Using explainable artificial intelligence to predict and forestall flare in rheumatoid arthritis

Stefano Alivernini,Juan D. Cañete,Jaume Bacardit,Mariola Kurowska-Stolarska

DOI: https://doi.org/10.1038/s41591-024-02818-w

IF: 82.9

2024-02-17

Nature Medicine

Abstract:Current treatments for rheumatoid arthritis aim to achieve and maintain disease remission, which is characterized by the absence of symptoms of inflammation 1 . However, remission is often fragile, with about 50% of patients experiencing disease flares after reducing or stopping medication, representing a burden for both patients and clinicians 1 . Although some predictors of flare exist, they do not confidently predict flare for most patients with rheumatoid arthritis in remission 2 . Recent studies suggest that deconvolution of patients' synovial tissue cellular and molecular signatures may offer more accurate prediction models 3,4 . Patients with rheumatoid arthritis in sustained remission, defined on the basis of clinical symptoms and ultrasound examination, may have residual synovial tissue inflammation that could determine the sustainability of remission 5 . Recent advances in single-cell transcriptomics have uncovered a previously unrecognized heterogeneity of rheumatoid arthritis synovial tissue during remission and identified distinct clusters of synovial macrophages and fibroblasts that are predictive of flare or sustained remission 4,6 .

biochemistry & molecular biology,cell biology,medicine, research & experimental

What problem does this paper attempt to address?

The paper primarily explores how to use explainable artificial intelligence (AI) techniques to predict and prevent disease flare-ups in patients with rheumatoid arthritis. Specifically, the research team aims to address the following key issues:

1. **Creating a comprehensive cellular and molecular atlas**: First, the researchers plan to construct a detailed cellular and molecular atlas of synovial tissue and blood immune cells from rheumatoid arthritis patients who have achieved disease remission through various immunosuppressive treatments. By analyzing tissue and blood samples from these patients, the researchers hope to identify key molecular markers that can predict whether the disease will relapse.

2. **Developing an explainable AI-based prediction tool**: Combining spatial transcriptomics data from synovial tissue and single-cell RNA sequencing data from blood, the researchers will establish a machine learning tool based on explainable AI techniques. The goal of this tool is to predict whether patients in remission will experience a disease flare-up after reducing or stopping treatment, based on specific molecular and clinical features, and to provide recommendations on whether treatment plans should be adjusted accordingly.

3. **Validating the effectiveness of the prediction tool**: The second phase of the research is to validate the effectiveness of the prediction tool in independent, machine learning-guided clinical trials. Eligible patients will undergo synovial tissue biopsy and peripheral blood collection before reducing or stopping treatment to obtain data for the machine learning tool. Subsequently, patients will be randomly assigned to one of two groups: one group will receive treatment adjustment strategies guided by the machine learning tool, while the other group will continue with standard treatment protocols. By monitoring the disease status of patients over 12 months, the effectiveness of the prediction tool in maintaining long-term remission will be evaluated.

In summary, the core objective of this study is to optimize the clinical decision-making process for rheumatoid arthritis patients using advanced biomedical data analysis methods, particularly explainable AI techniques, thereby improving the long-term management of the disease.