Predicting early mortality after CPR in the ICU: a multimodal analytical approach

Abstract

Aims: Mortality rates remain high among patients admitted to the intensive care unit (ICU) following successful return of spontaneous circulation (ROSC) after cardiopulmonary resuscitation (CPR). Identifying risk factors specific to this patient group may directly inform clinical decision-making processes. This study aimed to identify the clinical and laboratory parameters associated with mortality in post-CPR ICU patients and to compare machine learning models developed using these parameterswith traditional statistical analyses. Methods: This retrospective study included a total of 82 patients treated in a tertiary-level ICU between 2020 and 2023. The post-CPR group (n=41) consisted of patients admitted to the ICU following effective CPR and ROSC, while the control group (n=41) included randomly selected patients with similar clinical characteristics who had not undergone CPR. Demographic data, clinical scores (APACHE II, SOFA, NUTRIC), laboratory values, and survival outcomes were recorded. Mortality prediction models were developed using the Random Forest algorithm applied to class-balanced datasets generated with the ADASYN method. Results: The post-CPR group had significantly higher scores and biomarker levels, including APACHE II, SOFA, and CRP, whereas albumin and GFR levels were notably lower. Both ICU and hospital mortality rates were significantly elevated in this group (75.6% and 80.5%, respectively; p<0.001). In general ICU mortality models developed using Random Forest, variables such as inotropic support, APACHE II, SOFA, and CRP emerged as prominent predictors, and the model demonstrated high predictive performance (AUC: 0.914). In the subgroup of post-CPR patients, factors such as thrombocyte count, mean platelet volume, and sex were found to be particularly influential in predicting mortality. Conclusion: Both traditional statistical analyses and machine learning models provide clinically meaningful results in predicting early mortality among post-CPR patients. In particular, the need for inotropic support and elevated inflammatory markers appear to be strong predictors of mortality. The high predictive performance of AI-supported models, even with small sample sizes, highlights their potential clinical utility, though prospective observational studies are needed to further validate these models. However, the limited cohort size and the absence of resuscitation-specific variables such as initial CPR rhythm and duration represent important limitations that should be addressed in future prospective studies. The dataset used for model development, along with the executable Python scripts, is available for sharing.

Keywords

• Post-CPR
• early mortality prediction
• machine learning

Predicting early mortality after CPR in the ICU: a multimodal analytical approach

Abstract

Aims: Mortality rates remain high among patients admitted to the intensive care unit (ICU) following successful return of spontaneous circulation (ROSC) after cardiopulmonary resuscitation (CPR). Identifying risk factors specific to this patient group may directly inform clinical decision-making processes. This study aimed to identify the clinical and laboratory parameters associated with mortality in post-CPR ICU patients and to compare machine learning models developed using these parameters with traditional statistical analyses. Methods: This retrospective study included a total of 82 patients treated in a tertiary-level ICU between 2020 and 2023. The post-CPR group (n=41) consisted of patients admitted to the ICU following effective CPR and ROSC, while the control group (n=41) included randomly selected patients with similar clinical characteristics who had not undergone CPR. Demographic data, clinical scores (APACHE II, SOFA, NUTRIC), laboratory values, and survival outcomes were recorded. Mortality prediction models were developed using the Random Forest algorithm applied to class-balanced datasets generated with the ADASYN method. Results: The post-CPR group had significantly higher scores and biomarker levels, including APACHE II, SOFA, and CRP, whereas albumin and GFR levels were notably lower. Both ICU and hospital mortality rates were significantly elevated in this group (75.6% and 80.5%, respectively; p<0.001). In general ICU mortality models developed using Random Forest, variables such as inotropic support, APACHE II, SOFA, and CRP emerged as prominent predictors, and the model demonstrated high predictive performance (AUC: 0.914). In the subgroup of post-CPR patients, factors such as thrombocyte count, mean platelet volume, and sex were found to be particularly influential in predicting mortality. Conclusion: Both traditional statistical analyses and machine learning models provide clinically meaningful results in predicting early mortality among post-CPR patients. In particular, the need for inotropic support and elevated inflammatory markers appear to be strong predictors of mortality. The high predictive performance of AI-supported models, even with small sample sizes, highlights their potential clinical utility, though prospective observational studies are needed to further validate these models. However, the limited cohort size and the absence of resuscitation-specific variables such as initial CPR rhythm and duration represent important limitations that should be addressed in future prospective studies. The dataset used for model development, along with the executable Python scripts, is available for sharing.

Keywords

• Post-CPR
• early mortality prediction
• machine learning

References

1. Nolan JP, Sandroni C, Böttiger BW, et al. Cardiac arrest and cardiopulmonary resuscitation outcome reports: update of the Utstein Resuscitation Registry Templates. Resuscitation. 2019;144:166-177.
2. Berg KM, Cheng A, Panchal AR, et al. Part 7: systems of care: 2020 American Heart Association Guidelines for cardiopulmonary resuscitation and emergency cardiovascular care. Circulation. 2020; 142(16_suppl_2):S580-S604. doi:10.1161/CIR.0000000000000899
3. Schuetz P, Müller B, Christ-Crain M, et al. Procalcitonin to initiate or discontinue antibiotics in acute respiratory tract infections. Cochrane Database Syst Rev. 2012;2012(9):CD007498. doi:10.1002/14651858.CD 007498.pub2
4. Wacker C, Prkno A, Brunkhorst FM, Schlattmann P. Procalcitonin as a diagnostic marker for sepsis: a systematic review and meta-analysis. Lancet Infect Dis. 2013;13(5):426-435. doi:10.1016/S1473-3099 (12)70323-7
5. Erenler AK, Yapar D, Terzi Ö. Comparison of procalcitonin and C-reactive protein in differential diagnosis of sepsis and severe sepsis in emergency department. Dicle Med. J. 2017;44(2):175-182. doi:10.5798/dicletip.319750
6. Silvestre JP, Coelho LM, Póvoa PM. Impact of fulminant hepatic failure in C-reactive protein?. J Crit Care. 2010;25(4):657-e7. doi:10.1016/j.jcrc. 2010.02.004
7. Kan WC, Huang YT, Wu VC, Shiao CC. Predictive Ability of procalcitonin for acute kidney injury: a narrative review focusing on the interference of infection. Int J Mol Sci. 2021;22(13):6903. doi:10.3390/ijms22136903
8. Iesu E, Franchi F, Zama Cavicchi F, et al. Acute liver dysfunction after cardiac arrest. PLoS One. 2018;13(11):e0206655. doi:10.1371/journal.pone.0206655

9. Jeppesen KK, Rasmussen SB, Kjaergaard J, et al. Acute kidney injury after out-of-hospital cardiac arrest. Crit Care. 2024;28(1):169. doi:10. 1186/s13054-024-04936-w
10. Vardon-Bounes F, Gratacap MP, Groyer S, et al. Kinetics of mean platelet volume predicts mortality in patients with septic shock. PLoS One. 2019;14(10):e0223553. doi:10.1371/journal.pone.0223553
11. Ari M, Akinci Ozyurek B, Yildiz M, et al. Mean platelet volume-to-platelet count ratio (MPR) in acute exacerbations of idiopathic pulmonary fibrosis: a novel biomarker for icu mortality. Medicina (Kaunas). 2025;61(2):244. doi:10.3390/medicina61020244
12. Rimmer E, Garland A, Kumar A, et al. White blood cell count trajectory and mortality in septic shock: a historical cohort study. Évolution de la numération leucocytaire et mortalité en cas de choc septique: une étude de cohorte historique. Can J Anaesth. 2022;69(10):1230-1239. doi:10. 1007/s12630-022-02282-5
13. Asai N, Ohashi W, Sakanashi D, et al. Combination of sequential organ failure assessment (SOFA) score and Charlson comorbidity index (CCI) could predict the severity and prognosis of candidemia more accurately than the acute physiology, age, chronic health evaluation II (APACHE II) score. BMC Infect Dis. 2021;21(1):77. doi:10.1186/s12879-020-05719-8
14. Wang N, Qin Z, Liu H, Shang N, Wang Y, Xi X. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue. 2022;34(3):245-249. doi:10.3760/cma.j.cn121430- 20211019-01525
15. Seppä AMJ, Skrifvars MB, Vuopio H, Raj R, Reinikainen M, Pekkarinen PT. Association of white blood cell count with one-year mortality after cardiac arrest. Resusc Plus. 2024;20:100816. doi:10.1016/j.resplu.2024. 100816
16. Kong G, Lin K, Hu Y. Using machine learning methods to predict in-hospital mortality of sepsis patients in the ICU. BMC Med Inform Decis Mak. 2020;20(1):251. doi:10.1186/s12911-020-01271-2
17. Kim YT, Kim DK, Kim H, Kim DJ. A comparison of oversampling methods for constructing a prognostic model in the patient with heart failure. ICTC. 2020:379-383. doi:10.1109/ICTC49870.2020.9289522
18. Panda AR, Banerjee S, Naik S. Effect of different oversampling techniques to handle class imbalance challenges in coronary heart disease prediction. GCCIT. 2024:1-5. doi:10.1109/GCCIT63234.2024.10862873
19. Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ 2024;385:e078378. doi:10. 1136/bmj-2023-078378
20. de Vries MC, Koekkoek WK, Opdam MH, van Blokland D, van Zanten AR. Nutritional assessment of critically ill patients: validation of the modified NUTRIC score. Eur J Clin Nutr. 2018;72(3):428-435. doi:10. 1038/s41430-017-0008-7
21. Mukhopadhyay A, Henry J, Ong V, et al. Association of modified NUTRIC score with 28-day mortality in critically ill patients. Clin Nutr. 2017;36(4):1143-1148. doi:10.1016/j.clnu.2016.08.004
22. Soni KD, Rai N, Aggarwal R, Trikha A. Outcomes of trauma victims with cardiac arrest who survived to intensive care unit admission in a level 1 apex Indian Trauma Centre: a retrospective cohort study. Indian J Crit Care Med. 2021;25(12):1408-1412. doi:10.5005/jp-journals-10071-24057
23. Kim JH, Kwon YS, Baek MS. Machine learning models to predict 30-day mortality in mechanically ventilated patients. J Clin Med. 2021;10(10): 2172. doi:10.3390/jcm10102172
24. Xie Y, Lin L, Sun C, Chen L, Lv W. Association between serum alkaline phosphatase and clinical prognosis in patients with acute liver failure following cardiac arrest: a retrospective cohort study. Eur J Med Res. 2024;29(1):453. doi:10.1186/s40001-024-02049-2
25. Nanayakkara S, Fogarty S, Tremeer M, et al. Characterising risk of in-hospital mortality following cardiac arrest using machine learning: a retrospective international registry study. PLoS Med. 2018;15(11): e1002709. doi:10.1371/journal.pmed.1002709
26. Cotoia A, Franchi F, De Fazio C, Vincent JL, Creteur J, Taccone FS. Platelet indices and outcome after cardiac arrest. BMC Emerg Med. 2018; 18(1):31. doi:10.1186/s12873-018-0183-4
27. Duran M, Uludağ Ö. Can platelet count and mean platelet volume and red cell distribution width be used as a prognostic factor for mortality in intensive care unit?. Cureus. 2020;12(11):e11630. doi:10.7759/cureus. 11630
28. Annborn M, Dankiewicz J, Erlinge D, et al. Procalcitonin after cardiac arrest-an indicator of severity of illness, ischemia-reperfusion injury and outcome. Resuscitation. 2013;84(6):782-787. doi:10.1016/j.resuscitation.2013.01.004
29. Beumier M, Cortez DO, Donadello K, Vincent JL, Taccone FS. CRP levels after cardiac arrest. Crit Care Med. 2012;40(12):586. doi:10.1097/01.ccm. 0000424803.51586.56
30. Zhuang YG, Chen YZ, Zhou SQ, Peng H, Chen YQ, Li DJ. High plasma levels of pro-inflammatory factors interleukin-17 and interleukin-23 are associated with poor outcome of cardiac-arrest patients: a single center experience. BMC Cardiovasc Disord. 2020;20(1):170. doi:10.1186/s12872-020-01451-y
31. Li Y, She Y, Mo W, Jin B, Xiang W, Luo L. Albumin level at admission to the intensive care unit is associated with prognosis in cardiac arrest patients. Cureus. 2021;13(4):e14501. doi:10.7759/cureus.14501
32. Zheng H, Waqar MM, Arif S, Sherazi SWA, Son SH Lee JY. An explainable machine learning-based prediction model for in-hospital mortality in acute myocardial infarction patients with typical chest pain. Proc Int Conf Softw Eng Inf Manag. 2023:44-49. doi:10.1145/3584871.3584877
33. Wang Z, Li Y, Chen H, et al. Mortality prediction of ICU patients with rheumatic heart disease using machine learning on imbalanced data. AIMS Bioinf. Data Integr. 2024;3:1-15. doi:10.3934/bdia.2024003