No abstract available.

While digital health solutions have shown good outcomes in numerous studies, the adoption of digital health solutions in clinical practice faces numerous challenges. To prepare for widespread adoption of digital health, stakeholders in digital health will need to establish an objective evaluation process, consider uncertainty through critical evaluation, be aware of inequity, and consider patient engagement. By “making friends” with digital health, health care can be improved for patients.

Objectives: Many deep learning-based predictive models evaluate the waveforms of electrocardiograms (ECGs). Because deep learning-based models are data-driven, large and labeled biosignal datasets are required. Most individual researchers find it difficult to collect adequate training data. We suggest that transfer learning can be used to solve this problem and increase the effectiveness of biosignal analysis. Methods: We applied the weights of a pretrained model to another model that performed a different task (i.e., transfer learning). We used 2,648,100 unlabeled 8.2-second-long samples of ECG II data to pretrain a convolutional autoencoder (CAE) and employed the CAE to classify 12 ECG rhythms within a dataset, which had 10,646 10-second-long 12-lead ECGs with 11 rhythm labels. We split the datasets into training and test datasets in an 8:2 ratio. To confirm that transfer learning was effective, we evaluated the performance of the classifier after the proposed transfer learning, random initialization, and two-dimensional transfer learning as the size of the training dataset was reduced. All experiments were repeated 10 times using a bootstrapping method. The CAE performance was evaluated by calculating the mean squared errors (MSEs) and that of the ECG rhythm classifier by deriving F1-scores. Results: The MSE of the CAE was 626.583. The mean F1-scores of the classifiers after bootstrapping of 100%, 50%, and 25% of the training dataset were 0.857, 0.843, and 0.835, respectively, when the proposed transfer learning was applied and 0.843, 0.831, and 0.543, respectively, after random initialization was applied. Conclusions Transfer learning effectively overcomes the data shortages that can compromise ECG domain analysis by deep learning.

Background: Most studies of systematic drug repositioning have used drug-oriented data such as chemical structures, gene expression patterns, and adverse effect profiles. As it is often difficult to prove repositioning candidates’ effectiveness in real-world clinical settings, we used patient-centered real-world data for screening repositioning candidate drugs for multiple diseases simultaneously, especially for diabetic complications. Methods: Using the National Health Insurance Service-National Sample Cohort (2002 to 2013), we analyzed claims data of 43,048 patients with type 2 diabetes mellitus (age ≥40 years). To find repositioning candidate disease-drug pairs, a nested case-control study was used for 29 pairs of diabetic complications and the drugs that met our criteria. To validate this study design, we conducted an external validation for a selected candidate pair using electronic health records. Results: We found 24 repositioning candidate disease-drug pairs. In the external validation study for the candidate pair cerebral infarction and glycopyrrolate, we found that glycopyrrolate was associated with decreased risk of cerebral infarction (hazard ratio, 0.10; 95% confidence interval, 0.02 to 0.44). Conclusion To reduce risks of diabetic complications, it would be possible to consider these candidate drugs instead of other drugs, given the same indications. Moreover, this methodology could be applied to diseases other than diabetes to discover their repositioning candidates, thereby offering a new approach to drug repositioning.

Objectives: Cheyne-Stokes respiration (CSR) is frequently found in critically ill patients and is associated with poor prognosis. However, CSR has not been evaluated in neurocritical patients. This study investigated the frequency and prognostic impact of CSR in neurocritical patients using biosignal big data obtained from intensive care units. Methods: This study included all patients who received neurocritical care at the tertiary hospital from January 2018 to December 2019. Clinical information and biosignal data of intensive care units were used and analyzed. The respiratory curve was visually assessed to determine whether CSR and obstructive sleep apnea (OSA) were present, and a heart rate variability (HRV) was obtained from the electrocardiogram. Results: CSR was confirmed in 166 of 406 patients (40.9%). Patients with CSR were older, had a higher frequency of cardiovascular risk factors as well as heart failure, and had a poor outcome (modified Rankin scale ≥4). As a result of multiple regression analysis adjusted for other variables, CSR was significantly associated with poor outcome with an odds ratio of 2.27 times higher (95% confidence interval 1.25–4.14, p=0.007). HRV analysis demonstrated that CSR and OSA had distinct autonomic characteristics. Conclusions This study first revealed the substantial frequency of CSR in neurocritical patients and suggests that it can be used as a predictor of poor prognosis in neurocritical care.

Objectives: Early hemorrhage detection in intensive care units (ICUs) enables timely intervention and reduces the risk of irreversible outcomes. In this study, we aimed to develop a machine learning model to predict hemorrhage by learning the patterns of continuously changing, real-world clinical data. Methods: We used the Medical Information Mart for Intensive Care databases (MIMIC-III and MIMIC-IV). A recurrent neural network was used to predict severe hemorrhage in the ICU. We developed three machine learning models with an increasing number of input features and levels of complexity: model 1 (11 features), model 2 (18 features), and model 3 (27 features). MIMIC-III was used for model training, and MIMIC-IV was split for internal validation. Using the model with the highest performance, external verification was performed using data from a subgroup extracted from the eICU Collaborative Research Database. Results: We included 5,670 ICU admissions, with 3,150 in the training set and 2,520 in the internal test set. A positive correlation was found between model complexity and performance. As a measure of performance, three models developed with an increasing number of features showed area under the receiver operating characteristic (AUROC) curve values of 0.61–0.94 according to the range of input data. In the subgroup extracted from the eICU database for external validation, an AUROC value of 0.74 was observed. Conclusions Machine learning models that rely on real clinical data can be used to predict patients at high risk of bleeding in the ICU.

BACKGROUND: Injury severity scoring systems that quantify and predict trauma outcomes have not been established in Korea. This study was designed to determine the best system for use in the Korean trauma population. METHODS: We collected and analyzed the data from trauma patients admitted to our institution from January 2010 to December 2014. Injury Severity Score (ISS), Revised Trauma Score (RTS), and Trauma and Injury Severity Score (TRISS) were calculated based on the data from the enrolled patients. Area under the receiver operating characteristic (ROC) curve (AUC) for the prediction ability of each scoring system was obtained, and a pairwise comparison of ROC curves was performed. Additionally, the cut-off values were estimated to predict mortality, and the corresponding accuracy, positive predictive value, and negative predictive value were obtained. RESULTS: A total of 7,120 trauma patients (6,668 blunt and 452 penetrating injuries) were enrolled in this study. The AUCs of ISS, RTS, and TRISS were 0.866, 0.894, and 0.942, respectively, and the prediction ability of the TRISS was significantly better than the others (p < 0.001, respectively). The cut-off value of the TRISS was 0.9082, with a sensitivity of 81.9% and specificity of 92.0%; mortality was predicted with an accuracy of 91.2%; its positive predictive value was the highest at 46.8%. CONCLUSIONS: The results of our study were based on the data from one institution and suggest that the TRISS is the best prediction model of trauma outcomes in the current Korean population. Further study is needed with more data from multiple centers in Korea.
Area Under Curve ; Humans ; Injury Severity Score ; Korea ; Mortality ; ROC Curve ; Sensitivity and Specificity ; Trauma Centers

OBJECTIVES: Biosignal data include important physiological information. For that reason, many devices and systems have been developed, but there has not been enough consideration of how to collect and integrate raw data from multiple systems. To overcome this limitation, we have developed a system for collecting and integrating biosignal data from two patient monitoring systems. METHODS: We developed an interface to extract biosignal data from Nihon Kohden and Philips monitoring systems. The Nihon Kohden system has a central server for the temporary storage of raw waveform data, which can be requested using the HL7 protocol. However, the Philips system used in our hospital cannot save raw waveform data. Therefore, our system was connected to monitoring devices using the RS232 protocol. After collection, the data were transformed and stored in a unified format. RESULTS: From September 2016 to August 2017, we collected approximately 117 patient-years of waveform data from 1,268 patients in 79 beds of five intensive care units. Because the two systems use the same data storage format, the application software could be run without compatibility issues. CONCLUSIONS: Our system collects biosignal data from different systems in a unified format. The data collected by the system can be used to develop algorithms or applications without the need to consider the source of the data.
Electrocardiography ; Humans ; Information Storage and Retrieval ; Intensive Care Units ; Monitoring, Physiologic* ; Photoplethysmography

Objectives: Drug-induced QT prolongation can lead to life-threatening arrhythmia. In the intensive care unit (ICU), various drugs are administered concurrently, which can increase the risk of QT prolongation. However, no well-validated method to evaluate the risk of QT prolongation in real-world clinical practice has been established. We developed a risk scoring model to continuously evaluate the quantitative risk of QT prolongation in real-world clinical practice in the ICU. Methods: Continuous electrocardiogram (ECG) signals measured by patient monitoring devices and Electronic Medical Records data were collected for ICU patients. QT and RR intervals were measured from raw ECG data, and a corrected QT interval (QTc) was calculated by Bazett’s formula. A case-crossover study design was adopted. A case was defined as an occurrence of QT prolongation ≥12 hours after any previous QT prolongation. The patients served as their own controls. Conditional logistic regression was conducted to analyze prescription, surgical history, and laboratory test data. Based on the regression analysis, a QTc prolongation risk scoring model was established. Results: In total, 811 ICU patients who experienced QT prolongation were included in this study. Prescription information for 13 drugs was included in the risk scoring model. In the validation dataset, the high-risk group showed a higher rate of QT prolongation than the low-and low moderate-risk groups. Conclusions Our proposed model may facilitate risk stratification for QT prolongation during ICU care as well as the selection of appropriate drugs to prevent QT prolongation.

OBJECTIVES: Biosignal data captured by patient monitoring systems could provide key evidence for detecting or predicting critical clinical events; however, noise in these data hinders their use. Because deep learning algorithms can extract features without human annotation, this study hypothesized that they could be used to screen unacceptable electrocardiograms (ECGs) that include noise. To test that, a deep learning-based model for unacceptable ECG screening was developed, and its screening results were compared with the interpretations of a medical expert. METHODS: To develop and apply the screening model, we used a biosignal database comprising 165,142,920 ECG II (10-second lead II electrocardiogram) data gathered between August 31, 2016 and September 30, 2018 from a trauma intensive-care unit. Then, 2,700 and 300 ECGs (ratio of 9:1) were reviewed by a medical expert and used for 9-fold cross-validation (training and validation) and test datasets. A convolutional neural network-based model for unacceptable ECG screening was developed based on the training and validation datasets. The model exhibiting the lowest cross-validation loss was subsequently selected as the final model. Its performance was evaluated through comparison with a test dataset. RESULTS: When the screening results of the proposed model were compared to the test dataset, the area under the receiver operating characteristic curve and the F1-score of the model were 0.93 and 0.80 (sensitivity = 0.88, specificity = 0.89, positive predictive value = 0.74, and negative predictive value = 0.96). CONCLUSIONS: The deep learning-based model developed in this study is capable of detecting and screening unacceptable ECGs efficiently.
Dataset ; Electrocardiography ; Humans ; Learning ; Mass Screening ; Monitoring, Physiologic ; Noise ; ROC Curve ; Sensitivity and Specificity ; Signal Detection, Psychological