Skip to main content

Clinical Application of Artificial Intelligence in Emergency and Critical Care Medicine, Volume I

Cover image for Research Topic "Clinical Application of Artificial Intelligence in Emergency and Critical Care Medicine, Volume I"
203kviews
Editorial06 December 2021
Editorial: Clinical Application of Artificial Intelligence in Emergency and Critical Care Medicine, Volume I

Analytics based on artificial intelligence (AI) has greatly advanced a variety of scientific research fields such as natural language processing, imaging classification and signal processing (1). Clinical research is also revolutionized by the development of artificial intelligence (2), and conventional research paradigm is being supplemented by the new technology. Conventional treatment strategy based on evidence-based medicine typically exploits the average treatment effect in a population to dictate medical decision making(3). However, it is well known that a patient population is usually heterogeneous that one size does not fit all. In other words, although a treatment strategy is reported to be beneficial for the overall population, it might be harmful for a subgroup of patients. Thus, the idea of individualized treatment is proposed to address the problem of differential treatment effects in a heterogeneous population. Patients in emergency and critical care setting are usually heterogenous and the clinical condition changes rapidly (4,5), which highlights the importance of early risk stratification and individualized treatment.Artificial intelligence can be applied in three aspects in the emergency and critical care setting. These three aspects have been well captured in this research topic entitled "Clinical Application of Artificial Intelligence in Emergency and Critical Care Medicine, Volume I", which has been successfully launched in Frontiers in Medicine. First, several studies developed prediction models for risk stratification in the critical care setting. Different clinical risks are defined in a variety of study populations such as mortality prediction in surgical ICU patients, risk of blood transfusion in liver transplantation, and risk of coagulopathy in sepsis. Collectively, these studies exploited the supervised learning algorithm to train a prediction model (6). The clinical events of interest/labels must be unambiguously defined. Misclassification in the database will cause model instability or inaccuracy for the prediction in future samples (7). The second category of study is to disentangle heterogenous population into more homogenous subgroups by using unsupervised machine learning algorithms (8). The algorithms differ from the supervised learning methods in that they do not require the samples being labeled in advance. Instead, they exploit the features to classify samples into separatable subgroups/subtypes. The subgroups of patients can have prognostic and predictive enrichment. Prognostic enrichment indicates different subgroups have different risk of clinical outcome events, whereas the predictive enrichment indicates that different subgroups can have different responses to a particular intervention. In this collection of articles, Zhang S and colleagues explored the subphenotyes of acute respiratory distress syndrome and found that Rosuvastatin has differential treatment effect across these subphenotypes. Chen H and colleagues developed a novel clinical classification system for SARS-CoV-2 Pneumonia, which showed prognostic enrichment for mortality outcome. They further developed a parsimonious class membership prediction model for the ease of clinical utility. The third type of clinical scenario is to employ reinforcement learning algorithm to dictate treatment regimen in sequential manner (9,10). This methodology is not used in the current collection of articles. The key idea underlying this application is that the treatment strategy should be tailored sequentially according to the changes of patient's status. The interactions between treatment action, patient state, and reward are formalized in a dynamic process, so as to maximize the final outcome reward. Alternatively, the dynamic treatment regime (DTR) model adapts the idea of reinforcement learning to estimate a sequence of decision rules, one per stage of intervention, that dictate how to individualize treatments to patients based on evolving treatment and covariate history. DTR relaxes the model complexity and are more acceptable to the field of medical epidemiology. This model has been utilized in critical care setting to tailor fluid resuscitation in sepsis and ventilation strategy in acute respiratory failure (10,11).Since the advances in machine learning algorithms have greatly revolutionized the industry, the technology can surely influence how we treat patients in the emergency and critical care setting.However, the application of AI in clinical practice is still in its infancy and requires more research efforts. Several key aspects that hinder the utility of AI models in clinical practice include but not limit to the quality of training datasets, institutional idiosyncrasy, and model overfitting (12). That is why some models show good performance in the training dataset but perform poorly in new samples. The model might learn something specific to an institution/hospital, but not the underlying true pathophysiological processes. The second issue relates to the model interpretability. Although AI models can improve prediction accuracy in some situations, a notorious drawback of these models are their black box nature prohibiting easy interpretation of the predicted outcome (13,14). Physicians are less likely to adopt a recommendation made by the machine while the underlying pathophysiology is unknown or uninterpretable. Due to the importance and the potential impact of artificial intelligence on the emergency and critical care setting, we launch a second volume of the research topic. We welcome more studies to address the above-mentioned problems in applying ML in clinical practice. Successful settlement of these issues will hopefully transform more research models into real clinical practice.

2,438 views

2 citations

ORIGINAL RESEARCH
16 November 2021
Machine Learning Approach to Predict Positive Screening of Methicillin-Resistant Staphylococcus aureus During Mechanical Ventilation Using Synthetic Dataset From MIMIC-IV Database
Yohei Hirano, 7 more and Aiko Tanaka

7,056 views

28 citations

ORIGINAL RESEARCH
8 September 2021
Differing Visual Behavior Between Inexperienced and Experienced Critical Care Nurses While Using a Closed-Loop Ventilation System—A Prospective Observational Study
Philipp K. Buehler, 9 more and Daniel A. Hofmaenner

3,859 views

11 citations

ORIGINAL RESEARCH
16 August 2021
Ability of a Machine Learning Algorithm to Predict the Need for Perioperative Red Blood Cells Transfusion in Pelvic Fracture Patients: A Multicenter Cohort Study in China
Xueyuan Huang, 6 more and Xianjun Ma

4,844 views

16 citations

Machine Learning Prediction Models for Mechanically Ventilated Patients: Analyses of the MIMIC-III Database
ORIGINAL RESEARCH
1 July 2021
Machine Learning Prediction Models for Mechanically Ventilated Patients: Analyses of the MIMIC-III Database
Yibing Zhu, 17 more and Qian Yu

14,394 views

80 citations

ORIGINAL RESEARCH
28 June 2021
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
Longxiang Su, 11 more and Yun Long
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
Background Early prediction of the clinical outcome of patients with sepsis is of great significance, which can reduce the mortality of patients. But it is clinically difficult for clinicians. Methods A total of 2,224 patients with sepsis were involved over a 3-year period (2016– 2018) to the intensive care unit (ICU) of Peking Union Medical College Hospital. With all the key medical data from the first six-hour in ICU, three machine learning models including logistic regression, random forest and XGBoost were carried out to predict the mortality, severity (sepsis/septic shock), and length of ICU stay (LOS) (>6 days, ≤ 6days). Missing data imputation and oversampling were completed on the dataset before putting them into model. Results Compared to mortality and LOS prediction, the severity prediction achieves the best classification results (sensitivity = 0.65, specificity = 0.73, F1 score = 0.72, AUC = 0.79) with random forest classifier according to the area under operating receiver characteristics (AUC). Random forest model also showed the best performance (mortality prediction: sensitivity = 0.50, specificity = 0.84, F1 score = 0.66, AUC = 0.74; LOS prediction: sensitivity = 0.79, specificity = 0.66, F1 score = 0.69, AUC = 0.76) among three models. While the predictive ability of SOFA score is inferior to the above three models. Conclusions Using the random forest classifier in the first 6 hours of ICU admission can provide comprehensive early warning of sepsis, which contribute to the formulation and management of clinical decisions and the allocation and management of resources.

10,579 views

85 citations

Development and Validation of a Machine-Learning Model for Prediction of Extubation Failure in Intensive Care Units
Background: Extubation failure (EF) can lead to an increased chance of ventilator-associated pneumonia, longer hospital stays, and a higher mortality rate. This study aimed to develop and validate an accurate machine-learning model to predict EF in intensive care units (ICUs). Methods: Patients who underwent extubation in the Medical Information Mart for Intensive Care (MIMIC)-IV database were included. EF was defined as the need for ventilatory support (non-invasive ventilation or reintubation) or death within 48 hours following extubation. A machine-learning model called Categorical Boosting (CatBoost) was developed based on 89 clinical and laboratory variables. SHapley Additive exPlanations (SHAP) values were calculated to evaluate feature importance and the recursive feature elimination (RFE) algorithm was used to select key features. Hyperparameter optimization was conducted using an automated machine-learning toolkit (Neural Network Intelligence). The final model was trained based on key features and compared with 10 other models. The model was then prospectively validated in patients enrolled in the Cardiac Surgical ICU of Zhongshan Hospital, Fudan University. In addition, a web-based tool was developed to help clinicians use our model. Results: Of 16,189 patients included in the MIMIC-IV cohort, 2756 (17.0%) had EF. Nineteen key features were selected using the RFE algorithm, including age, body mass index, stroke, heart rate, respiratory rate, mean arterial pressure, peripheral oxygen saturation, temperature, pH, central venous pressure, tidal volume, positive end-expiratory pressure, mean airway pressure, pressure support ventilation (PSV) level, mechanical ventilation (MV) durations, spontaneous breathing trial success times, urine output, crystalloid amount, and antibiotic types. After hyperparameter optimization, our model had the greatest area under the receiver operating characteristic (AUROC: 0.835) in internal validation. Significant differences in mortality, reintubation rates, and NIV rates were shown between patients with a high predicted risk and those with a low predicted risk. In the prospective validation, the superiority of our model was also observed (AUROC: 0.803). According to the SHAP values, MV duration and PSV level were the most important features for prediction. Conclusions: In conclusion, this study developed and prospectively validated a CatBoost model, which better predicted EF in ICUs than other models.

16,566 views

102 citations

ORIGINAL RESEARCH
4 June 2021
Identification and Prediction of Novel Clinical Phenotypes for Intensive Care Patients With SARS-CoV-2 Pneumonia: An Observational Cohort Study
Hui Chen, 12 more and Yongsheng Li

4,082 views

7 citations

Artificial Intelligence for Clinical Decision Support in Sepsis
REVIEW
13 May 2021
Artificial Intelligence for Clinical Decision Support in Sepsis
Miao Wu, 2 more and Jie Wei

23,441 views

76 citations

Background Sepsis-induced coagulopathy (SIC) is a common cause for inducing poor prognosis in intensive care unit (ICU) patients with sepsis. This study aimed to develop a practical nomogram to predict the risk of 28-day mortality in SIC patients. Methods Data were retrieved from the Medical Information Mart for Intensive Care (MIMIC-III) database, according to predetermined inclusion and exclusion criteria. Sepsis was defined based on Sepsis 3.0 criterion and SIC based on Toshiaki Iba’s criterion. Further, SIC cohort was randomly divided into training or validation set, the predictive nomogram was constructed by using logistic regression analysis, then verified the ability of discrimination and calibration by internal validation. Results 9432 sepsis patients in MIMIC III were enrolled, in which 3280 (34.8%) patients were diagnosed as SIC during the first ICU admission. The multivariate regression analysis found that SIC was independently associated with the 7-day and 28-day mortality of sepsis patients. Parameters eligible for this nomogram including age, combined with liver disease, the administration of norepinephrine, and some clinical variables during the first 24 h since ICU admission, including mean arterial pressure (MAP), mean heart rate, mean respiratory rate, mean temperature, lactate-max, prothrombin time-max (PT-max), red cell distribution width-min (RDW-min), mean corpuscular volume-min (MCV-min) and platelet-min. This nomogram performed a better discrimination indicated by the area under the receiver operating characteristic curve (AUROC) of 0.78 (95%CI 0.76-0.80) and 0.81 (95%CI 0.78-0.84) in training and validation set respectively when predicted the risk of 28-day mortality. Meanwhile, the nomogram was closely related with an acceptable calibration reflected by the fitness of the calibration plot. In the clinical practice, the decision curve analysis of nomogram obtained more net benefit when compared with the sequential organ failure assessment (SOFA) score and the logistic organ dysfunction score (LODS). Conclusions SIC is independently related to the short-time mortality of sepsis patients. The nomogram achieved an optimal prediction of 28-day mortality in SIC patients. Using the model, the 28-day mortality risk of an individual SIC patient can be determined, which can lead to a better prognosis.

11,067 views

50 citations

Machine Learning for the Prediction of Red Blood Cell Transfusion in Patients During or After Liver Transplantation Surgery
ORIGINAL RESEARCH
22 February 2021
Machine Learning for the Prediction of Red Blood Cell Transfusion in Patients During or After Liver Transplantation Surgery
Le-Ping Liu, 6 more and Yong-Jun Wang

7,610 views

49 citations

A Machine-Learning Approach for Dynamic Prediction of Sepsis-Induced Coagulopathy in Critically Ill Patients With Sepsis
ORIGINAL RESEARCH
21 January 2021
A Machine-Learning Approach for Dynamic Prediction of Sepsis-Induced Coagulopathy in Critically Ill Patients With Sepsis
Qin-Yu Zhao, 7 more and Zhe Luo
Background: Sepsis-induced coagulopathy (SIC) denotes an increased mortality rate and poorer prognosis in septic patients. Objectives: Our study aimed to develop and validate machine-learning models to dynamically predict the risk of SIC in critically ill patients with sepsis. Methods: Machine-learning models were developed and validated based on two public databases named Medical Information Mart for Intensive Care (MIMIC)-IV and the eICU Collaborative Research Database (eICU-CRD). Dynamic prediction of SIC involved an evaluation of the risk of SIC each day after the diagnosis of sepsis using fifteen predictive models. The best model was selected based on its accuracy and area under the receiver operating characteristic curve (AUC), followed by fine-grained hyperparameter adjustment using the Bayesian Optimization Algorithm. A compact model was developed, based on 15 features selected according to their importance and clinical availability. These two models were compared with Logistic Regression and SIC scores in terms of SIC prediction. Results: Of 11362 patients in MIMIC-IV included in the final cohort, a total of 6744 (59%) patients developed SIC during sepsis. The model named Categorical Boosting (CatBoost) had the greatest AUC in our study (0.869; 95% CI: 0.850 to 0.886). Coagulation profile and renal function indicators were the most important features for predicting SIC. A compact model was developed with an AUC of 0.854 (95% CI: 0.832 to 0.872), while the AUCs of Logistic Regression and SIC scores were 0.746 (95% CI: 0.735 to 0.755) and 0.709 (95% CI: 0.687 to 0.733), respectively. A cohort of 35252 septic patients in eICU-CRD was analyzed. The AUCs of the full and the compact models in the external validation were 0.842 (95% CI: 0.837 to 0.846) and 0.803 (95% CI: 0.798 to 0.809), respectively, which were still larger than those of Logistic Regression (0.660; 95% CI: 0.653 to 0.667) and SIC scores (0.752; 95% CI: 0.747 to 0.757). Prediction results were illustrated by SHapley Additive exPlanations (SHAP) values, which made our models clinically interpretable. Conclusions: We developed two models which were able to dynamically predict the risk of SIC in septic patients better than conventional Logistic Regression and SIC scores.

14,558 views

60 citations

ORIGINAL RESEARCH
3 December 2020
Classification of Patients With Sepsis According to Immune Cell Characteristics: A Bioinformatic Analysis of Two Cohort Studies
Shi Zhang, 5 more and Haibo Qiu

4,367 views

17 citations

ORIGINAL RESEARCH
25 November 2020
Risk Factors for Patient–Ventilator Asynchrony and Its Impact on Clinical Outcomes: Analytics Based on Deep Learning Algorithm
Huiqing Ge, 8 more and Zhongheng Zhang

5,356 views

12 citations