<p>Respiratory diseases such as chronic obstructive pulmonary disease, asthma, and pulmonary fibrosis remain leading causes of global morbidity and mortality, necessitating intelligent, real-time monitoring systems. Traditional diagnostic approaches are largely reactive and fail to capture dynamic physiological variations. To address these limitations, this study proposes a novel fusion framework that integrates digital twin technology with an efficient sequence-modelling architecture based on the Mamba Transformer. The proposed system constructs a patient-specific digital twin by aggregating multimodal data, including electronic health record data, spirometry signals, wearable sensor data, and medical imaging. A selective state-space Mamba model is employed to capture long-range temporal dependencies in irregular physiological time series with linear computational complexity. The framework is evaluated on a hybrid dataset comprising 1000 patients from MIMIC-IV and SpiroHealth cohorts with curated spirometry subset derived from the Figshare Raw Spirometry dataset. Experimental results demonstrate promising performance under internal validation settings with an accuracy of 96.2%, sensitivity of 94.7%, specificity of 95.8%, and F1-score of 95.1%. Additionally, the model achieves a mean absolute error of 3.6% in pulmonary function test prediction and enables early exacerbation prediction with 87.2% accuracy up to seven days in advance. The proposed system enhances predictive capability, computational efficiency, and personalization in respiratory healthcare. Future work includes large-scale clinical validation and integration with federated and edge AI systems.</p>