Prediction of good neurological recovery after out-of-hospital cardiac arrest: A machine learning analysis

Prediction of neurologic outcome after out-of-hospital cardiac arrest: An interpretable approach with machine learning

Out-of-hospital cardiac arrest (OHCA) is a critical condition with low survival rates. In patients with a return of spontaneous circulation, brain injury is a leading cause of death. In this study, we propose an interpretable machine learning approach for predicting neurologic outcome after OHCA, using information available at the time of hospital admission.

METHODS

The study population were 55 615 OHCA cases registered in the Swedish Cardiopulmonary Resuscitation Registry between 2010 and 2020. The dataset was split to training and validation sets (for model development) and test set (for evaluation of the final model). We used an XGBoost algorithm with stratified, repeated 10-fold cross-validation along with Optuna framework for hyperparameters tuning. The final model was trained on 10 features selected based on the importance scores and evaluated on the test set in terms of discrimination, calibration and bias-variance tradeoff. We used SHapley Additive exPlanations to address the 'black-box' model and align with eXplainable artificial intelligence.

RESULTS

The final model achieved: area under the receiver operating characteristic value 0.964 (95% confidence interval (CI) [0.960-0.968]), sensitivity 0.606 (95% CI [0.573-0.634]), specificity 0.975 (95% CI [0.972-0.978]), positive predictive value (PPV) 0.664 (95% CI [0.625-0.696]), negative predictive value (NPV) 0.969 (95% CI [0.966-0.972]), macro F1 0.803 (95% CI [0.788-0.816]), and showed a very good calibration. SHAP features with the highest impact on the model's output were:'ROSC on arrival to hospital', 'Initial rhythm asystole' and 'Conscious on arrival to hospital'.

CONCLUSIONS

The XGBoost machine learning model with 10 features available at the time of hospital admission showed good performance for predicting neurologic outcome after OHCA, with no apparent signs of overfitting.

Improving EMS response times for out-of-hospital cardiac arrest in urban areas using drone-like vertical take-off and landing air ambulances: An international, simulation-based cohort study

BACKGROUND

Advances in vertical take-off and landing (VTOL) technologies may enable drone-like crewed air ambulances to rapidly respond to out-of-hospital cardiac arrest (OHCA) in urban areas. We estimated the impact of incorporating VTOL air ambulances on OHCA response intervals in two large urban centres in France and Canada.

METHODS

We included adult OHCAs occurring between Jan. 2017-Dec. 2018 within Greater Paris in France and Metro Vancouver in Canada. Both regions utilize tiered OHCA response with basic (BLS)- and advanced life support (ALS)-capable units. We simulated incorporating 1-2 ALS-capable VTOL air ambulances dedicated to OHCA response in each study region, and computed time intervals from call reception by emergency medical services (EMS) to arrival of the: (1) first ALS unit ("call-to-ALS arrival interval"); and (2) first EMS unit ("call-to-first EMS arrival interval").

RESULTS

There were 6,217 OHCAs included during the study period (3,760 in Greater Paris and 2,457 in Metro Vancouver). Historical median call-to-ALS arrival intervals were 21 min [IQR 16-29] in Greater Paris and 12 min [IQR 9-17] in Metro Vancouver, while median call-to-first EMS arrival intervals were 11 min [IQR 8-14] and 7 min [IQR 5-8] respectively. Incorporating 1-2 VTOL air ambulances improved median call-to-ALS arrival intervals to 7-9 min and call-to-first EMS arrival intervals to 6-8 min in both study regions (all P < 0.001).

CONCLUSION

VTOL air ambulances dedicated to OHCA response may improve EMS response intervals, with substantial improvements in ALS response metrics.

Machine learning-based analysis of regional differences in out-of-hospital cardiopulmonary arrest outcomes and resuscitation interventions in Japan

Refining out-of-hospital cardiopulmonary arrest (OHCA) resuscitation protocols for local emergency practices is vital. The lack of comprehensive evaluation methods for individualized protocols impedes targeted improvements. Thus, we employed machine learning to assess emergency medical service (EMS) records for examining regional disparities in time reduction strategies. In this retrospective study, we examined Japanese EMS records and neurological outcomes from 2015 to 2020 using nationwide data. We included patients aged ≥ 18 years with cardiogenic OHCA and visualized EMS activity time variations across prefectures. A five-layer neural network generated a neurological outcome predictive model that was trained on 80% of the data and tested on the remaining 20%. We evaluated interventions associated with changes in prognosis by simulating these changes after adjusting for time factors, including EMS contact to hospital arrival and initial defibrillation or drug administration. The study encompassed 460,540 patients, with the model's area under the curve and accuracy being 0.96 and 0.95, respectively. Reducing transport time and defibrillation improved outcomes universally, while combining transport time and drug administration showed varied efficacy. In conclusion, the association of emergency activity time with neurological outcomes varied across Japanese prefectures, suggesting the need to set targets for reducing activity time in localized emergency protocols.

Prediction model of in-hospital mortality in intensive care unit patients with cardiac arrest: a retrospective analysis of MIMIC -IV database based on machine learning

BACKGROUND

Both in-hospital cardiac arrest (IHCA) and out-of-hospital cardiac arrest (OHCA) have higher incidence and lower survival rates. Predictors of in-hospital mortality for intensive care unit (ICU) admitted cardiac arrest (CA) patients remain unclear.

METHODS

The Medical Information Mart for Intensive Care IV (MIMIC-IV) database was used to perform a retrospective study. Patients meeting the inclusion criteria were identified from the MIMIC-IV database and randomly divided into training set (n = 1206, 70%) and validation set (n = 516, 30%). Candidate predictors consisted of the demographics, comorbidity, vital signs, laboratory test results, scoring systems, and treatment information on the first day of ICU admission. Independent risk factors for in-hospital mortality were screened using the least absolute shrinkage and selection operator (LASSO) regression model and the extreme gradient boosting (XGBoost) in the training set. Multivariate logistic regression analysis was used to build prediction models in training set, and then validated in validation set. Discrimination, calibration and clinical utility of these models were compared using the area under the curve (AUC) of the receiver operating characteristic (ROC) curves, calibration curves and decision curve analysis (DCA). After pairwise comparison, the best performing model was chosen to build a nomogram.

RESULTS

Among the 1722 patients, in-hospital mortality was 53.95%. In both sets, the LASSO, XGBoost,the logistic regression(LR) model and the National Early Warning Score 2 (NEWS 2) models showed acceptable discrimination. In pairwise comparison, the prediction effectiveness was higher with the LASSO,XGBoost and LR models than the NEWS 2 model (p < 0.001). The LASSO,XGBoost and LR models also showed good calibration. The LASSO model was chosen as our final model for its higher net benefit and wider threshold range. And the LASSO model was presented as the nomogram.

CONCLUSIONS

The LASSO model enabled good prediction of in-hospital mortality in ICU admission CA patients, which may be widely used in clinical decision-making.

Neuroprognostication in the Post Cardiac Arrest Patient: A Canadian Cardiovascular Society Position Statement

Cardiac arrest (CA) is associated with a low rate of survival with favourable neurologic recovery. The most common mechanism of death after successful resuscitation from CA is withdrawal of life-sustaining measures on the basis of perceived poor neurologic prognosis due to underlying hypoxic-ischemic brain injury. Neuroprognostication is an important component of the care pathway for CA patients admitted to hospital but is complex, challenging, and often guided by limited evidence. Using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) system to evaluate the evidence underlying factors or diagnostic modalities available to determine prognosis, recommendations were generated in the following domains: (1) circumstances immediately after CA; (2) focused neurologic exam; (3) myoclonus and seizures; (4) serum biomarkers; (5) neuroimaging; (6) neurophysiologic testing; and (7) multimodal neuroprognostication. This position statement aims to serve as a practical guide to enhance in-hospital care of CA patients and emphasizes the adoption of a systematic, multimodal approach to neuroprognostication. It also highlights evidence gaps.

Artificial Intelligence in Resuscitation: A Scoping Review

Introduction: Cardiac arrest is a significant cause of premature mortality and severe disability. Despite the death rate steadily decreasing over the previous decade, only 22% of survivors achieve good clinical status and only 25% of patients survive until their discharge from the hospital. The objective of this scoping review was to review relevant AI modalities and the main potential applications of AI in resuscitation. Methods: We conducted the literature search for related studies in PubMed, EMBASE, and Google Scholar. We included peer-reviewed publications and articles in the press, pooling and characterizing the data by their model types, goals, and benefits. Results: After identifying 268 original studies, we chose 59 original studies (reporting 1,817,419 patients) to include in the qualitative synthesis. AI-based methods appear to be superior to traditional methods in achieving high-level performance. Conclusion: AI might be useful in predicting cardiac arrest, heart rhythm disorders, and post-cardiac arrest outcomes, as well as in the delivery of drone-delivered defibrillators and notification of dispatchers. AI-powered technologies could be valuable assistants to continuously track patient conditions. Healthcare professionals should assist in the research and development of AI-powered technologies as well as their implementation into clinical practice.

Tree-Based Algorithms and Association Rule Mining for Predicting Patients’ Neurological Outcomes After First-Aid Treatment for an Out-of-Hospital Cardiac Arrest During COVID-19 Pandemic: Application of Data Mining

Objective

The authors performed several tree-based algorithms and an association rules mining as data mining tools to find useful determinants for neurological outcomes in out-of-hospital cardiac arrest (OHCA) patients as well as to assess the effect of the first-aid and basic characteristics in the EMS system.

Patients and Methods

This was a retrospective cohort study. The outcome was Cerebral Performance Categories grading on OHCA patients at hospital discharge. Decision tree-based models inclusive of C4.5 algorithm, classification and regression tree and random forest were built to determine an OHCA patient’s prognosis. Association rules mining was another data mining method which we used to find the combination of prognostic factors linked to the outcome.

Results

The total of 3520 patients were included in the final analysis. The mean age was 67.53 (±18.4) year-old and 63.4% were men. To overcome the imbalance outcome issue in machine learning, the random forest has a better predictive ability for OHCA patients in overall accuracy (91.19%), weighted precision (88.76%), weighted recall (91.20%) and F1 score (0.9) by oversampling adjustment. Under association rules mining, patients who had any witness on the spot when encountering OHCA or who had ever ROSC during first-aid would be highly correlated with good CPC prognosis.

Conclusion

The random forest has a better predictive ability for OHCA patients. This paper provides a role model applying several machine learning algorithms to the first-aid clinical assessment that will be promising combining with Artificial Intelligence for applying to emergency medical services.

Development and validation of machine learning-driven prediction model for serious bacterial infection among febrile children in emergency departments

Serious bacterial infection (SBI) in children, such as bacterial meningitis or sepsis, is an important condition that can lead to fatal outcomes. Therefore, since it is very important to accurately diagnose SBI, SBI prediction tools such as 'Refined Lab-score' or 'clinical prediction rule' have been developed and used. However, these tools can predict SBI only when there are values of all factors used in the tool, and if even one of them is missing, the tools become useless. Therefore, the purpose of this study was to develop and validate a machine learning-driven model to predict SBIs among febrile children, even with missing values. This was a multicenter retrospective observational study including febrile children <6 years of age who visited Emergency departments (EDs) of 3 different tertiary hospitals from 2016 to 2018. The SBI prediction model was trained with a derivation cohort (data from two hospitals) and externally tested with a validation cohort (data from a third hospital). A total of 11,973 and 2,858 patient records were included in the derivation and validation cohorts, respectively. In the derivation cohort, the area under the receiver operating characteristic curve (AUROC) of the RF model was 0.964 (95% confidence interval [CI], 0.943-0.986), and the area under the precision-recall curve (AUPRC) was 0.753 (95% CI, 0.681-0.824). The conventional LR (CLR) model showed corresponding values of 0.902 (95% CI, 0.894-0.910) and 0.573 (95% CI, 0.560-0.586), respectively. In the validation cohort, the AUROC (95% CI) of the RF model was 0.950 (95% CI, 0.945-0.956), the AUPRC was 0.605 (95% CI, 0.593-0.616), and the CLR presented corresponding values of 0.815 (95% CI, 0.789-0.841) and 0.586 (95% CI, 0.553-0.619), respectively. We developed a machine learning-driven prediction model for SBI among febrile children, which works robustly despite missing values. And it showed superior performance compared to CLR in both internal validation and external validation.

Comparison of Machine Learning Methods for Predicting Outcomes After In-Hospital Cardiac Arrest

OBJECTIVES

Prognostication of neurologic status among survivors of in-hospital cardiac arrests remains a challenging task for physicians. Although models such as the Cardiac Arrest Survival Post-Resuscitation In-hospital score are useful for predicting neurologic outcomes, they were developed using traditional statistical techniques. In this study, we derive and compare the performance of several machine learning models with each other and with the Cardiac Arrest Survival Post-Resuscitation In-hospital score for predicting the likelihood of favorable neurologic outcomes among survivors of resuscitation.

DESIGN

Analysis of the Get With The Guidelines-Resuscitation registry.

SETTING

Seven-hundred fifty-five hospitals participating in Get With The Guidelines-Resuscitation from January 1, 2001, to January 28, 2017.

PATIENTS

Adult in-hospital cardiac arrest survivors.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Of 117,674 patients in our cohort, 28,409 (24%) had a favorable neurologic outcome, as defined as survival with a Cerebral Performance Category score of less than or equal to 2 at discharge. Using patient characteristics, pre-existing conditions, prearrest interventions, and periarrest variables, we constructed logistic regression, support vector machines, random forests, gradient boosted machines, and neural network machine learning models to predict favorable neurologic outcome. Events prior to October 20, 2009, were used for model derivation, and all subsequent events were used for validation. The gradient boosted machine predicted favorable neurologic status at discharge significantly better than the Cardiac Arrest Survival Post-Resuscitation In-hospital score (C-statistic: 0.81 vs 0.73; p < 0.001) and outperformed all other machine learning models in terms of discrimination, calibration, and accuracy measures. Variables that were consistently most important for prediction across all models were duration of arrest, initial cardiac arrest rhythm, admission Cerebral Performance Category score, and age.

CONCLUSIONS

The gradient boosted machine algorithm was the most accurate for predicting favorable neurologic outcomes in in-hospital cardiac arrest survivors. Our results highlight the utility of machine learning for predicting neurologic outcomes in resuscitated patients.

Development and validation of a prehospital-stage prediction tool for traumatic brain injury: a multicentre retrospective cohort study in Korea

OBJECTIVES

Predicting diagnosis and prognosis of traumatic brain injury (TBI) at the prehospital stage is challenging; however, using comprehensive prehospital information and machine learning may improve the performance of the predictive model. We developed and tested predictive models for TBI that use machine learning algorithms using information that can be obtained in the prehospital stage.

DESIGN

This was a multicentre retrospective study.

SETTING AND PARTICIPANTS

This study was conducted at three tertiary academic emergency departments (EDs) located in an urban area of South Korea. The data from adult patients with severe trauma who were assessed by emergency medical service providers and transported to three participating hospitals between 2014 to 2018 were analysed.

RESULTS

We developed and tested five machine learning algorithms-logistic regression analyses, extreme gradient boosting, support vector machine, random forest and elastic net (EN)-to predict TBI, TBI with intracranial haemorrhage or injury (TBI-I), TBI with ED or admission result of admission or transferred (TBI with non-discharge (TBI-ND)) and TBI with ED or admission result of death (TBI-D). A total of 1169 patients were included in the final analysis, and the proportions of TBI, TBI-I, TBI-ND and TBI-D were 24.0%, 21.5%, 21.3% and 3.7%, respectively. The EN model yielded an area under receiver-operator curve of 0.799 for TBI, 0.844 for TBI-I, 0.811 for TBI-ND and 0.871 for TBI-D. The EN model also yielded the highest specificity and significant reclassification improvement. Variables related to loss of consciousness, Glasgow Coma Scale and light reflex were the three most important variables to predict all outcomes.

CONCLUSION

Our results inform the diagnosis and prognosis of TBI. Machine learning models resulted in significant performance improvement over that with logistic regression analyses, and the best performing model was EN.

Machine learning methods applied to triage in emergency services: A systematic review

BACKGROUND

In emergency services is important to accurately assess and classify symptoms, which may be improved with the help of technology. One mechanism that could help and improve predictions from health records or patient flow is machine learning (ML).

AIM

To analyse the effectiveness of ML systems in triage for making predictions at the emergency department in comparison with other triage scales/scores.

METHODS

Following the PRISMA recommendations, a systematic review was conducted using CINAHL, Cochrane, Cuiden, Medline and Scopus databases with the search equation "Machine learning AND triage AND emergency".

RESULTS

Eleven studies were identified. The studies show that the use of ML methods consistently predict important outcomes like mortality, critical care outcomes and admission, and the need for hospitalization in comparison with scales like Emergency Severity Index or others. Among the ML models considered, XGBoost and Deep Neural Networks obtained the highest levels of prediction accuracy, while Logistic Regression performed obtained the worst values.

CONCLUSIONS

Machine learning methods can be a good instrument for helping triage process with the prediction of important emergency variables like mortality or the need for critical care or hospitalization.

Discriminating head trauma outcomes using machine learning and genomics

A percentage of the population suffers prolonged and persistent post-concussion symptoms (PCS) following average head injuries or develops severe neurological dysfunction following minor head trauma. Genetic variants that may contribute to individual response to head trauma have been investigated in some studies, but to date none have explored the use of machine learning (ML) methods with genomic data to specifically explore outcomes of head trauma. Whole exome sequencing (WES) was completed for three groups of individuals (N = 60): (a) 16 individuals with severe neurological responses to minor head trauma, (b) 26 individuals with persistent PCS and (c) 18 individuals with normal recovery from concussion or mTBI. Gradient boosted tree algorithms were applied to the data using XGBoost. By using variants with CADD scores above 15 in the training set (randomly sampled 70%), we identified signatures that accurately distinguish to accurately distinguish the test groups with an average area under the curve (AUC) of 0.8 (SE = 0.019). Metrics including positive and negative prediction values, as well as kappa were all within acceptable range to support the prediction accuracy. This study illustrates how ML methods in combination with WES data have the potential to predict severe or prolonged responses to head trauma from healthy recovery. KEY MESSAGES: Linear association analysis has been inconclusive in concussion genetics. Non-linear methods as boosted trees can offer better insights in small samples. Strong discrimination trends can be achieved from exome data of cases and controls.

Development and validation of early prediction for neurological outcome at 90 days after return of spontaneous circulation in out-of-hospital cardiac arrest

AIM

To develop and validate a model for the early prediction of long-term neurological outcome in patients with non-traumatic out-of-hospital cardiac arrest (OHCA).

METHODS

We analysed multicentre OHCA registry data of adult patients with non-traumatic OHCA who experienced return of spontaneous circulation (ROSC) and had been admitted to the intensive care unit between 2013 and 2017. We allocated 1329 (2013-2015) and 1025 patients (2016-2017) to the derivation and validation sets, respectively. The primary outcome was the dichotomized cerebral performance category (CPC) at 90 days, defined as good (CPC 1-2) or poor (CPC 3-5). We developed 2 models: model 1 included variables without laboratory data, and model 2 included variables with laboratory data available immediately after ROSC. Logistic regression with least absolute shrinkage and selection operator regularization was employed for model development. Measures of discrimination, accuracy, and calibration (C-statistics, Brier score, calibration plot, and net benefit) were assessed in the validation set.

RESULTS

The C-statistic (95% confidence intervals) of models 1 and 2 in the validation set was 0.947 (0.930-0.964) and 0.950 (0.934-0.966), respectively. The Brier score of models 1 and 2 in the validation set was 0.0622 and 0.0606, respectively. The calibration plot showed that both models were well-calibrated to the observed outcome. Decision curve analysis indicated that model 2 was similar to model 1.

CONCLUSION

The prediction tool containing detailed in-hospital information showed good performance for predicting neurological outcome at 90 days immediately after ROSC in patients with OHCA.

Use of Time-to-Event Analysis to Develop On-Scene Return of Spontaneous Circulation Prediction for Out-of-Hospital Cardiac Arrest Patients

STUDY OBJECTIVE

We aimed to train and validate the time to on-scene return of spontaneous circulation prediction models using time-to-event analysis among out-of-hospital cardiac arrest patients.

METHODS

Using a Korean population-based out-of-hospital cardiac arrest registry, we selected a total of 105,215 adults with presumed cardiac etiologies between 2013 and 2018. Patients from 2013 to 2017 and from 2018 were analyzed for training and test, respectively. We developed 4 time-to-event analyzing models (Cox proportional hazard [Cox], random survival forest, extreme gradient boosting survival, and DeepHit) and 4 classification models (logistic regression, random forest, extreme gradient boosting, and feedforward neural network). Patient characteristics and Utstein elements collected at the scene were used as predictors. Discrimination and calibration were evaluated by Harrell's C-index and integrated Brier score.

RESULTS

Among the 105,215 patients (mean age 70 years and 64% men), 86,314 and 18,901 patients belonged to the training and test sets, respectively. On-scene return of spontaneous circulation was achieved in 5,240 (6.1%) patients in the former set and 1,709 (9.0%) patients in the latter. The proportion of emergency medical services (EMS) management was higher and scene time interval longer in the latter. Median time from EMS scene arrival to on-scene return of spontaneous circulation was 8 minutes for both datasets. Classification models showed similar discrimination and poor calibration power compared to survival models; Cox showed high discrimination with the best calibration (C-index [95% confidence interval]: 0.873 [0.865 to 0.882]; integrated Brier score at 30 minutes: 0.060).

CONCLUSION

Incorporating time-to-event analysis could lead to improved performance in prediction models and contribute to personalized field EMS resuscitation decisions.

Prediction of Neurological Outcomes in Out-of-hospital Cardiac Arrest Survivors Immediately after Return of Spontaneous Circulation: Ensemble Technique with Four Machine Learning Models

BACKGROUND

We performed this study to establish a prediction model for 1-year neurological outcomes in out-of-hospital cardiac arrest (OHCA) patients who achieved return of spontaneous circulation (ROSC) immediately after ROSC using machine learning methods.

METHODS

We performed a retrospective analysis of an OHCA survivor registry. Patients aged ≥ 18 years were included. Study participants who had registered between March 31, 2013 and December 31, 2018 were divided into a develop dataset (80% of total) and an internal validation dataset (20% of total), and those who had registered between January 1, 2019 and December 31, 2019 were assigned to an external validation dataset. Four machine learning methods, including random forest, support vector machine, ElasticNet and extreme gradient boost, were implemented to establish prediction models with the develop dataset, and the ensemble technique was used to build the final prediction model. The prediction performance of the model in the internal validation and the external validation dataset was described with accuracy, area under the receiver-operating characteristic curve, area under the precision-recall curve, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Futhermore, we established multivariable logistic regression models with the develop set and compared prediction performance with the ensemble models. The primary outcome was an unfavorable 1-year neurological outcome.

RESULTS

A total of 1,207 patients were included in the study. Among them, 631, 139, and 153 were assigned to the develop, the internal validation and the external validation datasets, respectively. Prediction performance metrics for the ensemble prediction model in the internal validation dataset were as follows: accuracy, 0.9620 (95% confidence interval [CI], 0.9352-0.9889); area under receiver-operator characteristics curve, 0.9800 (95% CI, 0.9612-0.9988); area under precision-recall curve, 0.9950 (95% CI, 0.9860-1.0000); sensitivity, 0.9594 (95% CI, 0.9245-0.9943); specificity, 0.9714 (95% CI, 0.9162-1.0000); PPV, 0.9916 (95% CI, 0.9752-1.0000); NPV, 0.8718 (95% CI, 0.7669-0.9767). Prediction performance metrics for the model in the external validation dataset were as follows: accuracy, 0.8509 (95% CI, 0.7825-0.9192); area under receiver-operator characteristics curve, 0.9301 (95% CI, 0.8845-0.9756); area under precision-recall curve, 0.9476 (95% CI, 0.9087-0.9867); sensitivity, 0.9595 (95% CI, 0.9145-1.0000); specificity, 0.6500 (95% CI, 0.5022-0.7978); PPV, 0.8353 (95% CI, 0.7564-0.9142); NPV, 0.8966 (95% CI, 0.7857-1.0000). All the prediction metrics were higher in the ensemble models, except NPVs in both the internal and the external validation datasets.

CONCLUSION

We established an ensemble prediction model for prediction of unfavorable 1-year neurological outcomes in OHCA survivors using four machine learning methods. The prediction performance of the ensemble model was higher than the multivariable logistic regression model, while its performance was slightly decreased in the external validation dataset.

Machine Learning-Based Prediction of Korean Triage and Acuity Scale Level in Emergency Department Patients

Objectives

Triage is a process to accurately assess and classify symptoms to identify and provide rapid treatment to patients. The Korean Triage and Acuity Scale (KTAS) is used as a triage instrument in all emergency centers. The aim of this study was to train and compare machine learning models to predict KTAS levels.

Methods

This was a cross-sectional study using data from a single emergency department of a tertiary university hospital. Information collected during triage was used in the analysis. Logistic regression, random forest, and XGBoost were used to predict the KTAS level.

Results

The models with the highest area under the receiver operating characteristic curve (AUROC) were the random forest and XGBoost models trained on the entire dataset (AUROC = 0.922, 95% confidence interval 0.917-0.925 and AUROC = 0.922, 95% confidence interval 0.918-0.925, respectively). The AUROC of the models trained on the clinical data was higher than that of models trained on text data only, but the models trained on all variables had the highest AUROC among similar machine learning models.

Conclusions

Machine learning can robustly predict the KTAS level at triage, which may have many possibilities for use, and the addition of text data improves the predictive performance compared to that achieved by using structured data alone.