CARE-Former: Enhancing Prediction of Comorbidities in Rheumatoid Arthritis via Asymmetric Complication Associations

A significant proportion of rheumatoid arthritis patients experience multiple organ-related complications, severely affecting their quality of life. Establishing a dynamic prediction model that adequately considers the associative relationships of complications can contribute to the health management of patients. In this study, we propose to introduce a complication association enhancement module into the dynamic time-series prediction model to improve its performance. This method captures the triggering and dependency relationships between complications, reveals comorbidity interactions, and identifies high-risk complication combinations through clustering. We identified a cluster of comorbidities consisting of interstitial lung lesions, pulmonary hypertension, and cardiovascular disease and achieved better predictive performance than previously proposed methods on the CREDIT dataset (Macro accuracy=0.978 Micro F1=0.755 Micro AUC=0.910). To the best of our knowledge, this is the first study to introduce disease associations in dynamic medical time-series data to enhance early warning of complications in rheumatoid arthritis.