Intermittent manual measurement of vital signs may not rapidly predict sepsis development in febrile patients admitted to the emergency department (ED). We aimed to evaluate the predictive performance of a wireless monitoring device that continuously measures heart rate (HR) and respiratory rate (RR) and a machine learning analysis in febrile but stable patients in the ED. We analysed 468 patients (age, ≥18 years; training set, n = 277; validation set, n = 93; test set, n = 98) having fever (temperature >38 °C) and admitted to the isolation care unit of the ED. The AUROC of the fragmented model with device data was 0.858 (95% confidence interval [CI], 0.809–0.908), and that with manual data was 0.841 (95% CI, 0.789–0.893). The AUROC of the accumulated model with device data was 0.861 (95% CI, 0.811–0.910), and that with manual data was 0.853 (95% CI, 0.803–0.903). Fragmented and accumulated models with device data detected clinical deterioration in febrile patients at risk of septic shock 9 h and 5 h 30 min earlier, respectively, than those with manual data. Continuous vital sign monitoring using a wearable device could accurately predict clinical deterioration and reduce the time to recognise potential clinical deterioration in stable ED patients with fever.

Models for predicting acute myocardial infarction (AMI) at the prehospital stage were developed and their efficacy compared, based on variables identified from a nationwide systematic emergency medical service (EMS) registry using conventional statistical methods and machine learning algorithms. Patients in the EMS cardiovascular registry aged >15 years who were transferred from the public EMS to emergency departments in Korea from January 2016 to December 2018 were enrolled. Two datasets were constructed according to the hierarchical structure of the registry. A total of 184,577 patients (Dataset 1) were included in the final analysis. Among them, 72,439 patients (Dataset 2) were suspected to have AMI at prehospital stage. Between the models derived using the conventional logistic regression method, the B-type model incorporated AMI-specific variables from the A-type model and exhibited a superior discriminative ability (p = 0.02). The models that used extreme gradient boosting and a multilayer perceptron yielded a higher predictive performance than the conventional logistic regression-based models for analyses that used both datasets. Each machine learning algorithm yielded different classification lists of the 10 most important features. Therefore, prediction models that use nationwide prehospital data and are developed with appropriate structures can improve the identification of patients who require timely AMI management.

This study develops a CPR prediction model using initial admission data in the emergency department. We retrospectively analyzed EMR data from Severance Hospital (2018-2022), excluding patients aged under 18 or without initial vital signs and KTAS, and those with DNR orders. We developed and compared two models: one using patient information and initial vital signs and another incorporating KTAS. The model with KTAS showed significantly better predictive performance. Our findings suggest that combining vital signs and KTAS can effectively predict CPR, helping to prevent cardiac arrest.

Electronic Health Record (EHR) provides abundant information through various modalities. However, learning multi-modal EHR is currently facing two major challenges, namely, 1) data embedding and 2) cases with missing modality. A lack of shared embedding function across modalities can discard the temporal relationship between different EHR modalities. On the other hand, most EHR studies are limited to relying only on EHR Times-series, and therefore, missing modality in EHR has not been well-explored. Therefore, in this study, we introduce a Unified Multi-modal Set Embedding (UMSE) and Modality-Aware Attention (MAA) with Skip Bottleneck (SB). UMSE treats all EHR modalities without a separate imputation module or error-prone carry-forward, whereas MAA with SB learns missing modal EHR with effective modality-aware attention. Our model outperforms other baseline models in mortality, vasopressor need, and intubation need prediction with the MIMIC-IV dataset.

In this study, we analyzed voice of customer (VOC) data for an AI-based early warning system from healthcare providers using the BERTopic framework for effective topic modeling. A preprocessing pipeline was implemented, incorporating techniques such as lowercasing, stopword removal, and tokenization to prepare the text for analysis. To refine the model’s performance, hyperparameter optimization was conducted using Optuna, with a primary focus on maximizing the Silhouette Score and minimizing the Davies-Bouldin Index, while also assessing the diversity of the generated topics. Ultimately, we identified five major topics that encapsulate key themes within the VOC data.