CoMPuTE: Complex Multimorbidity Phenotypes, Trends, and Endpoints

Background Multimorbidity places a substantial burden of ill-health on individuals, and a substantial burden of health and social care need on the NHS and society.

A key gap in our understanding is how individual health trajectories – the patterns of disease, medications, biomarkers, health behaviours, and other characteristics – evolve over time, and how these trajectories relate to patient outcomes.

A patient s age and diagnosed diseases may offer some insights into their risks of further complications, but much more information is collected in UK healthcare which could be used to identify their biological age; that is, the effective age of a person (or body system) based not only on a snapshot of their health status, but on a more complete trajectory of health.

Longitudinal healthcare data present considerable challenges to conventional medical statistics due to high dimensionality, sporadic and error-prone measurements, and a high frequency of missing values.

Contemporary machine learning and AI methods offer the opportunity to meet these challenges and infer information about the latent processes of ageing, health and disease.

With such tools, it will be possible to better identify clusters of conditions, and to better organise healthcare resources in order to prevent and manage multimorbidity.

Aims and Objectives Development Award Work plan This Development Award will prototype statistical and AI tools to make use of longitudinal healthcare data to better prevent, manage and organise health services for multimorbidity.

Using UK primary care data from the QResearch database for patients in two age groups (18-44 and 65-84-years), we will: Construct Bayesian models of latent trajectories of health and biological ageing; Use these models with AI-based clustering to identify multi-system groups of conditions (with a focus, for this pilot work, on the 65-84-year age group); and Identify patterns and trends in multimorbidity and in biological ageing by socioeconomic status, ethnicity, region, and other social determinants of health.

Delivery Timelines The convening of Steering and Stakeholder groups and the data approval and specifications and will prior to start date.

The first four months will be used to develop in parallel the latent trajectories (65-84-years) and AI-based clustering.

The second four months will be used to explore the latent trajectories for the 18-44 and the AI-based clustering for the 65-84. At the end of this grant, we will submit a proposal for a full Programme to the NIHR.

Anticipated Impact and Dissemination Building on this Development Award, we aim to achieve impact through a Research Collaboration investigating patterns and trends of complex clusters, potential mechanisms, economic costs, health inequalities as well as the patient experience and pathways to care.

We plan to disseminate evidence, (via conferences, workshops, publications and social media), on how to predict multimorbidities and pinpoint areas of care, geography, socioeconomic status which need critical attention to reduce the substantial burden of multimorbidity on individuals and the health and social care system.

Our ultimate goal is to contribute to developing AI capability in the NHS by devising a user-friendly decision support tool for complex multimorbidity.