Machine Learning for Predicting neurological outcomes After Cardiac Arrest: A Systematic Review (Preprint)

Abstract

Cardiac arrest (CA) is still one of the most common causes of death in the world. However, CA patients survived in 45 to 70% of the patients suffering from hypoxic-ischemic encephalopathy after CA, which is manifested as severe neurological impairment and death, and the survival rate is also reduced. A severe poor prognosis will lead to patients' physical and psychological damage, also leading to a serious burden and impact on the patients' families and society as well as serious burden and impact on the patients' families and society.

Early clinical prediction of cardiac arrest (CA) can be beneficial and challenging. With the development of machine learning (ML) and a series of promising models to predict neurological outcomes after cardiac arrest, this study systematically reviewed the prediction value of these models.

PubMed, Embase, WanFang Data, and CNKI databases were electronically searched to retrieve all ML studies on predicting CA from January 2015 to February 2021. Two reviewers independently screened literature, extracted data, and assessed the risk of bias in included studies. The value of each model was evaluated based on the area under the receiver operating characteristic curve (AUROC), sensitivity, or specificity.

A total of 30 studies were eventually included. In terms of data sources, five kinds of research were based on the public database, and other studies retrospectively collected clinical data. A total of 52 models had been adopted, among which the most popular ML methods included logistic regression (n= 22), followed by random forest (n= 9). The most frequently used input features were age (n=30) and gender (n=30). Three studies compared the regression models with other ML models and the results presented that Machine learning shows no advantage over traditional algorithms.

The available evidence suggests that ML can be predicting cardiac arrest outcomes, but the current set of predictor models are primarily built on supervised learning and mostly used regression models built on retrospective data.

The study protocol was registered and approved on the international prospective register of systematic reviews PROSPERO before the start of the study (reference number CRD42021247323).