Causal Representation Learning for Predicting Autoimmune Disease Progression from Longitudinal Multimodal Clinical Data

Autoimmune diseases pose a persistent challenge to clinical decision-making due to their heterogeneous presentations, episodic flare-ups, and complex pathophysiological mechanisms. Existing deep learning models frequently rely on correlation-driven patterns, which often lack robustness and interpretability in real-world clinical settings. This study proposes a novel causal deep learning framework designed to model and predict the progression of autoimmune diseases using longitudinal multimodal clinical data. The proposed approach integrates a temporal transformer architecture with a structural causal model (SCM)-guided counterfactual inference module to capture latent disease mechanisms while mitigating spurious correlations introduced by observational confounders. The model is trained and evaluated on a real-world cohort consisting of patients diagnosed with systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA), incorporating electronic health records (EHRs), laboratory biomarkers, and immunological profiles. Empirical results demonstrate that the proposed framework outperforms conventional deep learning baselines in predicting disease transitions and flare-up events. In addition, interventional and counterfactual evaluation protocols are employed to validate the model’s causal consistency and interpretability. This work contributes to the advancement of clinically trustworthy AI systems by enabling robust, explainable, and data-efficient disease trajectory modeling in the autoimmune domain.