Machine Learning Driven Diagnostic and Prognostic Tools Targeting Rheumatology

Rheumatic diseases are autoimmune and inflammatory disorders with multifactorial nature characterized by a wide variety of symptoms such as joint pain, joint deformity, impaired physical function and comorbidities, including an increased risk of cardiovascular pathologies. The presence of chronic pain is also responsible for a significant decrease in quality of life for patients. Rheumatic diseases can have gestation periods that precede clinically relevant onset by many years, and disease development includes genetic and environmental factors as well as systemic autoimmunity. Instead, early diagnosis and immediate treatment of rheumatic diseases have been identified as key elements for a possible remission of the disease. Therefore, it is important to develop efficient tools for the early identification of the disease and the assessment of its evolution on the basis of the patient's clinical features. In addition, the determination of the risk of comorbidities is also very important for these patients, particularly with regard to the determination of the risk of related cardiovascular diseases. However, current cardiovascular risk predictors mostly fail in identifying events when applied to rheumatic patients due to the synergistic effect of many factors and the complex clinical scenario of this type of subjects. In this context, machine learning (ML) algorithms are proving to be a valuable clinical decision support as they are better suited than traditional algorithms to deal with non-linearly distributed features and are able to find complex relationships within many predictor variables. In this work, ML was applied to investigate on crucial clinical issues concerning early disease diagnosis and cardiovascular risk assessment in rheumatic patients. Firstly, a ML-based model was developed/tailored to diagnose rheumatic patients and classify them according to their specific inflammatory disease; secondly, several ML algorithms were tested as prognostic predictors of cardiovascular events in the case of rheumatic patients and were compared with the Framingham Risk Score approach. In addition, ML methods were used to investigate the predictive power of different clinical markers and their weight in risk evaluation. In particular, it was highlighted which patient characteristics are most relevant for the prediction of cardiovascular risk in systemic lupus erythematosus, psoriatic arthritis and ankylosing spondylitis. The identification of new biomarkers specific for autoimmune rheumatic diseases is crucial for improving prognostic performance and finding hidden relationships between biomarkers that may also help characterize one disease over another. This aspect is crucial in rheumatic diseases since an early identification of the possible evolution of the disease can help to promptly choose the appropriate therapeutic strategy.