Development and evaluation of interpretable machine learning models for predicting in-ICU cardiac arrest from non-cardiac causes using clinical biomarkers

Abstract

Background

In-hospital cardiac arrest (IHCA) from non-cardiac causes is a life-threatening condition characterized by high mortality and complex etiology. Despite advances in critical care monitoring, effective early warning systems remain lacking due to challenges in balancing sensitivity and specificity within imbalanced clinical data and the interpretability of machine learning models remains poorly defined.

Methods

We developed and evaluated six machine learning algorithms using data from 43,618 ICU patients in the MIMIC-IV database. Model performance was assessed through multiple metrics, and SHapley Additive exPlanations (SHAP) analysis was employed for model interpretability. Decision curve analysis was performed to evaluate clinical utility.

Results

Distinct performance trade-offs were identified across models. XGBoost achieved optimal discriminative ability (AUC = 0.730, 95% CI: 0.695–0.763), while CatBoost favored sensitivity (0.907) to meet clinical guidelines. SHAP analysis identified anion gap (AG), white blood cell count (WBC), red cell distribution width (RDW), Glasgow Coma Scale (GCS), platelet, partial thromboplastin time (PTT), and red blood cell count (RBC) as key predictors, with elevated inflammatory markers and reduced neurological parameters consistently associated with increased risk.

Conclusions

Gradient boosting algorithms demonstrate crucial utility in predicting non-cardiac IHCA. The identified interpretable biomarkers align with established pathophysiological mechanisms and may reflect downstream manifestations of genome instability pathways, representing potential targets for early clinical intervention.

Keywords

In-hospital cardiac arrest / Intensive care unit / Machine learning / Sensitivity / SHAP analysis

References

[1] Andersen LW, Holmberg MJ, Berg KM, Donnino MW, Granfeldt A. In-hospital cardiac arrest: A review. Journal Of The American Medical Association, 2019, 321: 1200-1210.

[2] Barbieri MC, Grisci BI, Dorn M. Analysis and comparison of feature selection methods towards performance and stability. Expert Systems with Applications, 2024, 249: 123667.

[3] Billet S, Vanbiervliet P, Remery M, Dekoninck J, Janssens W. The concomitant use of paracetamol and flucloxacillin: A rare cause of high aniongap metabolic acidosis in the frail oldest old. Acta Clinica Belgica, 2023, 78: 509-515.

[4] Brealey D, Brand M, Hargreaves I, Heales S, Land J, Smolenski R, Davies NA, Cooper CE, Singer M. Association between mitochondrial dysfunction and severity and outcome of septic shock. The Lancet, 2002, 360(9328): 219-223.

[5] Fenves AZ, Emmett M. Approach to patients with high anion gap metabolic acidosis: Core curriculum 2021. American Journal of Kidney Diseases, 2021, 78: 590-600.

[6] Gupta S, Gambhir JK, Kalra O, Gautam A, Shukla K, Mehndiratta M, Agarwal S, Shukla R. Association of biomarkers of inflammation and oxidative stress with the risk of chronic kidney disease in Type 2 diabetes mellitus in North Indian population. Journal of Diabetes Complications, 2013, 27(6): 548-552.

[7] Hammons L, Filopei J, Steiger D, Bondarsky E. A narrative review of red blood cell distribution width as a marker for pulmonary embolism. Journal of Thrombosis and Thrombolysis, 2019, 48(4): 638-647.

[8] Hou FF, Song Y, Du WN, Wang BB, Wang Q, Wu Q, Yan LN, Chen X. Predictive value of red blood cell distribution width in critically ill patients with acute respiratory distress syndrome: A meta-analysis. Medicine, 2025, 104: e42701.

[9] Huang Y, Li J, Li M, Aparasu RR. Application of machine learning in predicting survival outcomes involving real-world data: A scoping review. BMC Medical Research Methodology, 2023, 23: 268.

[10] Lawrence, D.W., Comper, P., Hutchison, M.G., Sharma, B. (2015). The role of apolipoprotein E epsilon (ε)-4 allele on outcome following traumatic brain injury: A systematic review. Brain Injury, 29(9), 1018-1031 https://doi.org/10.3109/02699052.2015.1005131

[11] Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B. MIMIC-IV, a freely accessible electronic health record dataset. Scientific Data, 2023, 10: 1.

[12] Keller MF, Reiner AP, Okada Y, van Rooij FJ, Johnson AD, Chen MH, Wilson JG. Genome-wide association analysis of red blood cell traits in African Americans. Human Molecular Genetics, 2014, 23(10): 2789-2799.

[13] Lundberg SM, Lee SI. A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 2017, 30: 4765-4774.

[14] Nagaraju R, Kalahasthi R, Balachandar R, Bagepally BS. Association between lead exposure and DNA damage (genotoxicity): Systematic review and meta-analysis. Archives of Toxicology, 2022, 96(11): 2899-2911.

[15] Ouyang Y, Cheng M, He B, Zhang F, Ouyang W, Zhao J, Qu Y. Interpretable machine learning models for predicting in-hospital death in patients in the intensive care unit with cerebral infarction. Computer Methods and Programs in Biomedicine, 2023, 231: 107431.

[16] Pavey H, Wood A, McEniery CM, AlGhatrif M, Arshi B, Brunner E, Chen CH, Cheng HM, Hansen TW, Ikram MK. Association between carotid-femoral pulse wave velocity and cardiovascular disease in individuals with moderate blood pressure: A systematic review and individual participant meta-analysis. British Medical Journal Open, 2025, 15: e101368.

[17] Penketh J, Nolan JP. In-hospital cardiac arrest: The state of the art. Critical Care, 2022, 26. ArticleID: 376

[18] Perman SM, Stanton E, Soar J, Berg RA, Donnino MW, Mikkelsen ME, Edelson DP, Churpek MM, Yang L, Merchant RM. Location of in-hospital cardiac arrest in the United States—Variability in event rate and outcomes. Journal of the American Heart Association, 2016, 5. ArticleID: e003638

[19] Petersen JA. Early warning score challenges and opportunities in the care of deteriorating patients. Danish Medical Journal, 2018, 65: B5439

[20] Rush CJ, Berry C, Oldroyd KG, Rocchiccioli JP, Lindsay MM, Touyz RM, Murphy CL, Ford TJ, Sidik N, McEntegart MB. Prevalence of coronary artery disease and coronary microvascular dysfunction in patients with heart failure with preserved ejection fraction. JAMA Cardiology, 2021, 6: 1130-1143.

[21] Sabater-Lleal, M., Almasy, L., Martínez-Marchán, E., ... & Tang, W. (2012). Genetic associations for activated partial thromboplastin time and prothrombin time, their gene expression profiles, and risk of coronary artery disease. American Journal of Human Genetics, 91(1), 108-119 https://doi.org/10.1016/j.ajhg.2012.05.015

[22] Schefold JC, Storm C, Krüger A, Ploner CJ, Hasper D. The Glasgow Coma Score is a predictor of good outcome in cardiac arrest patients treated with therapeutic hypothermia. Resuscitation, 2009, 80: 658-661.

[23] Semba RD, Patel KV, Ferrucci L, Sun K, Roy CN, Guralnik JM, Fried LP. Serum antioxidants and inflammation predict red cell distribution width in older women: The Women's Health and Aging Study I. Clinical Nutrition, 2010, 29(5): 600-604.

[24] Smit LCM, Bots ML, van der Leeuw J, Damen JAAG, Blankestijn PJ, Verhaar MC, Vernooij RWM. One heartbeat away from a prediction model for cardiovascular diseases in patients with chronic kidney disease: A systematic review. Cardiorenal Medicine, 2023, 13: 109-142.

[25] Smith RJ, Sarma D, Padkins MR, Gajic O, Lawler PR, Van Diepen S, Kashani KB, Jentzer JC. Admission total leukocyte count as a predictor of mortality in cardiac intensive care unit patients. JACC: Advances, 2024, 3. ArticleID: 100757

[26] Swanson K, Wu E, Zhang A, Alizadeh AA, Zou J. From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment. Cell, 2023, 186: 1772-1791.

[27] Tang, W., Schwienbacher, C., Lopez, L.M., Chen, M.H., Oudot-Mellkah, T., Johnson, A.D., Sabater-Lleal, M. (2012). Genetic associations for activated partial thromboplastin time and prothrombin time, their gene expression profiles, and risk of coronary artery disease. American Journal of Human Genetics, 91(1), 108-119. https://doi.org/10.1016/j.ajhg.2012.05.015

[28] Teasdale GM, Murray GD, Nicoll JAR. The association between APOE epsilon4, age and outcome after head injury: A prospective cohort study. Brain, 2005, 128(Pt 11): 2556-2561.

[29] Teasdale GM, Nicoll JA, Murray G, Fiddes M. Association of apolipoprotein E polymorphism with outcome after head injury. The Lancet, 1997, 350(9084): 1069-1071.

[30] Wang CH, Ho LT, Wu MC, Wu CY, Tay J, Su PI, Tsai MS, Wu YW, Chang WT, Huang CH. Prognostic implication of heart failure stage and left ventricular ejection fraction for patients with in-hospital cardiac arrest: A 16-year retrospective cohort study. Clinical Research in Cardiology, 2024, 114: 557-569.

[31] Yuan X, Xu Q, Du F, Gao X, Guo J, Zhang J, Wu Y, Zhou Z, Yu Y, Zhang Y. Development and validation of a model to predict cognitive impairment in traumatic brain injury patients: A prospective observational study. EClinicalMedicine, 2025, 80: 103023.

[32] Zhang K, Zhang X, Ding W, Xuan N, Tian B, Huang T, Zhang Z, Cui W, Huang H, Zhang G. National early warning score does not accurately predict mortality for patients with infection outside the intensive care unit: A systematic review and meta-analysis. Frontiers in Medicine, 2021, 8: 704358.

[33] Zou, X., Li, S., Fang, M., Hu, M., Bian, Y., Ling, J., Yu, S., Jing, L., Li, D., & Huang, J. (2020). Acute Physiology and Chronic Health Evaluation II Score as a Predictor of Hospital Mortality in Patients of Coronavirus Disease 2019. Critical Care Medicine, 48(8), e657-e665 https://doi.org/10.1097/CCM.0000000000004411

Funding

Medical-Engineering Interdisciplinary Research Foundation of Shenzhen University(2023YG004)

Shenzhen Science and Technology Program(KQTD20221101093605019)

RIGHTS & PERMISSIONS

Shenzhen University School of Medicine; Fondazione Istituto FIRC di Oncologia Molecolare