Prediction of cardiac arrest in critically ill patients based on bedside vital signs monitoring
Introduction
Cardiac arrest (CA) refers to sudden interruption of cardiac activity, usually caused by some specified anomalous events, such as ventricular fibrillation (VF), pulseless ventricular tachycardia (VT), asystole and pulseless electrical activity [1]. About 290,000 in-hospital CAs are reported in the United States each year, and the survival discharge rate of patients with CA is only 25% [2]. A study has shown that 59.4% of patients had at least one abnormal sign within 1–4 h before CA, such as respiratory problems or hemodynamic instability [3]. A study of Bergum et al. further highlighted that early recognition of the causes of CA has been shown to increase the survival rate of patients within an hour of episode by about 29% and by 19% until their discharge [4]. Thus, the early detection of CA is quite beneficial to reduce in-hospital mortality.
From a clinical point of view, the real-time as well as accuracy of predicting CA in the short time window before event have great impact. Additionally, interpretability is a significant factor for doctor to make medical decisions. Over the past decade, multiple attempts have been made to predict CA. Several studies managed to train models with more and more abundant features, e.g., demographic variables and laboratory values [5], [6], [7], [8], [9]. Although the predictions can be more accurate, the models are not real-time and overly rely on the features that are not commonly used in hospital. Other studies developed real-time early warning systems using deep learning networks [10,11]. However, due to the “black box” nature of deep learning, clinician often believe that the results of deep learning are unexplained, so they distrust the medical decision support of deep learning [12].
Bedside vital signs monitoring is the most common nursing assessment [13], which includes five vital signs: blood pressure, peripheral oxygen saturation (SpO2), heart rate (HR), respiratory rate (RR) and body temperature. These can help identify the deterioration of patients. Since they have been widely used in clinical practice, they are suitable to be used as the data source for the CA prediction.
Based on vital signs monitoring, this study aims to develop a real-time and interpretable machine learning model, namely cardiac arrest prediction index (CAPI), to predict the risk of CA in critically ill patients with only four vital signs (blood pressure, SpO2, RR, HR) and verify its clinical practicability.
Section snippets
Dataset
We tested our method using the multi-parameter waveform data in the Medical Information Mart for Intensive Care III (MIMIC III) database, which was developed by the MIT Lab [14]. The MIMIC III publication noted that, “the project was approved by the Institutional Review Boards of Beth Israel Deaconess Medical Center (Boston, MA) and the Massachusetts Institute of Technology (Cambridge, MA). Requirement for individual patient consent was waived because the project did not impact clinical care
Patient characteristics
A total of 1860 patients (169 CA patients and 1691 non-CA patients) were selected. We randomly assigned 80% of the patients for model development and 20% for testing (Table 1). Continuous and categorical characteristics of patients were presented as median with upper/lower quartiles and percentage, respectively. T-test and Chi-square test were conducted to analyze the grouping differences between CA and non-CA patients. The confidence interval and significance level of both sides were lower
Discussion
Based on the MIMIC III database, a real-time, high-accuracy, and interpretable CAPI which uses only four vital signs was developed. As shown in Table 3, compared with previous studies, CAPI can be more appropriate for bedside vital signs monitoring to predict CA. Although the studies of Hong S et al. [8] and Kennedy et al. [9] proved higher accuracy, they relied on some features that couldn't be obtained from vital signs monitoring and were not real-time. An example of a real-time prediction of
Conclusion
We validated the machine-learning algorithm, CAPI, in a study including an ICU population. CAPI used only four common vital signs extracted from vital signs monitoring, which provided high accuracy for the early detection of CA. Every 5 min, CAPI aims to determine the risk of the patient developing CA within the next 1 hour. In addition, we used SHAP value to explain the overall and real-time behavior of CAPI, which helped doctors to gain insights into the relationship between prediction index
Funding/support
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (29)
- et al.
Cardiac arrest and cardiopulmonary resuscitation outcome reports: update of the Utstein Resuscitation Registry template for in-hospital cardiac arrest: a consensus report from a task force of the International Liaison Committee on Resuscitation (American Heart Association, European Resuscitation Council, Australian and New Zealand Council on Resuscitation, Heart and Stroke Foundation of Canada, InterAmerican Heart Foundation, Resuscitation Council of Southern Africa, Resuscitation Council of Asia)
Resuscitation.
(2019 Nov) - et al.
American Heart Association's Get with the Guidelines (®) – Resuscitation Investigators. The prevalence and significance of abnormal vital signs prior to in-hospital cardiac arrest
Resuscitation
(2016 Jan) - et al.
Recognizing the causes of in-hospital cardiac arrest–a survival benefit
Resuscitation
(2015 Dec) - et al.
Strengths and limitations of early warning scores: a systematic review and narrative synthesis
Int. J. Nurs. Stud.
(2017 Nov) - et al.
An intelligent warning model for early prediction of cardiac arrest in sepsis patients
Comput. Methods Programs Biomed.
(2019 Sep) - et al.
Vital signs monitoring and nurse-patient interaction: a qualitative observational study of hospital practice
Int. J. Nurs. Stud.
(2016 Apr) Vital signs: vital for surviving in-hospital cardiac arrest?
Resuscitation
(2016 Jan)The role of central venous oximetry, lactic acid concentration and shock index in the evaluation of clinical shock: a review
Resuscitation
(1992 Aug-Sep)- et al.
Prediction of respiratory decompensation in Covid-19 patients using machine learning: the READY trial
Comput. Biol. Med.
(2020 Sep) - et al.
Electrocardiographic markers and the left ventricular ejection fraction have cumulative effects on risk of sudden cardiac death
JACC Clin. Electrophysiol
(2015 Dec 1)
In-hospital cardiac arrest: a review
JAMA
Multicenter comparison of machine learning methods and conventional regression for predicting clinical deterioration on the wards
Crit. Care Med.
Prediction of cardiac arrest in the emergency department based on machine learning and sequential characteristics: model development and retrospective clinical validation study
JMIR Med. Inform
Using time series analysis to predict cardiac arrest in a PICU
Pediatr. Crit. Care Med.
Cited by (12)
A machine learning approach to predict cerebral perfusion status based on internal carotid artery blood flow
2023, Computers in Biology and MedicineDynamic Prediction of Patient Outcomes in the Intensive Care Unit: A Scoping Review of the State-of-the-Art
2023, Journal of Intensive Care MedicineExplainable Artificial Intelligence Warning Model Using an Ensemble Approach for In-Hospital Cardiac Arrest Prediction: Retrospective Cohort Study
2023, Journal of Medical Internet ResearchA Retrospective Cohort Study: Predicting 90-Day Mortality for ICU Trauma Patients with a Machine Learning Algorithm Using XGBoost Using MIMIC-III Database
2023, Journal of Multidisciplinary Healthcare