Axial length, more than aging, decreases the thickness and superficial vessel density of retinal nerve fiber layer in non-glaucomatous eyes.

PURPOSE: This study seeks to build a normative database for the vessel density of the superficial retina (SVD) and evaluate how changes and trends in the retinal microvasculature may be influenced by age and axial length (AL) in non-glaucomatous eyes, as measured with optical coherence tomography angiography (OCTA). METHODS: We included 500 eyes of 290 healthy subjects visiting a county hospital. Each participant underwent comprehensive ophthalmological examinations and OCTA to measure the SVD and thickness of the macular and peripapillary areas. To analyze correlations between SVD and age or AL, multivariable linear regression models with generalized estimating equations were applied. RESULTS: Age was negatively correlated with the SVD of the superior, central, and inferior macular areas and the superior peripapillary area, with a decrease rate of 1.06%, 1.36%, 0.84%, and 0.66% per decade, respectively. However, inferior peripapillary SVD showed no significant correlation with age. AL was negatively correlated with the SVD of the inferior macular area and the superior and inferior peripapillary areas, with coefficients of -0.522%/mm, -0.733%/mm, and -0.664%/mm, respectively. AL was also negatively correlated with the thickness of the retinal nerve fiber layer and inferior ganglion cell complex (p = 0.004). CONCLUSION: Age and AL were the two main factors affecting changes in SVD. Furthermore, AL, a relative term to represent the degree of myopia, had a greater effect than age and showed a more significant effect on thickness than on SVD. This relationship has important implications because myopia is a significant issue in modern cities.

Wearable mobile health device for monitoring postoperative ambulation among patients with colorectal cancer undergoing minimally invasive surgery: A prospective comparison study.

BACKGROUND: This study aimed to evaluate the feasibility of using mHealth devices for monitoring postoperative ambulation among patients with colorectal cancer undergoing minimally invasive surgery (MIS). METHODS: Patients with colorectal cancer undergoing MIS were prospectively recruited to wear mHealth devices for recording postoperative ambulation between October 2018 and January 2021. The primary outcome was the compliance by evaluating the weekly submission rate of step counts. The secondary outcome was the association of weekly step counts and postoperative length of stay. RESULTS: Of 107 eligible patients, 53 patients wore mHealth devices, whereas 54 patients did not. The average weekly submission rate was 72.6% for the first month after surgery. The total step counts <4000 or >10 000 in the postoperative week one were negatively associated with postoperative length of stay (ß = -2.874, p = 0.038). CONCLUSIONS: mHealth devices provide an objective assessment of postoperative ambulation among patients with colorectal cancer undergoing MIS. CLINICAL TRIAL REGISTRATION: NCT03277235.

Machine learning models to evaluate mortality in pediatric patients with pneumonia in the intensive care unit.

OBJECTIVES: This study aimed to predict mortality in children with pneumonia who were admitted to the intensive care unit (ICU) to aid decision-making. STUDY DESIGN: Retrospective cohort study conducted at a single tertiary hospital. PATIENTS: This study included children who were admitted to the pediatric ICU at the National Taiwan University Hospital between 2010 and 2019 due to pneumonia. METHODOLOGY: Two prediction models were developed using tree-structured machine learning algorithms. The primary outcomes were ICU mortality and 24-h ICU mortality. A total of 33 features, including demographics, underlying diseases, vital signs, and laboratory data, were collected from the electronic health records. The machine learning models were constructed using the development data set, and performance matrices were computed using the holdout test data set. RESULTS: A total of 1231 ICU admissions of children with pneumonia were included in the final cohort. The area under the receiver operating characteristic curves (AUROCs) of the ICU mortality model and 24-h ICU mortality models was 0.80 (95% confidence interval [CI], 0.69-0.91) and 0.92 (95% CI, 0.86-0.92), respectively. Based on feature importance, the model developed in this study tended to predict increased mortality for the subsequent 24 h if a reduction in the blood pressure, peripheral capillary oxygen saturation (SpO2), or higher partial pressure of carbon dioxide (PCO2) were observed. CONCLUSIONS: This study demonstrated that the machine learning models for predicting ICU mortality and 24-h ICU mortality in children with pneumonia have the potential to support decision-making, especially in resource-limited settings.

Sleep, physical activity and panic attacks: A two-year prospective cohort study using smartwatches, deep learning and an explainable artificial intelligence model.

BACKGROUND: Sleep and physical activity suggestions for panic disorder (PD) are critical but less surveyed. This two-year prospective cohort study aims to predict panic attacks (PA), state anxiety (SA), trait anxiety (TA) and panic disorder severity (PDS) in the upcoming week. METHODS: We enrolled 114 PD patients from one general hospital. Data were collected using the DSM-5, the MINI, clinical app questionnaires (BDI, BAI, PDSS-SR, STAI) and wearable devices recording daily sleep, physical activity and heart rate from 16 June 2020 to 10 June 2022. Our teams applied RNN, LSTM, GRU deep learning and SHAP explainable methods to analyse the data. RESULTS: The 7-day prediction accuracies for PA, SA, TA, and PDS were 92.8 %, 83.6 %, 87.2 %, and 75.6 % from the LSTM model. Using the SHAP explainable model, higher initial BDI or BAI score and comorbidities with depressive disorder, generalized anxiety disorder or agoraphobia predict a higher chance of PA. However, PA decreased under the following conditions: daily average heart rate, 72-87 bpm; maximum heart rate, 100-145 bpm; resting heart rate, 55-60 bpm; daily climbing of more than nine floors; total sleep duration between 6 h 23 min and 10 h 50 min; deep sleep, >50 min; and awake duration, <53 min. LIMITATIONS: Moderate sample size and self-report questionnaires were the limitations. CONCLUSIONS: Deep learning predicts recurrent PA and various anxiety domains with 75.6-92.8 % accuracy. Recurrent PA decreases under adequate daily sleep and physical activity.

Enhancing panic disorder treatment with mobile-aided case management: an exploratory study based on a 3-year cohort analysis.

Background: Individuals with panic disorder frequently face ongoing symptoms, suboptimal treatment adherence, and increased relapse rates. Although mobile health interventions have shown promise in improving treatment outcomes for numerous mental health conditions, their effectiveness, specifically for panic disorder, has yet to be determined. Objective: This study investigates the effects of a mobile-aided case management program on symptom reduction and quality of care among individuals with panic disorder. Methods: This 3-year cohort study enrolled 138 participants diagnosed with panic disorder. One hundred and eight participants joined the mobile-aided case management group and 30 in the treatment-as-usual group. Data were collected at baseline, 3-month, 6-month, and 12-month treatment checkpoints using self-report questionnaires, in-depth interviews, direct observation, and medical record analysis. Results: During the maintenance treatment phase, the mobile-assisted case management group decreased both panic severity (p = 0.008) and state anxiety (p = 0.016) more than the control group at 6 months. Participants who underwent case management experienced enhanced control over panic symptoms, heightened self-awareness, and elevated interpersonal support. Conclusion: The mobile-aided case management is beneficial in managing panic disorder, especially maintenance treatment.

Applying a Smartwatch to Predict Work-related Fatigue for Emergency Healthcare Professionals: Machine Learning Method.

INTRODUCTION: Healthcare professionals frequently experience work-related fatigue, which may jeopardize their health and put patient safety at risk. In this study, we applied a machine learning (ML) approach based on data collected from a smartwatch to construct prediction models of work-related fatigue for emergency clinicians. METHODS: We conducted this prospective study at the emergency department (ED) of a tertiary teaching hospital from March 10-June 20, 2021, where we recruited physicians, nurses, and nurse practitioners. All participants wore a commercially available smartwatch capable of measuring various physiological data during the experiment. Participants completed the Multidimensional Fatigue Inventory (MFI) web form before and after each of their work shifts. We calculated and labeled the before-and-after-shift score differences between each pair of scores. Using several tree-based algorithms, we constructed the prediction models based on features collected from the smartwatch. Records were split into training/validation and testing sets at a 70:30 ratio, and we evaluated the performances using the area under the curve (AUC) measure of receiver operating characteristic on the test set. RESULTS: In total, 110 participants were included in this study, contributing to a set of 1,542 effective records. Of these records, 85 (5.5%) were labeled as having work-related fatigue when setting the MFI difference between two standard deviations as the threshold. The mean age of the participants was 29.6. Most of the records were collected from nurses (87.7%) and females (77.5%). We selected a union of 31 features to construct the models. For total participants, CatBoost classifier achieved the best performances of AUC (0.838, 95% confidence interval [CI] 0.742-0.918) to identify work-related fatigue. By focusing on a subgroup of nurses <35 years in age, XGBoost classifier obtained excellent performance of AUC (0.928, 95% CI 0.839-0.991) on the test set. CONCLUSION: By using features derived from a smartwatch, we successfully built ML models capable of classifying the risk of work-related fatigue in the ED. By collecting more data to optimize the models, it should be possible to use smartwatch-based ML models in the future to predict work-related fatigue and adopt preventive measures for emergency clinicians.

Evaluating the Potential of Machine Learning and Wearable Devices in End-of-Life Care in Predicting 7-Day Death Events Among Patients With Terminal Cancer: Cohort Study.

BACKGROUND: An accurate prediction of mortality in end-of-life care is crucial but presents challenges. Existing prognostic tools demonstrate moderate performance in predicting survival across various time frames, primarily in in-hospital settings and single-time evaluations. However, these tools may fail to capture the individualized and diverse trajectories of patients. Limited evidence exists regarding the use of artificial intelligence (AI) and wearable devices, specifically among patients with cancer at the end of life. OBJECTIVE: This study aimed to investigate the potential of using wearable devices and AI to predict death events among patients with cancer at the end of life. Our hypothesis was that continuous monitoring through smartwatches can offer valuable insights into the progression of patients at the end of life and enable the prediction of changes in their condition, which could ultimately enhance personalized care, particularly in outpatient or home care settings. METHODS: This prospective study was conducted at the National Taiwan University Hospital. Patients diagnosed with cancer and receiving end-of-life care were invited to enroll in wards, outpatient clinics, and home-based care settings. Each participant was given a smartwatch to collect physiological data, including steps taken, heart rate, sleep time, and blood oxygen saturation. Clinical assessments were conducted weekly. The participants were followed until the end of life or up to 52 weeks. With these input features, we evaluated the prediction performance of several machine learning-based classifiers and a deep neural network in 7-day death events. We used area under the receiver operating characteristic curve (AUROC), F1-score, accuracy, and specificity as evaluation metrics. A Shapley additive explanations value analysis was performed to further explore the models with good performance. RESULTS: From September 2021 to August 2022, overall, 1657 data points were collected from 40 patients with a median survival time of 34 days, with the detection of 28 death events. Among the proposed models, extreme gradient boost (XGBoost) yielded the best result, with an AUROC of 96%, F1-score of 78.5%, accuracy of 93%, and specificity of 97% on the testing set. The Shapley additive explanations value analysis identified the average heart rate as the most important feature. Other important features included steps taken, appetite, urination status, and clinical care phase. CONCLUSIONS: We demonstrated the successful prediction of patient deaths within the next 7 days using a combination of wearable devices and AI. Our findings highlight the potential of integrating AI and wearable technology into clinical end-of-life care, offering valuable insights and supporting clinical decision-making for personalized patient care. It is important to acknowledge that our study was conducted in a relatively small cohort; thus, further research is needed to validate our approach and assess its impact on clinical care. TRIAL REGISTRATION: ClinicalTrials.gov NCT05054907; https://classic.clinicaltrials.gov/ct2/show/NCT05054907.

Outcomes of pediatric community-acquired pneumonia before and after national pneumococcal immunization in Taiwan.

OBJECTIVE: In Taiwan, the incidence of invasive pneumococcal disease (IPD) in children declined after the catch-up primary vaccination programs and the full national immunization program (NIP) with PCV13. The objective of the study was to investigate the clinical outcomes of pediatric community-acquired pneumonia (CAP) before and after the NIP. METHODS: The study included patients aged 3 months to 17 years who were diagnosed with CAP and treated at the National Taiwan University Hospital between 2007 and 2019. Patients were assigned to three birth cohorts according to their birth years and vaccination eligibility: non-NIP, catch-up, and full NIP. We compared the rates of severe outcomes, including case fatality and pathogens. RESULTS: A total of 6557 patients who met the CAP criteria were enrolled during the study period. The case-fatality rate decreased from 3.2% (94/2984) in the non-NIP cohort to 0.3% (7/2176) in the catch-up cohort and 0.8% (11/1397) in the full NIP cohort (p < 0.001). Furthermore, there was a significant decrease in invasive ventilation from the non-NIP (17.9%) to both catch-up (6.8%) and full NIP cohorts (9.1%). The rate of IPD declined from the non-NIP cohort to the catch-up cohort (1.8% vs. 0.6%, p < 0.001) and from the catch-up to the full NIP cohort (0.6% vs. 0.07%, p = 0.014). In contrast, the rates of infections with other pathogens increased after NIP. CONCLUSION: The introduction of PCV13 showed significant reduction in case-fatality and IPD rates. The increasing rates of other pathogens warrant further surveillance for their clinical significance.

Prognostic Value of Pace Variability, a Novel 6MWT-Derived Feature, in Patients with Chronic Obstructive Pulmonary Disease.

Purpose: The 6-minute walk test (6MWT) is often used to evaluate chronic obstructive pulmonary disease (COPD) patients' functional capacity, with 6-minute walk distance (6MWD) and related measures being linked to mortality and hospitalizations. This study investigates the prognostic value of pace variability, a significant indicator in sports medicine, during the 6MWT for COPD patients. Patients and Methods: We retrospectively screened consecutive COPD patients who had been prospectively enrolled in a pay-for-performance program from January 2019 to May 2020 to determine their eligibility. Patient characteristics, including demographics, exacerbation history, and 6MWT data, were analyzed to investigate their potential associations with prognosis. The primary outcome was a composite of adverse events, including overall mortality or hospitalizations due to exacerbations during a 1-year follow-up period. To analyze the 6MWT data, we divided it into three 2-minute epochs and calculated the average walk speed for each epoch. We defined pace variability as the difference between the maximum and minimum average speed in a single 2-minute epoch, divided by the average speed for the entire 6-minute walk test. Results: A total of 163 patients with COPD were included in the study, and 19 of them (12%) experienced the composite adverse outcome. Multivariable logistic regression analyses revealed that two predictors were independently associated with the composite outcome: % predicted 6MWD <72 (adjusted odds ratio [aOR] 7.080; 95% confidence interval [CI] 1.481-33.847) and pace variability ≥0.39 (aOR 9.444; 95% CI 2.689-33.170). Patients with either of these adverse prognostic features had significantly worse composite outcome-free survival, with both log-rank P values less than 0.005. Notably, COPD patients with both adverse features experienced an especially poor outcome after 1 year. Conclusion: Patients with COPD who exhibited greater pace variability during the 6MWT had a significantly higher risk of overall mortality and COPD-related hospitalizations, indicating a worse prognosis.

Neurobiological mechanisms of dialectical behavior therapy and Morita therapy, two psychotherapies inspired by Zen.

Psychotherapy is a learning process. Updating the prediction models of the brain may be the mechanism underlying psychotherapeutic changes. Although developed in different eras and cultures, dialectical behavior therapy (DBT) and Morita therapy are influenced by Zen principles, and both emphasize the acceptance of reality and suffering. This article reviews these two treatments, their common and distinct therapeutic factors, and their neuroscientific implications. Additionally, it proposes a framework that includes the predictive function of the mind, constructed emotions, mindfulness, therapeutic relationship, and changes enabled via reward predictions. Brain networks, including the Default Mode Network (DMN), amygdala, fear circuitry, and reward pathways, contribute to the constructive process of brain predictions. Both treatments target the assimilation of prediction errors, gradual reorganization of predictive models, and creation of a life with step-by-step constructive rewards. By elucidating the possible neurobiological mechanisms of these psychotherapeutic techniques, this article is expected to serve as the first step towards filling the cultural gap and creating more teaching methods based on these concepts.

Clinical characteristics of hospitalized children with community-acquired pneumonia and respiratory infections: Using machine learning approaches to support pathogen prediction at admission.

BACKGROUND: Acute respiratory infections (ARIs) are common in children. We developed machine learning models to predict pediatric ARI pathogens at admission. METHODS: We included hospitalized children with respiratory infections between 2010 and 2018. Clinical features were collected within 24 h of admission to construct models. The outcome of interest was the prediction of 6 common respiratory pathogens, including adenovirus, influenza virus types A and B, parainfluenza virus (PIV), respiratory syncytial virus (RSV), and Mycoplasma pneumoniae (MP). Model performance was estimated using area under the receiver operating characteristic curve (AUROC). Feature importance was measured using Shapley Additive exPlanation (SHAP) values. RESULTS: A total of 12,694 admissions were included. Models trained with 9 features (age, event pattern, fever, C-reactive protein, white blood cell count, platelet count, lymphocyte ratio, peak temperature, peak heart rate) achieved the best performance (AUROC: MP 0.87, 95% CI 0.83-0.90; RSV 0.84, 95% CI 0.82-0.86; adenovirus 0.81, 95% CI 0.77-0.84; influenza A 0.77, 95% CI 0.73-0.80; influenza B 0.70, 95% CI 0.65-0.75; PIV 0.73, 95% CI 0.69-0.77). Age was the most important feature to predict MP, RSV and PIV infections. Event patterns were useful for influenza virus prediction, and C-reactive protein had the highest SHAP value for adenovirus infections. CONCLUSION: We demonstrate how artificial intelligence can assist clinicians identify potential pathogens associated with pediatric ARIs upon admission. Our models provide explainable results that could help optimize the use of diagnostic testing. Integrating our models into clinical workflows may lead to improved patient outcomes and reduce unnecessary medical costs.

Automatic segmentation and measurement of pressure injuries using deep learning models and a LiDAR camera.

Pressure injuries are a common problem resulting in poor prognosis, long-term hospitalization, and increased medical costs in an aging society. This study developed a method to do automatic segmentation and area measurement of pressure injuries using deep learning models and a light detection and ranging (LiDAR) camera. We selected the finest photos of patients with pressure injuries, 528 in total, at National Taiwan University Hospital from 2016 to 2020. The margins of the pressure injuries were labeled by three board-certified plastic surgeons. The labeled photos were trained by Mask R-CNN and U-Net for segmentation. After the segmentation model was constructed, we made an automatic wound area measurement via a LiDAR camera. We conducted a prospective clinical study to test the accuracy of this system. For automatic wound segmentation, the performance of the U-Net (Dice coefficient (DC): 0.8448) was better than Mask R-CNN (DC: 0.5006) in the external validation. In the prospective clinical study, we incorporated the U-Net in our automatic wound area measurement system and got 26.2% mean relative error compared with the traditional manual method. Our segmentation model, U-Net, and area measurement system achieved acceptable accuracy, making them applicable in clinical circumstances.

Application of multiple deep learning models for automatic burn wound assessment.

PURPOSE: Accurate assessment of the percentage of total body surface area (%TBSA) burned is crucial in managing burn injuries. It is difficult to estimate the size of an irregular shape by inspection. Many articles reported the discrepancy of estimating %TBSA burned by different doctors. We set up a system with multiple deep learning (DL) models for %TBSA estimation, as well as the segmentation of possibly poor-perfused deep burn regions from the entire wound. METHODS: We proposed boundary-based labeling for datasets of total burn wound and palm, whereas region-based labeling for the dataset of deep burn wound. Several powerful DL models (U-Net, PSPNet, DeeplabV3+, Mask R-CNN) with encoders ResNet101 had been trained and tested from the above datasets. With the subject distances, the %TBSA burned could be calculated by the segmentation of total burn wound area with respect to the palm size. The percentage of deep burn area could be obtained from the segmentation of deep burn area from the entire wound. RESULTS: A total of 4991 images of early burn wounds and 1050 images of palms were boundary-based labeled. 1565 out of 4994 images with deep burn were preprocessed with superpixel segmentation into small regions before labeling. DeeplabV3+ had slightly better performance in three tasks with precision: 0.90767, recall: 0.90065 for total burn wound segmentation; precision: 0.98987, recall: 0.99036 for palm segmentation; and precision: 0.90152, recall: 0.90219 for deep burn segmentation. CONCLUSION: Combining the segmentation results and clinical data, %TBSA burned, the volume of fluid for resuscitation, and the percentage of deep burn area can be automatically diagnosed by DL models with a pixel-to-pixel method. Artificial intelligence provides consistent, accurate and rapid assessments of burn wounds.

Successful real-world application of an osteoarthritis classification deep-learning model using 9210 knees-An orthopedic surgeon's view.

This study aimed to evaluate the performance of a deep-learning model to evaluate knee osteoarthritis using Kellgren-Lawrence grading in real-life knee radiographs. A deep convolutional neural network model was trained using 8964 knee radiographs from the osteoarthritis initiative (OAI), including 962 testing set images. Another 246 knee radiographs from the Far Eastern Memorial Hospital were used for external validation. The OAI testing set and external validation images were evaluated by experienced specialists, two orthopedic surgeons, and a musculoskeletal radiologist. The accuracy, interobserver agreement, F1 score, precision, recall, specificity, and ability to identify surgical candidates were used to compare the performances of the model and specialists. Attention maps illustrated the interpretability of the model classification. The model had a 78% accuracy and consistent interobserver agreement for the OAI (model-surgeon 1 К = 0.80, model-surgeon 2 К = 0.84, model-radiologist К = 0.86) and external validation (model-surgeon 1 К = 0.81, model-surgeon 2 К = 0.82, model-radiologist К = 0.83) images. A lower interobserver agreement was found in the images misclassified by the model (model-surgeon 1 К = 0.57, model-surgeon 2 К = 0.47, model-radiologist К = 0.65). The model performed better than specialists in identifying surgical candidates (Kellgren-Lawrence Stages 3 and 4) with an F1 score of 0.923. Our model not only had comparable results with specialists with respect to the ability to identify surgical candidates but also performed consistently with open database and real-life radiographs. We believe the controversy of the misclassified knee osteoarthritis images was based on a significantly lower interobserver agreement.

Use of multimodal dataset in AI for detecting glaucoma based on fundus photographs assessed with OCT: focus group study on high prevalence of myopia.

BACKGROUND: Glaucoma is one of the major causes of blindness; it is estimated that over 110 million people will be affected by glaucoma worldwide by 2040. Research on glaucoma detection using deep learning technology has been increasing, but the diagnosis of glaucoma in a large population with high incidence of myopia remains a challenge. This study aimed to provide a decision support system for the automatic detection of glaucoma using fundus images, which can be applied for general screening, especially in areas of high incidence of myopia. METHODS: A total of 1,155 fundus images were acquired from 667 individuals with a mean axial length of 25.60 ± 2.0 mm at the National Taiwan University Hospital, Hsinchu Br. These images were graded based on the findings of complete ophthalmology examinations, visual field test, and optical coherence tomography into three groups: normal (N, n = 596), pre-perimetric glaucoma (PPG, n = 66), and glaucoma (G, n = 493), and divided into a training-validation (N: 476, PPG: 55, G: 373) and test (N: 120, PPG: 11, G: 120) sets. A multimodal model with the Xception model as image feature extraction and machine learning algorithms [random forest (RF), support vector machine (SVM), dense neural network (DNN), and others] was applied. RESULTS: The Xception model classified the N, PPG, and G groups with 93.9% of the micro-average area under the receiver operating characteristic curve (AUROC) with tenfold cross-validation. Although normal and glaucoma sensitivity can reach 93.51% and 86.13% respectively, the PPG sensitivity was only 30.27%. The AUROC increased to 96.4% in the N + PPG and G groups. The multimodal model with the N + PPG and G groups showed that the AUROCs of RF, SVM, and DNN were 99.56%, 99.59%, and 99.10%, respectively; The N and PPG + G groups had less than 1% difference. The test set showed an overall 3%-5% less AUROC than the validation results. CONCLUSION: The multimodal model had good AUROC while detecting glaucoma in a population with high incidence of myopia. The model shows the potential for general automatic screening and telemedicine, especially in Asia. TRIAL REGISTRATION: The study was approved by the Institutional Review Board of the National Taiwan University Hospital, Hsinchu Branch (no. NTUHHCB 108-025-E).

Training a Deep Contextualized Language Model for International Classification of Diseases, 10th Revision Classification via Federated Learning: Model Development and Validation Study.

BACKGROUND: The automatic coding of clinical text documents by using the International Classification of Diseases, 10th Revision (ICD-10) can be performed for statistical analyses and reimbursements. With the development of natural language processing models, new transformer architectures with attention mechanisms have outperformed previous models. Although multicenter training may increase a model's performance and external validity, the privacy of clinical documents should be protected. We used federated learning to train a model with multicenter data, without sharing data per se. OBJECTIVE: This study aims to train a classification model via federated learning for ICD-10 multilabel classification. METHODS: Text data from discharge notes in electronic medical records were collected from the following three medical centers: Far Eastern Memorial Hospital, National Taiwan University Hospital, and Taipei Veterans General Hospital. After comparing the performance of different variants of bidirectional encoder representations from transformers (BERT), PubMedBERT was chosen for the word embeddings. With regard to preprocessing, the nonalphanumeric characters were retained because the model's performance decreased after the removal of these characters. To explain the outputs of our model, we added a label attention mechanism to the model architecture. The model was trained with data from each of the three hospitals separately and via federated learning. The models trained via federated learning and the models trained with local data were compared on a testing set that was composed of data from the three hospitals. The micro F1 score was used to evaluate model performance across all 3 centers. RESULTS: The F1 scores of PubMedBERT, RoBERTa (Robustly Optimized BERT Pretraining Approach), ClinicalBERT, and BioBERT (BERT for Biomedical Text Mining) were 0.735, 0.692, 0.711, and 0.721, respectively. The F1 score of the model that retained nonalphanumeric characters was 0.8120, whereas the F1 score after removing these characters was 0.7875-a decrease of 0.0245 (3.11%). The F1 scores on the testing set were 0.6142, 0.4472, 0.5353, and 0.2522 for the federated learning, Far Eastern Memorial Hospital, National Taiwan University Hospital, and Taipei Veterans General Hospital models, respectively. The explainable predictions were displayed with highlighted input words via the label attention architecture. CONCLUSIONS: Federated learning was used to train the ICD-10 classification model on multicenter clinical text while protecting data privacy. The model's performance was better than that of models that were trained locally.

Fast Healthcare Interoperability Resources for Inpatient Deterioration Detection With Time-Series Vital Signs: Design and Implementation Study.

BACKGROUND: Vital signs have been widely adopted in in-hospital cardiac arrest (IHCA) assessment, which plays an important role in inpatient deterioration detection. As the number of early warning systems and artificial intelligence applications increases, health care information exchange and interoperability are becoming more complex and difficult. Although Health Level 7 Fast Healthcare Interoperability Resources (FHIR) have already developed a vital signs profile, it is not sufficient to support IHCA applications or machine learning-based models. OBJECTIVE: In this paper, for IHCA instances with vital signs, we define a new implementation guide that includes data mapping, a system architecture, a workflow, and FHIR applications. METHODS: We interviewed 10 experts regarding health care system integration and defined an implementation guide. We then developed the FHIR Extract Transform Load to map data to FHIR resources. We also integrated an early warning system and machine learning pipeline. RESULTS: The study data set includes electronic health records of adult inpatients who visited the En-Chu-Kong hospital. Medical staff regularly measured these vital signs at least 2 to 3 times per day during the day, night, and early morning. We used pseudonymization to protect patient privacy. Then, we converted the vital signs to FHIR observations in the JSON format using the FHIR Extract Transform Load application. The measured vital signs include systolic blood pressure, diastolic blood pressure, heart rate, respiratory rate, and body temperature. According to clinical requirements, we also extracted the electronic health record information to the FHIR server. Finally, we integrated an early warning system and machine learning pipeline using the FHIR RESTful application programming interface. CONCLUSIONS: We successfully demonstrated a process that standardizes health care information for inpatient deterioration detection using vital signs. Based on the FHIR definition, we also provided an implementation guide that includes data mapping, an integration process, and IHCA assessment using vital signs. We also proposed a clarifying system architecture and possible workflows. Based on FHIR, we integrated the 3 different systems in 1 dashboard system, which can effectively solve the complexity of the system in the medical staff workflow.

A Precision Health Service for Chronic Diseases: Development and Cohort Study Using Wearable Device, Machine Learning, and Deep Learning.

This paper presents an integrated and scalable precision health service for health promotion and chronic disease prevention. Continuous real-time monitoring of lifestyle and environmental factors is implemented by integrating wearable devices, open environmental data, indoor air quality sensing devices, a location-based smartphone app, and an AI-assisted telecare platform. The AI-assisted telecare platform provided comprehensive insight into patients' clinical, lifestyle, and environmental data, and generated reliable predictions of future acute exacerbation events. All data from 1,667 patients were collected prospectively during a 24-month follow-up period, resulting in the detection of 386 abnormal episodes. Machine learning algorithms and deep learning algorithms were used to train modular chronic disease models. The modular chronic disease prediction models that have passed external validation include obesity, panic disorder, and chronic obstructive pulmonary disease, with an average accuracy of 88.46%, a sensitivity of 75.6%, a specificity of 93.0%, and an F1 score of 79.8%. Compared with previous studies, we establish an effective way to collect lifestyle, life trajectory, and symptom records, as well as environmental factors, and improve the performance of the prediction model by adding objective comprehensive data and feature selection. Our results also demonstrate that lifestyle and environmental factors are highly correlated with patient health and have the potential to predict future abnormal events better than using only questionnaire data. Furthermore, we have constructed a cost-effective model that needs only a few features to support the prediction task, which is helpful for deploying real-world modular prediction models.

Improved inpatient deterioration detection in general wards by using time-series vital signs.

Although in-hospital cardiac arrest is uncommon, it has a high mortality rate. Risk identification of at-risk patients is critical for post-cardiac arrest survival rates. Early warning scoring systems are generally used to identify hospitalized patients at risk of deterioration. However, these systems often require clinical data that are not always regularly measured. We developed a more accurate, machine learning-based model to predict clinical deterioration. The time series early warning score (TEWS) used only heart rate, systolic blood pressure, and respiratory data, which are regularly measured in general wards. We tested the performance of the TEWS in two tasks performed with data from the electronic medical records of 16,865 adult admissions and compared the results with those of other classifications. The TEWS detected more deteriorations with the same level of specificity as the different algorithms did when inputting vital signs data from 48 h before an event. Our framework improved in-hospital cardiac arrest prediction and demonstrated that previously obtained vital signs data can be used to identify at-risk patients in real-time. This model may be an alternative method for detecting patient deterioration.

Automatic International Classification of Diseases Coding System: Deep Contextualized Language Model With Rule-Based Approaches.

BACKGROUND: The tenth revision of the International Classification of Diseases (ICD-10) is widely used for epidemiological research and health management. The clinical modification (CM) and procedure coding system (PCS) of ICD-10 were developed to describe more clinical details with increasing diagnosis and procedure codes and applied in disease-related groups for reimbursement. The expansion of codes made the coding time-consuming and less accurate. The state-of-the-art model using deep contextual word embeddings was used for automatic multilabel text classification of ICD-10. In addition to input discharge diagnoses (DD), the performance can be improved by appropriate preprocessing methods for the text from other document types, such as medical history, comorbidity and complication, surgical method, and special examination. OBJECTIVE: This study aims to establish a contextual language model with rule-based preprocessing methods to develop the model for ICD-10 multilabel classification. METHODS: We retrieved electronic health records from a medical center. We first compared different word embedding methods. Second, we compared the preprocessing methods using the best-performing embeddings. We compared biomedical bidirectional encoder representations from transformers (BioBERT), clinical generalized autoregressive pretraining for language understanding (Clinical XLNet), label tree-based attention-aware deep model for high-performance extreme multilabel text classification (AttentionXLM), and word-to-vector (Word2Vec) to predict ICD-10-CM. To compare different preprocessing methods for ICD-10-CM, we included DD, medical history, and comorbidity and complication as inputs. We compared the performance of ICD-10-CM prediction using different preprocesses, including definition training, external cause code removal, number conversion, and combination code filtering. For the ICD-10 PCS, the model was trained using different combinations of DD, surgical method, and key words of special examination. The micro F1 score and the micro area under the receiver operating characteristic curve were used to compare the model's performance with that of different preprocessing methods. RESULTS: BioBERT had an F1 score of 0.701 and outperformed other models such as Clinical XLNet, AttentionXLM, and Word2Vec. For the ICD-10-CM, the model had an F1 score that significantly increased from 0.749 (95% CI 0.744-0.753) to 0.769 (95% CI 0.764-0.773) with the ICD-10 definition training, external cause code removal, number conversion, and combination code filter. For the ICD-10-PCS, the model had an F1 score that significantly increased from 0.670 (95% CI 0.663-0.678) to 0.726 (95% CI 0.719-0.732) with a combination of discharge diagnoses, surgical methods, and key words of special examination. With our preprocessing methods, the model had the highest area under the receiver operating characteristic curve of 0.853 (95% CI 0.849-0.855) and 0.831 (95% CI 0.827-0.834) for ICD-10-CM and ICD-10-PCS, respectively. CONCLUSIONS: The performance of our model with the pretrained contextualized language model and rule-based preprocessing method is better than that of the state-of-the-art model for ICD-10-CM or ICD-10-PCS. This study highlights the importance of rule-based preprocessing methods based on coder coding rules.