Prediction of rheumatoid arthritis flares after treatment withdrawal

Increasingly, rheumatoid arthritis patients in remission discontinue treatment to mitigate risks associated with long-term drug exposure. Yet, shortly after stopping therapy, approximately half of these patients experience a disease flare. Without reliable predictors, deciding who can safely discontinue medication poses a significant clinical challenge.

In this talk, I will give a brief overview of my PhD dissertation, where I investigate whether multi-source biological data from the BIO-FLARE cohort can improve the prediction of flares. I will discuss the data preprocessing challenges and the algorithm I developed (FlowTree) to automatically detect immune cell populations, including comparison of its results on several benchmarks. Then, I will discuss the flare prediction itself and compare the best models against the clinical baseline. Finally, I will show how I explored the best model explanations to identify distinct disease subtypes, and how I linked them to specific inflammatory patterns using the enrichment analysis.