A multimodal embedding model for sepsis data representation - Takeaways - MDSpire

A multimodal embedding model for sepsis data representation

  • By

  • Tuo Liu

  • Yonglin Li

  • Hongyi Chen

  • Naiqing Li

  • Yan Zhang

  • Xuanqi Huang

  • Jin Wang

  • Rui Chen

  • Yuping Zeng

  • Yuntao Liu

  • Danwen Zheng

  • Darong Wu

  • Changdong Wang

  • Tao Yu

  • Xiaotu Xi

  • Zhongde Zhang

  • February 23, 2026

  • 0 min

Share

  • 1

    SepsisDRM is an embedding model that integrates tabular and textual data to enhance patient representation in sepsis research.

  • 2

    Trained on 19,526 sepsis patients, SepsisDRM shows strong generalization across various sepsis-related tasks without specific tuning.

  • 3

    The model stratifies patients into four clinically interpretable phenotypes and predicts 28-day outcomes with high AUC scores.

  • 4

    SepsisDRM is the first embedding model specifically developed for sepsis, establishing a new paradigm for related research.

  • 5

    The complete source code and pre-trained model weights for SepsisDRM are archived on Zenodo to ensure accessibility and reproducibility.

Original Source(s)

A multimodal embedding model for sepsis data representation

  • by Tuo Liu, Yonglin Li, Hongyi Chen, Naiqing Li, Yan Zhang, Xuanqi Huang, Jin Wang, Rui Chen, Yuping Zeng, Yuntao Liu, Danwen Zheng, Darong Wu, Changdong Wang, Tao Yu, Xiaotu Xi, Zhongde Zhang

  • February 23, 2026

Related Content

The Prognostic Value of Red Cell Distribution Width-to-Albumin Ratio for 28-Day Mortality in Sepsis Patients: A Multicenter Analysis Based on the eICU Collaborative Research Database

Acute liver failure and secondary TMA-like phenomenon following SARS-CoV-2 and Mycoplasma pneumoniae co-detection in a child: a case report

Immunometabolic reprogramming and glycolysis-associated signatures in sepsis: insights from single-cell RNA sequencing and machine learning