Abstract
Sepsis research has long been constrained by limited labeled data and models designed for specific tasks that primarily rely on tabular inputs, overlooking the valuable insights contained in clinical text. To address these limitations, we propose the Sepsis Data Representation Model (SepsisDRM), an embedding model that jointly processes tabular and textual data to capture comprehensive patient representations. Trained on a dataset comprising 19,526 sepsis patients, SepsisDRM demonstrates strong generalization across diverse sepsis-related tasks without task-specific tuning. It effectively stratifies patients into four clinically interpretable phenotypes and achieves robust performance in predicting 28-day outcomes, with AUC scores of 0.92, 0.94, and 0.78 on retrospective, prospective, and external datasets, respectively. As the first embedding model developed specifically for sepsis, SepsisDRM establishes a novel paradigm for sepsis research and offers a promising approach for studies in other fields that involve the integration of both tabular and textual data.
Data availability
GDHCM dataset used to train SepsisDRM, and GDHCM retrospective dataset, GDHCM prospective dataset, SYSMH external validation dataset used to test SepsisDRM, are not publicly available due to its potentially identifiable nature.
Code availability
To ensure long-term accessibility and facilitate reproducibility, the complete source code of SepsisDRM, along with the synthetically generated toy datasets and pre-trained model weights, has been archived on Zenodo with the persistent identifier https://doi.org/10.5281/zenodo.17828465. The repository includes detailed documentation and environment configuration files.
References
Evans, L. et al. Surviving sepsis campaign: International guidelines for management of sepsis and septic shock 2021. Intensiv. Care Med. 47, 1181–1247 (2021).
Levy, M. M. et al. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Crit. Care Med. 31, 1250 (2003).
Rudd, K. E. et al. Global, regional, and national sepsis incidence and mortality, 1990–2017: analysis for the Global Burden of Disease Study. Lancet (Lond., Engl.) 395, 200–211 (2020).
Vincent, J.-L. et al. Sepsis in European intensive care units: results of the SOAP study. Crit. Care Med. 34, 344–353 (2006).
Leligdowicz, A. et al. Association between source of infection and hospital mortality in patients who have septic shock. Am. J. Respir. Crit. Care Med. 189, 1204–1213 (2014).
Antonucci, E. et al. Myocardial depression in sepsis: from pathogenesis to clinical manifestations and treatment. J. Crit. Care 29, 500–511 (2014).
Seymour, C. W. et al. Derivation, validation, and potential treatment implications of novel clinical phenotypes for sepsis. JAMA 321, 2003–2017 (2019).
Zhang, Z. et al. Exploring disease axes as an alternative to distinct clusters for characterizing sepsis heterogeneity. Intensiv. Care Med. 49, 1349–1359 (2023).
Guo, F. et al. Clinical applications of machine learning in the survival prediction and classification of sepsis: coagulation and heparin usage matter. J. Transl. Med. 20, 265 (2022).
Yan, F. et al. Association between the stress hyperglycemia ratio and 28-day all-cause mortality in critically ill patients with sepsis: a retrospective cohort study and predictive model establishment based on machine learning. Cardiovasc. Diabetol. 23, 163 (2024).
Ibarra-Estrada, M. et al. Early adjunctive methylene blue in patients with septic shock: a randomized controlled trial. Crit. Care 27, 110 (2023).
Gabarre, P. et al. Albumin versus saline infusion for sepsis-related peripheral tissue hypoperfusion: a proof-of-concept prospective study. Crit. Care 28, 43 (2024).
Moor, M. et al. Foundation models for generalist medical artificial intelligence. Nature 616, 259–265 (2023).
Xu, H. et al. A whole-slide foundation model for digital pathology from real-world data. Nature 630, 181–188 (2024).
Pai, S. et al. Foundation model for cancer imaging biomarkers. Nat. Mach. Intell. 6, 354–367 (2024).
Lu, M. Y. et al. A visual-language foundation model for computational pathology. Nat. Med. 30, 863–874 (2024).
Christensen, M., Vukadinovic, M. & Yuan, N. Vision-language foundation model for echocardiogram interpretation. Nat. Med. 30, 1481–1488 (2024).
Desautels, T. et al. Prediction of early unplanned intensive care unit readmission in a UK tertiary care hospital: a cross-sectional machine learning approach. BMJ Open 7, e017199 (2017).
Shashikumar, S. P., Shah, A. J., Li, Q., Clifford, G. D. & Nemati, S. A deep learning approach to monitoring and detecting atrial fibrillation using wearable technology. In 2017 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI), 141–144 (IEEE, 2017).
Huang, K., Altosaar, J. & Ranganath, R. Clinicalbert: Modeling clinical notes and predicting hospital readmission. arXiv preprint arXiv:1904.05342 (2019).
Alsentzer, E. et al. Publicly available clinical bert embeddings. In Proceedings of the 2nd Clinical Natural Language Processing Workshop, 72–78 (2019).
Cheerla, A. & Gevaert, O. Deep learning with multimodal representation for pancancer prognosis prediction. Bioinformatics 35, i446–i454 (2019).
Kline, A. et al. Multimodal machine learning in precision health: a scoping review. NPJ Digit. Med. 5, 171 (2022).
Fleuren, L. M. et al. Machine learning for the prediction of sepsis: a systematic review and meta-analysis of diagnostic test accuracy. Intensiv. Care Med. 46, 383–400 (2020).
Zuin, G. et al. Prediction of SARS-CoV-2 positivity from million-scale complete blood counts using machine learning. Commun. Med. 2, 72 (2022).
Liu, Y. et al. Roberta: a robustly optimized BERT pretraining approach. arXiv:1907.11692 (2019).
Song, Y. et al. Xuebijing injection versus placebo for critically ill patients with severe community-acquired pneumonia: a randomized controlled trial. Crit. Care Med. 47, e735–e743 (2019).
Li, C. et al. The current evidence for the treatment of sepsis with xuebijing injection: bioactive constituents, findings of clinical studies and potential mechanisms. J. Ethnopharmacol. 265, 113301 (2021).
Sinha, P. et al. Identifying molecular phenotypes in sepsis: an analysis of two prospective observational cohorts and secondary analysis of two randomised controlled trials. Lancet Respir. Med. 11, 965–974 (2023).
G, E. et al. Sepsis-induced endothelial dysfunction drives acute-on-chronic liver failure through angiopoietin-2-HGF-C/EBPβ pathway. Hepatology (Baltimore, MD) 78, (2023) https://pubmed.ncbi.nlm.nih.gov/36943063/.
Cheng, C. et al. Pharmacologically significant constituents collectively responsible for anti-sepsis action of XueBiJing, a Chinese herb-based intravenous formulation. Acta Pharmacol. Sin. 45, 1077–1092 (2024).
Rey, C. et al. Procalcitonin and c-reactive protein as markers of systemic inflammatory response syndrome severity in critically ill children. Intensiv. Care Med. 33, 477–484 (2007).
Pierrakos, C. & Vincent, J.-L. Sepsis biomarkers: a review. Crit. Care 14, R15 (2010).
Cawley, G. C. & Talbot, N. L. On over-fitting in model selection and subsequent selection bias in performance evaluation. J. Mach. Learn. Res. 11, 2079–2107 (2010).
Kenward, M. G. & Carpenter, J. Multiple imputation: current perspectives. Stat. Methods Med. Res. 16, 199–218 (2007).
MacQueen, J. Some methods for classification and analysis of multivariate observations. In Proc. of the Fifth Berkeley Symposium on Mathematical Statistics and Probability Vol. 1 (eds Le Cam, L. M. & Neyman, J.), 281–297 (University of California Press, 1967).
Ward Jr, J. H. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc. 58, 236–244 (1963).
Ng, A. Y., Jordan, M. I. & Weiss, Y. On spectral clustering: analysis and an algorithm. Adv. Neural Inf. Process. Syst. (NeurIPS) 14, 849–856 (2002).
Reynolds, D. A. Gaussian mixture models. In Encyclopedia of Biometrics (eds Li, S. Z. & Jain, A.), 659–663 (Springer US, 2009).
Ester, M., Kriegel, H.-P., Sander, J. & Xu, X. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proc. of the Second International Conference on Knowledge Discovery and Data Mining (KDD) (eds Simoudis, E., Han, J. & Fayyad, U. M.), 226–231 (AAAI Press, 1996).
Rousseeuw, P. J. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. J. Comput. Appl. Math. 20, 53–65 (1987).
Caliński, T. & Harabasz, J. A dendrite method for cluster analysis. Commun. Stat.-theory Methods 3, 1–27 (1974).
Davies, D. L. & Bouldin, D. W. A cluster separation measure. IEEE Trans. Pattern Anal. Mach. Intell. PAMI-1, 224–227 (1979).
Hosmer Jr, D. W., Lemeshow, S. & Sturdivant, R. X. Applied Logistic Regression (John Wiley & Sons, 2013).
Gorishniy, Y., Rubachev, I., Khrulkov, V. & Babenko, A. Revisiting deep learning models for tabular data. Adv. Neural Inf. Process. Syst. 34, 18932–18943 (2021).
Salton, G. & Buckley, C. Term-weighting approaches in automatic text retrieval. In Information Processing & Management Vol. 24, 513–523 (Elsevier, 1988).
Devlin, J., Chang, M.-W., Lee, K. & Toutanova, K. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proc. of NAACL-HLT, 4171–4186 (2019).
Beltagy, I., Peters, M. E. & Cohan, A. Longformer: the long-document transformer. arXiv preprint arXiv:2004.05150 (2020).
Zaheer, M. et al. Big Bird: transformers for longer sequences. In Advances in Neural Information Processing Systems Vol. 33 (eds Larochelle, H. et al.), 17283–17297 (Curran Associates, Inc., 2020).
Ngiam, J. et al. Multimodal deep learning. In Proc. of the 28th International Conference on Machine Learning (ICML) (eds Getoor, L. & Scheffer, T.) 689–696 (International Machine Learning Society, 2011).
Baltrušaitis, T., Ahuja, C. & Morency, L.-P. Multimodal machine learning: a survey and taxonomy. IEEE Trans. Pattern Anal. Mach. Intell. 41, 423–443 (2019).
Tsai, Y.-H. H. et al. Multimodal transformer for unaligned multimodal language sequences. In Proc. of ACL (eds Korhonen, A., Traum, D. & Márton, G.) 6558–6569 (Association for Computational Linguistics, 2019).
Lu, J., Batra, D., Parikh, D. & Lee, S. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. In Advances in Neural Information Processing Systems Vol. 32 (eds Wallach, H. et al.) (Curran Associates, Inc., 2019).
Alberti, C. et al. Fusion of detected objects in text for visual question answering. In Proc. of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP-IJCNLP) (eds Inui, K. et al.) 2131–2140 (Association for Computational Linguistics, 2019).
Loshchilov, I. & Hutter, F. Decoupled weight decay regularization. In International Conference on Learning Representations (ICLR) (eds Bach, F. & Blei, D.) (OpenReview.net, 2019).
Lin, T.-Y., Goyal, P., Girshick, R., He, K. & Dollár, P. Focal loss for dense object detection. In Proc. of the IEEE International Conference on Computer Vision (ICCV) (eds Venice, G. et al.) 2980–2988 (IEEE Computer Society, 2017).
Gao, T., Yao, X. & Chen, D. Simcse: Simple contrastive learning of sentence embeddings. In Proc. of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP) (eds Moens, M., Huang, X., Specia, L. & Yih, S. W.) 6894–6910 (Association for Computational Linguistics, 2021).
Lilliefors, H. W. On the Kolmogorov–Smirnov test for normality with mean and variance unknown. J. Am. Stat. Assoc. 62, 399–402 (1967).
Fisher, R. A. Statistical methods for research workers. In Breakthroughs in Statistics, Springer Series in Statistics (eds Kotz, S. & Johnson, N. L.) 66–70 (Springer, 1992).
Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B (Methodological) 57, 289–300 (1995).
Kruskal, W. H. & Wallis, W. A. Use of ranks in one-criterion variance analysis. J. Am. Stat. Assoc. 47, 583–621 (1952).
Dunn, O. J. Multiple comparisons using rank sums. Technometrics 6, 241–252 (1964).
Paszke, A. et al. PyTorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems, Vol. 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8–14, 2019, Vancouver, BC, Canada 8024–8035 (2019).
Acknowledgements
This work was supported by the National Key Research and Development Program of China (2024YFA1011900), Science and Technology Program of Guangzhou, China (2024A03J1188), Guangdong Provincial Key Laboratory of Research on Emergency in TCM (2023B1212060062), National Natural Science Foundation of China (82374392), the Incubation Program for the Science and Technology Development of Chinese Medicine Guangdong Laboratory (HQL2024PZ022), National Major Projects for Science and Technology Development (2025ZD01903002), and Guangdong Healthcare Talent Development Project (0720240226). The authors sincerely thank all clinicians and data management staff involved in this study for their valuable assistance. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Author information
Authors and Affiliations
Contributions
T.L.: study conceptualization and design, construction of model, technical implementation, data analysis, statistical analysis, manuscript drafting; Y.L.: study conceptualization and design, data preparation, resources, statistical analysis, manuscript drafting; H.C.: study conceptualization and design, construction of model, data analysis; N.L.: study conceptualization and design, data preparation, statistical analysis; Y.Z., X.H.: data analysis, statistical analysis; J.W., R.C., Y.Z., and Y.L.: resources; D.Z.: data analysis; D.W.: resources; C.W.: study conceptualization and design, construction of model, manuscript drafting; T.Y. and X.X.: study conceptualization and design, data preparation, resources; Z.Z.: study conceptualization and design, data preparation, resources, manuscript drafting. All authors reviewed and approved the final manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Liu, T., Li, Y., Chen, H. et al. A multimodal embedding model for sepsis data representation. npj Digit. Med. (2026). https://doi.org/10.1038/s41746-026-02446-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41746-026-02446-3
