New framework tackles missing data in multimodal clinical records

Researchers have developed a novel causal representation learning framework designed to handle missing data in multimodal clinical records. This framework addresses the challenge of non-random modality missingness, where the availability of data like clinical notes, structured data, or imaging is influenced by clinical decisions. The proposed system includes a component for fusing modalities while accounting for missingness patterns, another for preserving modality information in learned representations, and a multitask prediction model to correct for residual bias. Evaluations on MIMIC-IV and eICU datasets demonstrated significant improvements in predicting hospital readmission and ICU admission. AI

IMPACT This research could improve the accuracy of AI models in healthcare by enabling better use of incomplete patient data.

RANK_REASON The cluster contains a research paper detailing a new methodology for handling missing data in clinical records. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• MIMIC-IV
• Zihan Liang

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →