Fig 1.
The structure of the model proposed in this study.
It is an end-to-end pipeline integrating Transformer-based text embedding, multi-task learning, and transfer learning modules.
Fig 2.
The data preprocessing process in this study (This figure details the specialized clinical text normalization workflow, including medical concept recognition, temporal marker preservation, and ontology-based noise reduction to prepare EHR data for Transformer processing).
Fig 3.
The principle of Transformer-based text embedding process in this study.
Fig 4.
The multi-task learning framework in this study.
Fig 5.
The transfer learning framework in this study (This figure explains the two-phase optimization with clinical domain adapters and confusion loss, showing how pretrained ClinicalBERT weights are adapted to target institutions while mitigating catastrophic forgetting through curriculum learning).
Table 1.
Logistic regression performance across datasets.
Table 2.
Random forest performance across datasets.
Table 3.
CNN performance across datasets.
Table 4.
Robustness evaluation of our model across datasets.
Fig 6.
Comparative analysis reveals fundamental differences in how each baseline handles dataset variations.
Table 5.
Logistic regression temporal performance.
Table 6.
Random forest temporal performance.
Table 7.
CNN temporal performance.
Table 8.
Model performance across different temporal periods.
Fig 7.
Comparative temporal analysis reveals fundamental architectural differences in handling concept drift.
Table 9.
Training vs. validation performance on MIMIC-III dataset.
Table 10.
Ablation study results on MIMIC-III dataset.