Artificial intelligence for identifying new disease clusters in patients with immune-mediated inflammatory conditions: A Proof-of-Concept Study

The overall prevalence of immune-mediated inflammatory conditions (IMIDs) is estimated to affect 1 in 10 people in Europe and the United States.

The most common IMIDs include rheumatoid arthritis (RA), inflammatory bowel disease, systemic lupus erythematosus, psoriasis, and psoriatic arthritis (PsA).

Although the epidemiology of multimorbidity in the RA population is well characterised, it is poorly understood in other IMIDs where only sparse data are available.

The overarching aim of our study is to leverage longitudinal primary care health data from a comprehensive electronic health record database, Clinical Practice Research Datalink (CPRD), to characterise the epidemiology of clusters of medical conditions amongst the IMID population.

There are significant gaps in our understanding of which cluster of medical conditions occur commonly together in the IMID population and whether these clusters vary over the life course.

There is an urgent, yet unmet need to identify clusters of medical conditions in IMIDs accurately, and understanding their life course.

The CPRD comprises 50 million patients, including 15 million currently registered patients in GP practices across the UK, with up to 15-years follow-up.

We will define "multimorbidity" as the presence of two or more diseases in an individual without reference to any index disease, and the study will focus on patients with complex multiple long-term conditions.

We will analyse 40 long-term conditions as outlined by Barnett et al. to identify and interpret key and complex clusters of multimorbidity.

We propose a diverse range of clustering methods as a potential technique to characterise the complex interactions of multimorbidity. We will explore the clusters by age, sex and ethnicity in the IMID population. We will complete this proposed development research project in eight months.

Over this period, we will acquire and harmonise the CPRD data for the application of the artificial intelligence process.

The analyses will be carried out by an experienced team of investigators, including experts in computer science, data analysts, epidemiologists and overseen by experts in IMIDs. In the short-term, our study is timely.

It will shed light on new disease clusters in the IMID population, and this will help to tailor specific interventions to prevent multimorbidity.

In the long-term, the findings will inform the design of a longitudinal study using inception cohorts of IMID patients to understand the occurrence of disease clusters.

Our recent work has shown fluctuations in body mass index may be associated with developing psoriatic arthritis in people with psoriasis.

The plan is to explore if metabolic multimorbidity is associated with the development of PsA and whether the concept could be extended to other IMID's. We have engaged with the patients and the public (PPE) via the Bath Institute for Rheumatic Diseases.

The local NHS R&D has also supported patient and public involvement via their PPE network to inform the design of the study and develop a dissemination plan with clinicians, GPs and members of the public to promote the findings of the research at the end of the study.