Background/Purpose: Rheumatic and musculoskeletal diseases (RMD) affect up to one-third of the UK population and are the number one cause of disability and one of the most common reasons for people of working age having an increased number of sick days or unemployment. Patients with RMD experience periods of exacerbation of disease, known flare-ups associated with pain and function limitations for patients. If the disease flare is detected and treated late, this can lead to joint damage and reduced function.

RMD flare-ups are unpredictable, and there is no marker identified for an impending flare. Sudden flares may lead to additional hospitalisations, follow-up visits and decreased quality of life in general.  However there is currently no risk stratification model that can predict patient flare early and accurately in practice. 

We aimed to develop and evaluate a machine learning-based risk stratification model that can predict the risk of future disease flare-ups and guide early and accurate intervention to prevent flare-ups for patients with RMD.

Methods: 142067 patients who have been diagnosed as RMD including RA, PsA and axSpA in the rheumatology department of Royal Berkshire NHS Foundation Trust (UK) from 01/01/2016 to 07/05/2024 were included in the study. There are 58727 patients who had flares and 83340 patients who did not have flares.  We utilized patient blood test results, demographic information, electronic patient-reported outcomes scores, comorbidity, weight and height. We developed a time series dataset and applied Long Short-Term Memory (LSTM) networks and neural networks (NN) to forecast the risk of flare before their future clinics.

Results: We conducted predictive modelling to forecast flare and non-flare events using data collected prior to patients’ upcoming clinic appointments. The dataset was split into training and testing sets based on observations taken 3 months and 6 months before the clinic visit. Our LSTM based model achieved 89.32% accuracy and 0.76 Area Under the Receiver Operating Characteristic Curve (AUROC) predicting flare and non-flare events 6 months before the clinics, and 81.93% accuracy and 0.68 AUROC predicting flare and non-flare events 3 months before the clinics. The results are summarized in Table 1.

Conclusion: Our model can predict future flare and non-flare events with 89.93% accuracy 6 months in advance. Further work includes using advanced data processing methods to handle missing values to improve accuracy, especially for predictions 3 months in advance. If implemented in practice, this flare-risk prediction can aid the patients and clinicians in recognising if they are at a trend of flare-up or reduction in the efficacy of their treatment. With the informed risk prediction presented early and more accurately, the patient will be able to book an early appointment with a specialist so that early interventions can be put in place to prevent flare-ups timely. Clinicians can identify patients at high-risk of flare-up so interventions such as intensifying or changing medication can be in place before flare happens.

« Back to ACR Convergence 2024

ACR Meeting Abstracts - https://acrabstracts.org/abstract/machine-learning-based-risk-stratification-tool-to-predict-early-flare-for-rheumatic-and-musculoskeletal-diseases/