Causal Representation Learning from Multimodal Clinical Records under Non-Random Modality Missingness
Read the original on arXiv Computation and Language →The paper introduces a causal representation learning framework for multimodal clinical records that addresses non-random missingness of modalities. It combines an MMNAR-aware fusion component, a modality reconstruction module with latent alignment, and a multitask outcome predictor with bias correction. Experiments on MIMIC-IV and eICU datasets show consistent performance gains, achieving up to 4.3% higher AUC for hospital readmission and 2.7% for ICU admission compared to strong baselines.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computation and Language.