Treatment of Rheumatoid Arthritis (RA) has improved markedly, with many patients on biologic (b)DMARDs nowadays reaching and maintaining a low level of disease activity. This raises the question whether bDMARD dose can be reduced (tapering) to prevent AE's and high drug costs. Recent studies have shown that tapering of bDMARDs guided by disease activity is indeed possible, but at an increased risk of losing remission status and time in sustained disease control. No clear predictors for successful dose reduction exist. These are likely reasons for the fact that tapering is not performed frequently, although recommended in guidelines.
To improve the risk-benefit ratio of tapering we developed and validated a so-called dynamic prediction model using machine learning to predict the probability of a flare occurring within 3 months (i.e. typical time to next clinic visit). This model uses an indivudual patients' course of disease activity during the process of dose reduction to predict a flare repeatedly at every clinic visit. Simulation of the use of these predictions to guide dose-reduction showed potential to maintain most of the reduction in bDMARD dose with a significant reduction in flares.
Here we will develop a decision aid tool for real-time implementation of a dose reduction strategy using our predictions, compare the impact of this strategy with the current standard of disease activity guided dose reduction in a randomized clinical trial, and use the obtained data to further develop the model to enable the use of home-based monitoring device data on disease activity next to the clinical disease activity measurements. This last step could improve the models performance and usefulness in combination with eHealth solutions for the future.