A feasibility study of a machine-learning-based risk stratification tool for enhancing the rheumatology referral pathway

Summary

An evaluation of a machine-learning-based risk stratification tool assessed the acceptance and potential barriers to adopting this tool, designed to support the rheumatology referral pathway. The study involved gathering insights from GPs, rheumatology registrars and rheumatology consultants. The findings of our feasibility study will assist RMD-Health developers in finalising the functionality of the tool. Further evidence is required for more widespread adoption.

What is the challenge?

The current rheumatology referral pathway faces difficulties to correctly diagnose patients and differentiating between inflammatory arthritis (IA) and non-inflammatory conditions (NIC). Patients with IA are difficult to diagnose due to nonspecific symptoms; hence, there can be inaccuracies in diagnosis and an inefficient allocation of resources as a result.

What did we do?

The University of Reading, alongside Royal Berkshire NHS Foundation Trust (RBFT), developed RMD-Health, a machine learning risk stratification tool to assist in detecting and differentiating between IA and NIC at the point of referral and benchmarked against confirmed diagnoses. The model utilised a large language model (LLM) and scalable multimodal machine learning to provide an AI-based risk stratification tool that aimed to uncover hidden patterns for RMD (rheumatic and musculoskeletal diseases). It was designed to read data available at referrals, predict the likelihood of an individual patient having either IA or NIC and explain key risk factors of the predictions.

HIOTV explored stakeholder perspectives of this tool within the current referral pathway, the perceived utility of RMD-Health, and the practicalities of its implementation. Using the Lean Assessment Process (LAP) methodology, semi-structured interviews were conducted with 15 clinicians across primary and secondary care. Topics included current challenges, potential benefits, barriers to adoption and the evidence required to support adoption.

What has been achieved?

The feasibility study conducted by us demonstrated considerable agreement across stakeholders that RMD-Health could improve the referral pathway. The tool was seen as particularly beneficial for GPs, offering decision support and potentially reducing unnecessary referrals. Stakeholders emphasised the importance of conducting future pilot studies to demonstrate usability, diagnostic accuracy and cost-effectiveness across diverse NHS settings.

What they said

“The feasibility study report demonstrates that RMD-Health is a methodologically robust and clinically credible innovation with clear potential to improve referral triage, reduce clinician workload, and support earlier, more accurate decision-making across primary and secondary care. Underpinned by a strong health economic framework, the findings provide compelling evidence that these clinical benefits can translate into meaningful system-level efficiencies, making a strong case for further piloting and wider adoption of RMD-Health within NHS rheumatology services. Importantly, the study also highlights the evidence, integration, and trust requirements that now shape our next phase of real-world evaluation and implementation.”

Professor Weizi (Vicky) Li (University of Reading) and Professor Antoni Chan (RBFT)

Next steps

The findings of the feasibility study conducted we conducted will assist the University of Reading and RBFT in finalising the functionality of the tool. The tool will then need to be piloted and utilised for further evidence generation towards the adoption in the rheumatology referral pathway at the point of referrals.