.

Conventional statistical methods focus on average data, limiting our ability to predict individual behaviour with major ramifications for education and clinical management. In contrast, identifying the source of individual variability in health and disease requires predictive models that mine the interplay of diverse factors (social, genetic, brain). We address this challenge by developing predictive models based on machine learning approaches that synthesise multivariate data and longitudinal measurements from the same individuals to define profiles of individual health and predict changes over time. In our ongoing studies on large-scale population data, we use machine learning approaches to: 1) understand the dynamic interplay of the key factors that underlie individualised ageing profiles, 2) differentiate patients with Mild Cognitive Impairment (i.e. MCI patients at the symptomatic pre-dementia stage) at high and low risk of developing dementia and interrogate the neurocognitive factors that underlie disease progression.

Alan Turing Institute Fellowship to Zoe Kourtzi: TU/B/000095