Predictive model for daily risk alerts in sepsis patients in the ICU: visualization and clinical analysis of risk indicators

Abstract

This study introduces a novel Transformer-based time-series framework designed to revolutionize risk stratification in Intensive Care Units (ICUs) by predicting patient outcomes with high temporal precision. Leveraging sequential data from the eICU database, our two-stage architecture dynamically captures evolving health trajectories throughout a patient’s ICU stay, enabling real-time identification of high-risk individuals and actionable insights for personalized interventions. The model demonstrated exceptional predictive power, achieving a progressive AUC increase from 0.87 (±0.021) on admission day to 0.92 (±0.009) by day 5, reflecting its capacity to assimilate longitudinal physiological patterns. Rigorous external validation across geographically diverse cohorts—including an 81.8% accuracy on Chinese sepsis data (AUC=0.73) and 76.56% accuracy on MIMIC-IV-3.1 (AUC=0.84)—confirmed robust generalizability. Crucially, SHAP-derived temporal heatmaps unveiled mortality-associated feature dynamics over time, bridging the gap between model predictions and clinically interpretable biomarkers. These findings establish a new paradigm for ICU prognostics, where data-driven temporal modeling synergizes with clinician expertise to optimize triage, reduce diagnostic latency, and ultimately improve survival outcomes in critical care.

Keywords

sepsis / Transformer / time-series / visualization / predicting mortality

Availability of data and material

The data used in this study were sourced from the publicly available eICU database (https://eicu-crd.mit.edu/about/eicu/). For access to the related code, please contact the corresponding author.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant Nos. 32200545 and 32270690) and the 1·3·5 Project for Disciplines of Excellence, West China Hospital, Sichuan University (Grant Nos. ZYGD23012 and ZYAI24044), and was funded by the EU and the Xunta de Galicia (Spain) through grant ED431C2022/46 for Competitive Reference Groups (GRC). It was also supported by CITIC-UDC and INIBIC. We are grateful to the staff in our research groups involved in the study for their valuable contributions and discussions.

Author contributions

Hao Yang (Data curation, Methodology, Writing—original draft), Jiaxi Li (Writing—original draft, Writing—review & editing), Chi Zhang (Supervision, Validation, Visualization), Alejandro Pazos Sierra (Conceptualization), and Bairong Shen (Conceptualization).

Supplementary data

Supplementary data is available at PCMEDI online.

Conflict of interest

There are no conflicts of interest related to this manuscript. In addition, as an Editorial Board Member of Precision Clinical Medicine, the corresponding author Bairong Shen was blinded from reviewing and making decisions on this manuscript.

Ethics approval

This study was approved by the Medical Ethics Committee of the West China Hospital, Sichuan University (2024-126). All procedures were conducted in accordance with relevant ethical guidelines. No patient-identifiable data were used in this study, and all data were anonymized to ensure privacy and confidentiality.

References

[1] Singer M, Deutschman CS, Seymour CW et al. The third international consensus definitions for sepsis and septic shock (Sepsis3). JAMA 2016;315:801-10. https://doi.org/10.1001/jama.2016.0287.

[2] Iba T, Levi M, Levy JH. Sepsis-induced Coagulopathy and Disseminated Intravascular Coagulation. In Seminars in Thrombosis and Hemostasis. Stuttgart, Germany Thieme Medical Publishers, 2020.

[3] Patel R, Arisoyin AE, Okoronkwo OU et al. Trends and factors associated with the mortality rate of depressive episodes: an analysis of the CDC wide-ranging online data for epidemiological research (WONDER) database. Cureus 2023; 15(7):e41627.

[4] Jones AE, Trzeciak S, Kline JA. The sequential organ failure assessment score for predicting outcome in patients with severe sepsis and evidence of hypoperfusion at the time of emergency department presentation. Crit Care Med 2009;37:1649-54. https://doi.org/10.1097/CCM.0b013e31819def97.

[5] Zeng W, Mao P, Huang Y et al. Analyses of factors affecting prognosis of patients with sepsis and evaluation of their predicting values. Chinese Journal of Integrated Traditional and Western Medicine in Intensive and Critical Care 2015; (6): 118-23.

[6] Chen T, Guestrin C. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on Knowledge Discovery and Data Mining. 2016. ACM (Association for Computing Machinery), New York, NY, USA.

[7] Rhee C, Wang R, Song Y et al. Risk adjustment for sepsis mortality to facilitate hospital comparisons using centers for disease control and prevention's adult sepsis event criteria and routine electronic clinical data. Critical Care Explorations 2019;1:e0049. https://doi.org/10.1097/CCE.0000000000000049.

[8] Park JY, Hsu T-C, Hu J-R et al. Predicting sepsis mortality in a population-based national database: machine learning approach. Journal of Medical Internet Research 2022;24:e29982. https://doi.org/10.2196/29982.

[9] Su L, Xu Z, Chang F et al. Early prediction of mortality, severity, and length of stay in the intensive care unit of sepsis patients based on sepsis 3.0 by machine learning models. Frontiers in Medicine 2021;8:664966. https://doi.org/10.3389/fmed.2021.664966.

[10] Pollard TJ, Johnson AEW, Raffa JD et al. The eICU Collaborative Research Database, a freely available multi-center database for critical care research. Sci Data 2018;5:1-13. https://doi.org/10.1038/sdata.2018.178.

[11] Vaswani A. Attention is all you need. arXiv preprint arXiv:1706.03762, 2017.

[12] Harris CR, Millman KJ, Van Der Walt SJ et al. Array programming with NumPy. Nature 2020;585:357-62. https://doi.org/10.1038/s41586-020-2649-2.

[13] Reback J, McKinney W. pandas-dev/pandas: Pandas 1.0.5. Zenodo, 2020. https://doi.org/10.5281/zenodo.3898987.

[14] Ketkar N, Moolayil J. Introduction to pytorch. Deep Learning with Python: Learn Best Practices of Deep Learning Models with PyTorch, 2021, pp. 27-91. Apress, New York.

[15] Lundberg S. A unified approach to interpreting model predictions. arXiv preprint arXiv:1705.07874, 2017.

[16] Lin T. Focal loss for dense object detection. arXiv preprint arXiv:1708.02002, 2017.

[17] Worsley J, Drake JD. Practical PostgreSQL. 2002: "O'Reilly Media, Inc.".

[18] Pramesti G. Kupas tuntas data penelitian dengan SPSS 22. 2015: Elex Media Komputindo.

[19] McKnight PE, Najab J. Mann-Whitney U Test. The Corsini Encyclopedia of Psychology, 2010, pp. 1-1. Wiley, Hoboken.

[20] McHugh ML. The chi-square test of independence. Biochemia medica 2013;23:143-9. https://doi.org/10.11613/BM.2013.018.

[21] Song Y-Y, Ying L. Decision tree methods: applications for classification and prediction. Shanghai Archives of Psychiatry 2015;27:130.

[22] Graves A. Long short-term memory. Supervised Sequence Labelling with Recurrent Neural Networks, 2012, pp. 37-45. Springer, Berlin.

[23] Johnson A, Bulgarelli L. Mimic-iv. PhysioNet. Available online at: https://physionet.org/content/mimiciv/3.1/(accessed August 23, 2021).

[24] Zeng Q, Zhang L, Wu Y et al., Association of number of days with severe anemia in the first week with 28-day mortality in patients with sepsis: propensity score-based analyses. https://doi.org/10.21203/rs.3.rs-4553695/v1.

[25] Donnino MW, Miller J, Goyal N et al. Effective lactate clearance is associated with improved outcome in post-cardiac arrest patients. Resuscitation 2007;75:229-34. https://doi.org/10.1016/j.resuscitation.2007.03.021.

[26] Régnier M-A, Raux M, Manach LE et al. Prognostic significance of blood lactate and lactate clearance in trauma patients. The Journal of the American Society of Anesthesiologists 2012;117:127688.

[27] Wardi G, Wali AR, Villar J et al. Unexpected intensive care transfer of admitted patients with severe sepsis. Journal of Intensive Care 2017;5:1-9. https://doi.org/10.1186/s40560-017-0239-7.

[28] Wardi G, Brice J, Correia M et al. Demystifying lactate in the emergency department. Ann Emerg Med 2020;75:287-98. https://doi.org/10.1016/j.annemergmed.2019.06.027.

[29] Shapiro NI, Howell MD, Talmor D et al. Serum lactate as a predictor of mortality in emergency department patients with in-fection. Ann Emerg Med 2005;45:524-8. https://doi.org/10.1016/j.annemergmed.2004.12.006.

[30] GD R, Herridge MS. Epidemiology and outcomes of acute lung injury. Chest 2007;131:554-62.

[31] Rose L, Gray S, Burns K et al. Emergency department length of stay for patients requiring mechanical ventilation: a prospective observational study. Scandinavian Journal of Trauma, Resuscitation and Emergency Medicine 2012;20:1-7. https://doi.org/10.1186/1757-7241-20-30.

[32] Pan C, Liu L, Xie J-F et al. Acute respiratory distress syndrome: challenge for diagnosis and therapy. Chin Med J (Engl) 2018;131:1220-4. https://doi.org/10.4103/0366-6999.228765.

[33] Serpa Neto A, Cardoso SO, Manetta JA et al. Association between use of lung-protective ventilation with lower tidal volumes and clinical outcomes among patients without acute respiratory distress syndrome: a meta-analysis. JAMA 2012;308:1651-9. https://doi.org/10.1001/jama.2012.13730.

[34] Waikar SS, Mount DB, Curhan GC. Mortality after hospitalization with mild, moderate, and severe hyponatremia. Am J Med 2009;122:857-65. https://doi.org/10.1016/j.amjmed.2009.01.027.

[35] Berend K, Van Hulsteijn LH, Gans RO. Chloride: the queen of electrolytes? European Journal of Internal Medicine 2012;23:203-11. https://doi.org/10.1016/j.ejim.2011.11.013.

[36] Grodin JL, Simon J, Hachamovitch R et al. Prognostic role of serum chloride levels in acute decompensated heart failure. J Am Coll Cardiol 2015;66:659-66. https://doi.org/10.1016/j.jacc.2015.06.007.

[37] Rivera FB, Alfonso P, Golbin JM et al. The role of serum chloride in acute and chronic heart failure: a narrative review. Cardiorenal Medicine 2021;11:87-98. https://doi.org/10.1159/000515604.

[38] Hou D, Zhong P, Ye X et al. Persistent hyperglycemia is a useful glycemic pattern to predict stroke mortality: a systematic review and meta-analysis. BMC Neurology 2021;21:1-13. https://doi.org/10.1186/s12883-021-02512-1.

[39] Lu Z, Tao G, Sun X et al. Association of blood glucose level and glycemic variability with mortality in sepsis patients during ICU hospitalization. Frontiers in Public Health 2022;10:857368. https://doi.org/10.3389/fpubh.2022.857368.

[40] Tutak AS, Avni Findikli H. Can RDW be added to intensive care disease severity scoring systems? Archives of Clinical and Biomedical Research 2018;2:183-7. https://doi.org/10.26502/acbr.50170057.

[41] Fernandez R, Cano S, Catalan I et al. High red blood cell distribution width as a marker of hospital mortality after ICU discharge: a cohort study. Journal of Intensive Care 2018;6:1-7. https://doi.org/10.1186/s40560-018-0343-3.

[42] Özdemir R, Mutlu NM, Özdemir M et al. The importance of red cell distribution width (RDW) in patient follow up in intensive care unit (ICU). Acta Medica Mediterranea 2016. 66(2):151-4.

[43] Safdar SA, Modi T, Sriramulu L et al., The role of red cell distribution width as a predictor of mortality for critically ill patients in an inner-city hospital. Journal of natural science, biology, and medicine 2017;8:154.

[44] Salmorán HO, Martínez I, Ramírez S et al. Extent of erythrocyte distribution as a predictor of mortality on admission to the Intensive Care Unit of the Angeles Pedregal Hospital. Acta Médica Grupo Angeles 2019;17:230-6.

[45] Garcia-Martinez R, Caraceni P, Bernardi M et al. Albumin: pathophysiologic basis of its role in the treatment of cirrhosis and its complications. Hepatology 2013;58:1836-46. https://doi.org/10.1002/hep. 26338.

[46] Yin M, Si L, Qin W et al. Predictive value of serum albumin level for the prognosis of severe sepsis without exogenous human albumin administration: a prospective cohort study. J Intensive Care Med 2018;33:687-94. https://doi.org/10.1177/0885066616685300.

[47] Pan SW, Kao H, Yu W et al. Synergistic impact of low serum albumin on intensive care unit admission and high blood urea nitrogen during intensive care unit stay on post-intensive care unit mortality in critically ill elderly patients requiring mechanical ventilation. Geriatrics & Gerontology International 2013;13:107-15.

[48] Arques S, Roux E, Stolidi P et al. Usefulness of serum albumin and serum total cholesterol in the prediction of hospital death in older patients with severe, acute heart failure. Arch Cardiovasc Dis 2011;104:502-8. https://doi.org/10.1016/j.acvd.2011.06.003.

[49] Baek SD, Kang J-Y, Yu H et al. Change in alkaline phosphatase activity associated with intensive care unit and hospital length of stay in patients with septic acute kidney injury on continuous renal replacement therapy. BMC Nephrology 2018;19:1-7. https://doi.org/10.1186/s12882-018-1028-9.

[50] Zahmatkeshan M, Serati Z, Freydooni S et al. Prediction of early liver failure in pediatric patients admitted to intensive care unit. Middle East Journal of Digestive Diseases 2019;11:141. https://doi.org/10.15171/mejdd.2019.140.

[51] Daviaud F, Grimaldi D, Dechartres A et al. Timing and causes of death in septic shock. Annals of Intensive Care 2015;5:1-9. https://doi.org/10.1186/s13613-015-0058-8.

[52] Baranovskii DS, Klabukov ID, Krasilnikova OA et al. Prolonged prothrombin time as an early prognostic indicator of severe acute respiratory distress syndrome in patients with COVID19 related pneumonia. Curr Med Res Opin 2021; 37:21-5. https://doi.org/10.1080/03007995.2020.1853510.

[53] Tang N, Li D, Wang X et al. Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia. J Thromb Haemost 2020;18:844-7. https://doi.org/10.1111/jth.14768.

[54] Prucnal CK, Jansson PS, Deadmon E et al. Analysis of partial thromboplastin times in patients with pulmonary embolism during the first 48 hours of anticoagulation with unfractionated heparin. Acad Emerg Med 2020;27:117-27. https://doi.org/10.1111/acem.13872.

[55] Lippi G, Favaloro EJ, Franchini M. Paradoxical thrombosis part 1: factor replacement therapy, inherited clotting factor deficiencies and prolonged APTT. J Thromb Thrombolysis 2012;34:360-6. https://doi.org/10.1007/s11239-012-0753-3.

[56] Matsubara S, Usui R, Ohkuchi A et al. Prolonged activated partial thromboplastin time in thromboprophylaxis with unfractionated heparin in patients undergoing cesarean section. J Obstet Gynaecol Res 2010;36:58-63. https://doi.org/10.1111/j.1447-0756.2009.01107.x.

[57] Huang C-B, Hong C-X, Xu T-H et al. Risk factors for pulmonary embolism in ICU patients: a retrospective cohort study from the MIMIC-III database. Clin Appl Thromb Hemost 2022; 28:10760296211073925. https://doi.org/10.1177/10760296 211073925.

[58] Yasuda H, Sanui M, Nishimura T et al. Optimal upper limits of Plateau pressure for patients with Acute Respiratory Distress Syndrome during the first seven days: a meta-regression analysis. Journal of Clinical Medicine Research 2021;13:48. https://doi.org/10.14740/jocmr4390.

[59] Cretikos MA, Bellomo R, Hillman K et al. Respiratory rate: the neglected vital sign. Med J Aust 2008;188:657-9. https://doi.org/10.5694/j.1326-5377.2008.tb01825.x.

[60] Maraziti G, Becattini C. Early variation of respiratory indexes to predict death or ICU admission in severe acute respiratory syndrome coronavirus-2-related respiratory failure. Respiration 2022;101:632-7. https://doi.org/10.1159/000522275.

[61] Taneja R, Parodo J, Jia SH et al. Delayed neutrophil apoptosis in sepsis is associated with maintenance of mitochondrial transmembrane potential and reduced caspase-9 activity. Crit Care Med 2004;32:1460-9. https://doi.org/10.1097/01.CCM.000012997 5.26905.77.

[62] Huang Z, Fu Z, Huang W et al. Prognostic value of neutrophil-to-lymphocyte ratio in sepsis: a meta-analysis. Am J Emerg Med 2020;38:641-7. https://doi.org/10.1016/j.ajem.2019.10.023.

[63] Lameire NH, Bagga A, Cruz D et al. Acute kidney injury: an increasing global concern. The Lancet 2013;382:170-9. https://doi.org/10.1016/S0140-6736(13)60647-9.

[64] Hoste EAJ, Bagshaw SM, Bellomo R et al. Epidemiology of acute kidney injury in critically ill patients: the multinational AKI-EPI study. Intensive Care Med 2015;41:1411-23. https://doi.org/10.1007/s00134-015-3934-7.

[65] Pickering JW, Ralib AM, Endre ZH. Combining creatinine and volume kinetics identifies missed cases of acute kidney injury following cardiac arrest. Critical Care 2013;17:1-18. https://doi.org/10.1186/cc11931.

[66] Pickkers P, Mehta RL, Murray PT et al. Effect of human recombinant alkaline phosphatase on 7-day creatinine clearance in patients with sepsis-associated acute kidney injury: a randomized clinical trial. JAMA 2018;320:1998-2009. https://doi.org/10.1001/jama.2018.14283.

[67] Otto GP, Sossdorf M, Claus RA et al. The late phase of sepsis is characterized by an increased microbiological burden and death rate. Critical Care 2011;15:1-8. https://doi.org/10.1186/cc10332.

[68] Weiss SL, Balamuth F, Hensley J et al. The epidemiology of hospital death following pediatric severe sepsis: when, why, and how children with sepsis die. Pediatr Crit Care Med 2017;18:823-30. https://doi.org/10.1097/PCC.0000000000001222.

[69] Remick DG, Bolgos GR, Siddiqui J et al. Six at six: interleukin6 measured 6 h after the initiation of sepsis predicts mortality over 3 days. Shock 2002;17:463-7. https://doi.org/10.1097/00024382-200206000-00004.

[70] Noritomi DT, Soriano FG, Kellum JA et al. Metabolic acidosis in patients with severe sepsis and septic shock: a longitudinal quantitative study. Crit Care Med 2009;37:2733-9. https://doi.org/10.1097/CCM.0b013e3181a59165.

[71] Zhang C, Zhang X, Sun Z et al. MetaSepsisBase: a biomarker database for systems biological analysis and personalized diagnosis of heterogeneous human sepsis. Intensive Care Med 2023;49:1015-7. https://doi.org/10.1007/s00134-023-07126-4.