AI-driven models for predicting flare risk in RA patients to guide treatment tapering

In recent years, artificial intelligence found its way to biomedical research, and is particularly popular in precision medicine. Still, making reliable predictions from complex biological data remains challenging, due to data heterogeneity, high dimensionality, and small cohort sizes. An interesting unresolved clinical challenge, is a prediction of the effect of treatment tapering, where doctors try to decide whether patients in remission can safely stop taking the medication. I will look at this problem in context of rheumatoid arthritis and disease flares.

In this talk, I will present a machine learning approach for predicting rheumatoid arthritis flare risk using different data sources, including histology, flow cytometry, spatial transcriptomics, and single-cell RNA sequencing. I will describe how these datasets were analysed and used in the learning process, and show how explainable AI methods were used to identify biologically meaningful features. Finally, I will discuss the current results, ongoing work on flare prediction, and future directions towards improving prediction and supporting clinical decision-making.