Artificial Intelligence and Multimorbidity: Clustering in Individuals, Space and Clinical Context (AIM-CISC)

Multimorbidity is a major challenge to health systems internationally, but previous research examining how morbidities cluster within individuals has been limited in scope, has not been replicated, largely ignores social and geographical context, and has seldom translated into improvements in care. This research aims to use artificial intelligence and state-of-the art data science, social science, genomics, and health service research methods to understand clustering of morbidities within individuals, within communities, and in key clinical contexts. Objective 1: To use unsupervised and supervised machine learning including multilayer networks to examine clustering of morbidities in individuals. We will identify morbidity clusters that are consistent, operationalisable, stable, and reproducible across multiple methods and datasets. We will use these clusters in subsequent objectives along with clusters defined a priori based on theory or domain knowledge. Objective 2: To examine the genomics of morbidity clustering in individuals to support validation of objective 1 cluster solutions, and examine aetiology. We will map objective 1 clusters into UK Biobank data as traits and identify genetic loci that explain why individuals belong to one or more clusters, using polygenic models to predict an individual s cluster membership. This will underpin future mechanistic and potentially interventional research, based on mapping potential causal pathways of morbidity clustering. Objective 3: To use AI and machine learning to examine clustering of complex multimorbidity in communities and places. We will enrich existing locality data using novel geo-profiling based on news and social media free-text data, and examine how multimorbidity clusters spatially using existing social science methods for spatial analysis, and machine learning including multilayer networks. Objective 4: To develop and optimise AI tools to underpin complex interventions to reduce adverse events in people with complex multimorbidity and polypharmacy. We will improve ascertainment of serious adverse events using natural language processing to extract information from free-text data, and use knowledge graphs, and machine learning including multilayer networks to understand and predict adverse events. Quantitative modelling will be paralleled by complex intervention development work to design two interventions with AI prediction tools at their heart (one targeting people with complex multimorbidity and polypharmacy in the community, the other targeting the same people at the point of hospital admission). Our team has interdisciplinary expertise, and includes clinical and genetic researchers studying complex multimorbidity, public partners, social scientists researching wider social and spatial determinants of health and care, and informatics and data science academics with AI expertise across multiple domains including natural language processing, machine-learning including multilayer network analysis, and applied AI. We will actively collaborate with the AIM Research Support Facility and other AIM teams in order to resolve conceptual and methodological problems in this field, including through mutual reproduction of findings and reuse of methods and models. As well as conventional academic outputs, we are committed to engaging multiple stakeholders including the public, clinicians and public health professionals, managers and policymakers. We will ensure that all of our data assets are appropriately shared to be as findable, accessible, interoperable and reusable as possible.