Work Package 2: Predictive markers

To understand who is at risk of cognitive complaints and decline and who will benefit the most from a lifestyle intervention to reduce cognitive decline, we need markers. Preferably, these markers are non-invasive, such that they can be used easily in a home setting for self-care, and open to intervention. Ultimately, these markers could even be used to give personalised feedback of intervention necessity and response in a home setting.

Past and ongoing initiatives generated promising results and insights on the existence of non-invasive markers for the study of cognitive decline. However, the complexity of cognitive decline, the variability of symptoms, risk factor exposure and interactions among these risk factors provide a strong rationale for addressing the problem by modelling cognitive decline in a broader perspective.

In WP2 we aim to find easy, cost-effective, non-invasive ways to predict cognitive decline in an early stage and to predict the response to a lifestyle intervention response to prevent cognitive decline. Specifically, we will combine non-invasive modifiable risk and protective factors in more reliable scoring tools through predictive computational models, to quantify risk of cognitive complaints and decline on a personal level, and to analyse the effect of an intervention in relatively short periods (span of a few years). We will use the developed scoring tools to detect multi-modal, non-invasive markers.

Marco Loog
Academic leader
Radboud University

Jesse Krijthe
Academic leader
Radboud University

Tom Houslay
Impact leader
Reckitt

Design / methods

We will develop and test predictive computational models by combining advanced machine learning techniques with conventional statistical approaches that have been employed in most current studies for precision medicine, like multi-factor analysis, mixed-linear models, and Cox regression. Also, we will investigate the use of multi-modal machine learning in order to exploit heterogenous and large sources (modalities) of data. In particular, we will use advanced artificial intelligence (AI) methods based on deep learning to build computational models that integrate different modalities for predicting Mild Cognitive Impairment (MCI) onset in existing data sets, and cognitive function in response to the lifestyle intervention in WP1.

Societal relevance of outcomes

The relevant markers from this intervention identified within WP2 will not only be beneficial for health care providers to understand who is/will be responding, but can also be provided to individuals for self-management as motivating, intermediate measures when changes in subjective cognitive decline are not yet observed.

Involved partners

Academic partners:

• Radboud University; Faculty of Science, Institute for Computing and Information Sciences, Dept. of Data Science, Donders Institute for Brain, Cognition and Behaviour
  • Marco Loog (academic leader)
  • Jesse Krijthe (academic leader)
  • Joukje Oosterman
  • Wieske de Swart
  • Wouter Kant
  • Parisa Naseri
• Radboud University Medical Center; Dept. Geriatrics, Alzheimer Center
  • Jurgen Claassen
• Maastricht University; Dept. Psychiatry and Neuropsychology, Alzheimer Center Limburg, School for Mental Health and Neuroscience (MHeNs)
  • Martin van Boxtel
  • Kay Deckers
  • Sebastian Köhler
• AmsterdamUMC, location VUmc; Dept. Neurology, Alzheimer Center Amsterdam

Industrial partner:

• Reckitt
  • Tom Houslay (impact leader)

News