Causal Representation Learning from Multimodal Clinical Records under Non-Random Modality Missingness

2025-09-21

AI Summary

Clinical records often exhibit multivariate missing-not-at-random (MMNAR) patterns—where missingness in modalities such as structured vital signs, X-ray images, and radiology reports is driven by clinical decision-making—leading to biased representation learning and degraded generalization. To address this, we propose a causal representation learning framework explicitly designed for MMNAR settings. First, we model the missingness mechanism to approximate the underlying clinical assignment process. Second, we integrate a large language model with a medical image encoder, incorporating contrastive reconstruction and bias-correction modules to disentangle confounding factors and align causal signals across modalities under missing data conditions. Evaluated on MIMIC-IV and eICU, our method significantly improves downstream predictive performance: readmission and ICU admission prediction AUC increase by 13.8% and 13.1%, respectively. The framework further demonstrates robustness to distribution shifts and fairness across patient subgroups.

Technology Category

• Machine Learning: Causal Learning
• Computer Vision: Multi-modal Vision
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Clinical notes contain rich patient information, such as diagnoses or medications, making them valuable for patient representation learning. Recent advances in large language models have further improved the ability to extract meaningful representations from clinical texts. However, clinical notes are often missing. For example, in our analysis of the MIMIC-IV dataset, 24.5% of patients have no available discharge summaries. In such cases, representations can be learned from other modalities such as structured data, chest X-rays, or radiology reports. Yet the availability of these modalities is influenced by clinical decision-making and varies across patients, resulting in modality missing-not-at-random (MMNAR) patterns. We propose a causal representation learning framework that leverages observed data and informative missingness in multimodal clinical records. It consists of: (1) an MMNAR-aware modality fusion component that integrates structured data, imaging, and text while conditioning on missingness patterns to capture patient health and clinician-driven assignment; (2) a modality reconstruction component with contrastive learning to ensure semantic sufficiency in representation learning; and (3) a multitask outcome prediction model with a rectifier that corrects for residual bias from specific modality observation patterns. Comprehensive evaluations across MIMIC-IV and eICU show consistent gains over the strongest baselines, achieving up to 13.8% AUC improvement for hospital readmission and 13.1% for ICU admission.

Problem

• Addressing non-random missing modalities in clinical records
• Learning patient representations from incomplete multimodal data
• Correcting bias from clinical decision-making in representation learning

Innovation

• Causal framework leveraging observed data and missingness
• MMNAR-aware fusion conditioning on missingness patterns
• Multitask prediction with rectifier correcting residual bias