Event Prediction Model Research Articles

Study on predicting new onset heart failure events in patients with hypertrophic cardiomyopathy using machine learning algorithms based on clinical and magnetic resonance features

Objective: To explore the value of predicting new-onset heart failure events in patients with hypertrophic cardiomyopathy (HCM) using clinical and cardiac magnetic resonance (CMR) features based on machine learning algorithms. Methods: The study was a retrospective cohort study. Patients with a confirmed diagnosis of HCM who underwent CMR examinations at Beijing Anzhen Hospital from May 2017 to March 2021 were selected and randomly divided into the training set and the validation set in a ratio of 7∶3. Clinical data and CMR parameters (including conventional parameters and radiomics features) were collected. The endpoint events were heart failure hospitalization and heart failure death, with follow-up ending in January 2023. Features with high stability and P value<0.05 in univariate Cox regression analysis were selected. Subsequently, three machine learning algorithms-random forest, decision tree, and XGBoost-were used to build heart failure event prediction models in the training set. The model performance was then evaluated using the independent validation set, with the performance assessed based on the concordance index. Results: A total of 462 patients were included, with a median age of 51 (39, 62) years, of whom 332 (71.9%) were male. There were 323 patients in the training set and 139 in the validation set. The median follow-up time was 42 (28, 52) months. A total of 44 patients (9.5% (44/462)) experienced endpoint events (8 cases of heart failure death and 36 cases of heart failure hospitalization), with 31 events in the training set and 13 in the validation set. Univariate Cox regression analysis identified 39 radiomic features, 4 conventional CMR parameters (left ventricular end-diastolic volume index, left ventricular end-systolic volume index, left ventricular ejection fraction, and late gadolinium enhancement ratio), and 1 clinical feature (history of non-sustained ventricular tachycardia) that could be included in the machine learning model. In the prediction models built with the training set, the concordance indices for the random forest, decision tree, and XGBoost models were 0.966 (95%CI 0.813-0.995), 0.956 (95%CI 0.796-0.992), and 0.973 (95%CI 0.823-0.996), respectively. In the validation set, the concordance indices for the random forest, decision tree, and XGBoost models were 0.854 (95%CI 0.557-0.964), 0.706 (95%CI 0.399-0.896), and 0.703 (95%CI 0.408-0.890), respectively. Conclusion: Integrating clinical and CMR features of HCM patients through machine learning aids in predicting heart failure events, with the random forest model showing superior performance.

Time-varying cox model for heart failure prediction in phospholamban p.(Arg14del)-positive individuals

Abstract Background Phospholamban (PLN) p.(Arg14del)-positive individuals are at risk for developing severe heart failure (HF), yet a comprehensive risk model for this outcome is missing. Our PLN registry contains longitudinal data which enables us to capture dynamic changes in covariables over time, while prior prediction models rely on baseline clinical information which may change over time. Using this longitudinal clinical information will enhance the accuracy of HF prediction and may aid patients selection for future (gene)therapy. Purpose The aim of this study is to develop a dynamic HF risk prediction model for PLN p.(Arg14del)-positive individuals. Method Data were collected of 594 PLN p.(Arg14del)-positive individuals who had no documented history of HF or myocardial infarction at time of the first echocardiography. We incorporated two time-dependent covariates in our time-varying Cox regression model: longitudinal individual predicted left ventricular ejection fraction (LVEF) and longitudinal individual predicted relative risk for major ventricular arrhythmia (VA). Predicted LVEF values were derived using observed LVEF values in a linear mixed-effect model, while the relative risk for major VA was predicted using a Cox regression model with major VA as the outcome. These time-dependent covariates were included in the time-varying Cox regression along with baseline covariables age at inclusion and previous major VA. The results were validated using 5-fold cross-validation. Results Over a median follow-up period of 3.6 years (interquartile range 1.4-6.8), 77 (13%) individuals developed HF, defined as a composite endpoint of HF hospitalisation, implantation of left ventricular assist device, heart transplantation and HF-related mortality. Our time-varying Cox regression model demonstrated a robust 5-year mean C-statistic of 0.904 (95% CI 0.901-0.908). The hazard ratio for time-dependent LVEF predictions was 0.89 per % LVEF increase (95% CI 0.87-0.91, p &amp;lt;0.001) and for time-dependent relative major VA risk predictions 3.59 (95% CI 1.49-8.69, p = 0.005). Regarding baseline covariates, age at inclusion had a hazard ratio of 1.01 (95% CI 0.99-1.03, p = 0.47) and previous major VA 0.48 (95% CI 0.21-1.1, p = 0.08). Combining the hazard ratios of time-dependent predicted relative risk for major VA and major VA in history results in a hazard ratio of 1.73. Conclusion Our study introduces a novel time-varying Cox regression model for the individual prediction of heart failure events in PLN p.(Arg14del)-positive individuals which may aid patient selection for future (gene)therapy. Results demonstrate robust predictive performance (C-statistic of 0.904, 95% CI 0.901-0.908), highlighting the significance of incorporating longitudinal data for enhanced prediction accuracy.

Machine-Learning for Phenotyping and Prognostication of Myocardial Infarction and Injury in Suspected Acute Coronary Syndrome

BackgroundClinical work-up for suspected acute coronary syndrome (ACS) is resource intensive. ObjectivesThis study aimed to develop a machine learning model for digitally phenotyping myocardial injury and infarction and predict 30-day events in suspected ACS patients. MethodsTraining and testing data sets, predominantly derived from electronic health records, included suspected ACS patients presenting to 6 and 26 South Australian hospitals, respectively. All index presentations and 30-day death and myocardial infarction (MI) were adjudicated using the Fourth Universal Definition of MI. We developed 2 diagnostic prediction models which phenotype myocardial injury and infarction according to the Fourth UDMI (chronic myocardial injury vs acute myocardial injury patterns, the latter further differentiated into acute non-ischaemic myocardial injury, Types 1 and 2 MI) using eXtreme Gradient Boosting (XGB) and deep-learning (DL). We also developed an event prediction model for risk prediction of 30-day death or MI using XGB. Analyses were performed in Python 3.6. ResultsThe training and testing data sets had 6,722 and 8,869 participants, respectively. The diagnostic prediction XGB and deep learning models achieved an area under the curve of 99.2% ± 0.1% and 98.8% ± 0.2%, respectively, for differentiating an acute myocardial injury pattern from no injury or chronic myocardial injury pattern and achieved 95.5% ± 0.2% and 94.6% ± 0.9%, respectively, for differentiating type 1 MI from type 2 MI or acute nonischemic myocardial injury. The 30-day death/MI event prediction model achieved an area under the curve of 88.5% ± 0.5%. ConclusionsMachine learning models can digitally phenotype suspected ACS patients at index presentation and predict subsequent events within 30 days. These models require external validation in a randomized clinical trial to evaluate their impact in clinical practice.

The adverse inpatient medication event and frailty (AIME-frail) risk prediction model

Medication harm affects between 5 and 15% of hospitalised patients, with approximately half of the harm events considered preventable through timely intervention. The Adverse Inpatient Medication Event (AIME) risk prediction model was previously developed to guide a systematic approach to patient prioritisation for targeted clinician review, but frailty was not tested as a candidate predictor variable. To evaluate the predictive performance of an updated AIME model, incorporating a measure of frailty, when applied to a new multisite cohort of hospitalised adult inpatients. A retrospective cohort study was conducted at two tertiary Australian hospitals on patients discharged between 1st January and April 31, 2020. Data were extracted from electronic medical records (EMRs) and clinical coding databases. Medication harm was identified using ICD-10 Y-codes and confirmed by senior pharmacist review of medical records. The Hospital Frailty Risk Score (HFRS) was calculated for each patient. Logistic regression analysis was used to construct a modified AIME model. Candidate variables of the original AIME model, together with new variables including HFRS were tested. Performance of the final model was reported using area under the curve (AUC) and decision curve analysis (DCA). A total of 4089 patient admissions were included, with a mean age±standard deviation (SD) of 64 years (±19 years), 2050 patients (50%) were males, and mean HFRS was 6.2 (±5.9). 184 patients (4.5%) experienced one or more medication harm events during hospitalisation. The new AIME-Frail risk model incorporated 5 of the original variables: length of stay (LOS), anti-psychotics, antiarrhythmics, immunosuppressants, and INR greater than 3, as well as 5 new variables: HFRS, anticoagulants, antibiotics, insulin, and opioid use. The AUC was 0.79 (95% CI: 0.76-0.83) which was superior to the original model (AUC=0.70, 95% CI: 0.65-0.74) with a sensitivity of 69%, specificity of 81%, positive predictive value of 0.14 (95% CI: 0.10-0.17) and negative predictive value of 0.98 (95% CI: 0.97-0.99). The DCA identified the model as having potential clinical utility between the probability thresholds of 0.05-0.4. The inclusion of a frailty measure improved the predictive performance of the AIME model. Screening inpatients using the AIME-Frail tool could identify more patients at high-risk of medication harm who warrant timely clinician review.

Development of a deep learning model that predicts critical events of pediatric patients admitted to general wards

Early detection of deteriorating patients is important to prevent life-threatening events and improve clinical outcomes. Efforts have been made to detect or prevent major events such as cardiopulmonary resuscitation, but previously developed tools are often complicated and time-consuming, rendering them impractical. To overcome this problem, we designed this study to create a deep learning prediction model that predicts critical events with simplified variables. This retrospective observational study included patients under the age of 18 who were admitted to the general ward of a tertiary children’s hospital between 2020 and 2022. A critical event was defined as cardiopulmonary resuscitation, unplanned transfer to the intensive care unit, or mortality. The vital signs measured during hospitalization, their measurement intervals, sex, and age were used to train a critical event prediction model. Age-specific z-scores were used to normalize the variability of the normal range by age. The entire dataset was classified into a training dataset and a test dataset at an 8:2 ratio, and model learning and testing were performed on each dataset. The predictive performance of the developed model showed excellent results, with an area under the receiver operating characteristics curve of 0.986 and an area under the precision-recall curve of 0.896. We developed a deep learning model with outstanding predictive power using simplified variables to effectively predict critical events while reducing the workload of medical staff. Nevertheless, because this was a single-center trial, no external validation was carried out, prompting further investigation.

Attention-Based Deep Learning Model for Prediction of Major Adverse Cardiovascular Events in Peritoneal Dialysis Patients.

Major adverse cardiovascular events (MACE) encompass pivotal cardiovascular outcomes such as myocardial infarction, unstable angina, and cardiovascular-related mortality. Patients undergoing peritoneal dialysis (PD) exhibit specific cardiovascular risk factors during the treatment, which can escalate the likelihood of cardiovascular events. Hence, the prediction and key factor analysis of MACE have assumed paramount significance for peritoneal dialysis patients. Current pathological methodologies for prognosis prediction are not only costly but also cumbersome in effectively processing electronic health records (EHRs) data with high dimensionality, heterogeneity, and time series. Therefore in this study, we propose the CVEformer, an attention-based neural network designed to predict MACE and analyze risk factors. CVEformer leverages the self-attention mechanism to capture temporal correlations among time series variables, allowing for weighted integration of variables and estimation of the probability of MACE. CVEformer first captures the correlations among heterogeneous variables through attention scores. Then, it analyzes the correlations within the time series data to identify key risk variables and predict the probability of MACE. When trained and evaluated on data from a large cohort of peritoneal dialysis patients across multiple centers, CVEformer outperforms existing models in terms of predictive performance.

Machine learning models for early prediction of potassium lowering effectiveness and adverse events in patients with hyperkalemia

The aim of this study was to develop a model for early prediction of adverse events and treatment effectiveness in patients with hyperkalemia. We collected clinical data from patients with hyperkalemia in the First Hospital of Zhejiang University School of Medicine between 2015 and 2021. The least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression were used to analyze the predictors on the full dataset. We randomly divided the data into a training group and a validation group, and used LASSO to filter variables in the training set. Six machine learning methods were used to develop the models. The best model was selected based on the area under the curve (AUC). Shapley additive exPlanations (SHAP) values were used to explain the best model. A total of 1074 patients with hyperkalemia were finally enrolled. Diastolic blood pressure (DBP), breathing, oxygen saturation (SPO2), Glasgow coma score (GCS), liver disease, oliguria, blood sodium, international standardized ratio (ISR), and initial blood potassium were the predictors of the occurrence of adverse events; peripheral edema, estimated glomerular filtration rate (eGFR), blood sodium, actual base residual, and initial blood potassium were the predictors of therapeutic effect. Extreme gradient boosting (XGBoost) model achieved the best performance (adverse events: AUC = 0.87; therapeutic effect: AUC = 0.75). A model based on clinical characteristics was developed and validated with good performance.

Building and Validation of an Acute Event Prediction Model for Severe Mental Disorders.

The global incidence of acute events in psychiatric patients is intensifying, and models to successfully predict acute events have attracted much attention. To explore the influence factors of acute incident severe mental disorders (SMDs) and the application of Rstudio statistical software, and build and verify a nomogram prediction model. SMDs were taken as research objects. The questionnaire survey method was adopted to collect data. Patients with acute event independent factors were screened. R software multivariable Logistic regression model was constructed and a nomogram was drawn. A total of 342 patients with SMDs were hospitalized, and the number of patients who encountered acute events was 64, which accounted for 18.70% of all patients. Statistical significances were found in many aspects (all P ˂ 0.05). Such aspects included Medication adherence, disease diagnosis, marital status, caregivers, social support and the hospitalization environment (odds ratio (OR) = 4.08, 11.62, 12.06, 10.52, 0.04 and 0.61, respectively) were independent risk factors for the acute events of patients with SMDs. The prediction model was modeled, and the AUC was 0.77 and 0.80. The calibration curve shows that the model has good calibration. The clinical decision curve shows that the model has a good clinical effect. The constructed risk prediction model shows good prediction effectiveness in the acute events of patients with SMDs, which is helpful for the early detection of clinical mental health staff at high risk of acute events.

Aggregate Clinical and Biomarker-Based Model Predicts Adverse Outcomes in Patients With Coronary Artery Disease

Despite guideline-based therapy, patients with coronary artery disease (CAD) are at widely variable risk for cardiovascular events. This variability demands a more individualized risk assessment. Herein, we evaluate the prognostic value of 6 biomarkers: high-sensitivity C-reactive protein, heat shock protein-70, fibrin degradation products, soluble urokinase plasminogen activator receptor, high-sensitivity troponin I, and B-type natriuretic peptide. We then develop a multi-biomarker-based cardiovascular event prediction model for patients with stable CAD. In total, 3,115 subjects with stable CAD who underwent cardiac catheterization at Emory (mean age 62.8years, 17% Black, 35% female, 57% obstructive CAD, 31% diabetes mellitus) were randomized into a training cohort to identify biomarker cutoff values and a validation cohort for prediction assessment. Main outcomes included (1) all-cause death and (2) a composite of cardiovascular death and nonfatal myocardial infarction (MI) within 5years. Elevation of each biomarker level was associated with higher event rates in the training cohort. A biomarker risk score was created using optimal cutoffs, ranging from 0 to 6 for each biomarker exceeding its cutoff. In the validation cohort, each unit increase in the biomarker risk score was independently associated with all-cause death (hazard ratio 1.62, 95% confidence interval [CI] 1.45 to 1.80) and cardiovascular death/MI (hazard ratio 1.52, 95% CI 1.35 to 1.71). A biomarker risk prediction model for cardiovascular death/MI improved the c-statistic (∆ 6.4%, 95% CI 3.9 to 8.8) and net reclassification index by 31.1% (95% CI 24 to 37), compared with clinical risk factors alone. Integrating multiple biomarkers with clinical variables refines cardiovascular risk assessment in patients with CAD.

Machine Learning for Prediction of Adverse Cardiovascular Events in Adults With Repaired Tetralogy of Fallot Using Clinical and Cardiovascular Magnetic Resonance Imaging Variables

Background: Existing models for prediction of major adverse cardiovascular events (MACE) after repair of tetralogy of Fallot have been limited by modest predictive capacity and limited applicability to routine clinical practice. We hypothesized that an artificial intelligence model using an array of parameters would enhance 5-year MACE prediction in adults with repaired tetralogy of Fallot. Methods: A machine learning algorithm was applied to 2 nonoverlapping, institutional databases of adults with repaired tetralogy of Fallot: (1) for model development, a prospectively constructed clinical and cardiovascular magnetic resonance registry; (2) for model validation, a retrospective database comprised of variables extracted from the electronic health record. The MACE composite outcome included mortality, resuscitated sudden death, sustained ventricular tachycardia and heart failure. Analysis was restricted to individuals with MACE or followed ≥5 years. A random forest model was trained using machine learning (n=57 variables). Repeated random sub-sampling validation was sequentially applied to the development dataset followed by application to the validation dataset. Results: We identified 804 individuals (n=312 for development and n=492 for validation). Model prediction (area under the curve [95% CI]) for MACE in the validation dataset was strong (0.82 [0.74–0.89]) with superior performance to a conventional Cox multivariable model (0.63 [0.51–0.75]; P =0.003). Model performance did not change significantly with input restricted to the 10 strongest features (decreasing order of strength: right ventricular end-systolic volume indexed, right ventricular ejection fraction, age at cardiovascular magnetic resonance imaging, age at repair, absolute ventilatory anaerobic threshold, right ventricular end-diastolic volume indexed, ventilatory anaerobic threshold % predicted, peak aerobic capacity, left ventricular ejection fraction, and pulmonary regurgitation fraction; 0.81 [0.72–0.89]; P =0.232). Removing exercise parameters resulted in inferior model performance (0.75 [0.65–0.84]; P =0.002). Conclusions: In this single-center study, a machine learning-based prediction model comprised of readily available clinical and cardiovascular magnetic resonance imaging variables performed well in an independent validation cohort. Further study will determine the value of this model for risk stratification in adults with repared tetralogy of Fallot.

Rainfall-driven machine learning models for accurate flood inundation mapping in Karachi, Pakistan

Urban pluvial flooding (UPF) has emerged as a serious natural hazard, especially in recent years. Previous research on UPF prediction has mainly focused on hydrological models, which required a large amount of data. However, a data-driven method can significantly reduce the computational cost by using rainfall amounts to predict pluvial flooding. Intensity-duration-frequency (IDF) curves using the Gumbel method can provide a better interpretation of the correlation between rainfall intensity, duration, and probability of occurrence of a given rainfall amount. In this study, machine learning models (ML) for rainfall amounts were used to identify flood points in a case study conducted in Karachi, Pakistan. Thirteen inundation factors were used for the ML models, including a new factor, curve number. Ten ML models were applied first on training and then on validation data, yielding the inundation points. The training and validation process of the model included 384 flood points. Several statistics were used to verify the performance and accuracy of the model. We found that the Light Gradient Boost Machine and Random Forest Classifier models were the most accurate in training and validating the model, while the Decision Tree and K-Nearest Neighbor models were the least accurate in training and validating the model. The study provides valuable information for decision makers to protect communities from flood hazards by incorporating the likely intensity and duration of rainfall events and carefully selecting influencing factors into flood event prediction models.

Machine learning and artificial intelligence models development in rainfall-induced landslide prediction

&lt;span lang="EN-US"&gt;In Indonesia, rainfall is one crucial triggering factor for landslides. This paper aims to build landslide event prediction models using several machine learning and artificial intelligence algorithms. The algorithms were trained with two different methods. The input of the algorithms was precipitation data obtained from the global satellite mapping of precipitation satellite observation, and the target was landslide event occurrence data obtained from the Indonesian National Board for Disaster Management. Each algorithm provided some model candidates with different parameter settings for each method. As a result, there were 52 and 72 model candidates for both methods. The best model was then chosen from each method. The result shows that the model generated by generalized linear model was the best model for the first method and deep learning for the second one. Furthermore, the best models at each method gained 0.828 and 0.836 for the area under receiver operating characteristics curve, and their log-loss were 0.156 and 0.154. The second method, which used input data transformation, provided better performance.&lt;/span&gt;

Hi-BEHRT: Hierarchical Transformer-based model for accurate prediction of clinical events using multimodal longitudinal electronic health records.

Electronic health records (EHR) represent a holistic overview of patients' trajectories. Their increasing availability has fueled new hopes to leverage them and develop accurate risk prediction models for a wide range of diseases. Given the complex interrelationships of medical records and patient outcomes, deep learning models have shown clear merits in achieving this goal. However, a key limitation of current study remains their capacity in processing long sequences, and long sequence modelling and its application in the context of healthcare and EHR remains unexplored. Capturing the whole history of medical encounters is expected to lead to more accurate predictions, but the inclusion of records collected for decades and from multiple resources can inevitably exceed the receptive field of the most existing deep learning architectures. This can result in missing crucial, long-term dependencies. To address this gap, we present Hi-BEHRT, a hierarchical Transformer-based model that can significantly expand the receptive field of Transformers and extract associations from much longer sequences. Using a multimodal large-scale linked longitudinal EHR, the Hi-BEHRT exceeds the state-of-the-art deep learning models 1% to 5% for area under the receiver operating characteristic (AUROC) curve and 1% to 8% for area under the precision recall (AUPRC) curve on average, and 2% to 8% (AUROC) and 2% to 11% (AUPRC) for patients with long medical history for 5-year heart failure, diabetes, chronic kidney disease, and stroke risk prediction. Additionally, because pretraining for hierarchical Transformer is not well-established, we provide an effective end-to-end contrastive pre-training strategy for Hi-BEHRT using EHR, improving its transferability on predicting clinical events with relatively small training dataset.

Finding low CHA2DS2-VASc scores unreliable? Why not give morphological and hemodynamic methods a try?

Patients with atrial fibrillation (AF) suffer from a high risk of thrombosis. Currently, the CHA2DS2-VASc score is the most widely used tool for risk stratification in patients with AF, but it has disappointing accuracy and limited predictive value, especially in those with low scores. Thrombi in patients with AF mostly grow in their left atrial appendages (LAA), which is directly related to the abnormal morphology of the LAA or the left atrium and the unusual hemodynamic state around LAA, which may sensitively evaluate the risk of thrombosis complications in patients with AF and bring bases to clinical plans of medication and operation. Therefore, we investigated the research progress of hemodynamic and morphological studies about the predictive value of thrombosis risk in patients with AF, intending to discuss the prediction potential of morphological and hemodynamic indexes when compared with the presently used CHA2DS2-VASc system and how to build a more precise thromboembolic event prediction model for patients with AF.

Artificial Intelligent Models for Detection and Prediction of Lost Circulation Events: A Review

Lost circulation or losses in drilling fluid is one of the most important problems in the oil and gas industry, and it appeared at the beginning of this industry, which caused many problems during the drilling process, which may lead to closing the well and stopping the drilling process. The drilling muds are relatively expensive, especially the muds that contain oil-based mud or that contain special additives, so it is not economically beneficial to waste and lose these muds. The treatment of drilling fluid losses is also somewhat expensive as a result of the wasted time that it caused, as well as the high cost of materials used in the treatment such as heavy materials, cement, and others. The best way to deal with drilling fluid losses is to prevent them. Drilling fluid loss is a complex problem that is difficult to predict using simple and traditional methods. Artificial intelligence represents a modern and accurate technology for solving complex problems such as drilling fluid loss. Artificial intelligence through supervised machine learning provides the possibility of predicting these losses before they occur based on field data such as drilling fluid properties, drilling parameters, rock properties, and geomechanical parameters that are related to the loss of circulation of the wells suffered from losses problem located in the same area. In this paper, several supervised machine learning models have been reviewed that were used for detecting and predicting of loss of drilling fluids during the drilling process. The paper provides an inclusive review of drilling fluid prediction and detection from simplest to more complected intelligent models.

Comparison metrics for multi-step prediction of rare events in vital sign signals

Prediction of changes in biomedical signals, such as vital signs, is useful for many clinical applications. Several signal prediction (forecasting) tools were developed, but their evaluation and applicability to a specific clinical use is context dependent. In this work, we propose a novel method to tackle the problem of evaluation and comparison of vital sign predictors for intervention based clinical studies. The proposed prediction quality measures are particularly well-suited for forecasting rare events scenarios. Specifically, using the novel metrics, we measure the prediction statistics and compare nine deep learning and autoregressive forecasting models for multi-step prediction of rare bradycardia events in preterm infants, however the new concepts allow applications to other biomedical signals. We validated the novel metrics with experimental results on testing sets with several days of vital sign recordings. Our results show that simple statistical predictors could outperform state-of-the-art deep learning architectures for low-dimensional signals.

Sudden Event Prediction Based on Event Knowledge Graph

Event prediction is a knowledge inference problem that predicts the consequences or effects of an event based on existing information. Early work on event prediction typically modeled the event context to predict what would happen next. Moreover, the predicted outcome was often singular. These studies had difficulty coping with both the problems of predicting sudden events in unknown contexts and predicting outcomes consisting of multiple events. To address these two problems better, we present the heterogeneous graph event prediction model (HGEP), which is based on an event knowledge graph. To cope with the situation of missing contexts, we propose a representation learning method based on the heterogeneous graph transformer. We generate event scenario representations from arguments of the initial event and other related concurrent events for the event prediction. We improve the prediction ability of the HGEP model through prior knowledge provided by scenario models in the event knowledge graph. To obtain multiple prediction outcomes, we design a scoring function to calculate the score of the occurrence probability of each event class. The event classes with scores higher than a priori values are adopted as prediction outcomes. In this paper, we create an event knowledge graph in the domain of transportation for an event prediction experiment. The experimental results show that the HGEP model can effectively make predictions with event scenario representations and has a more accurate matching rate and higher precision than the baseline model.

Externally Validated Postoperative Long-Term Adverse Event Prediction Model for Non-Cardiac Surgery: An Extension of Simple Postoperative AKI Risk (SPARK) Model

Externally Validated Postoperative Long-Term Adverse Event Prediction Model for Non-Cardiac Surgery: An Extension of Simple Postoperative AKI Risk (SPARK) Model

Multicenter validation of a deep-learning-based pediatric early-warning system for prediction of deterioration events.

Early recognition of deterioration events is crucial to improve clinical outcomes. For this purpose, we developed a deep-learning-based pediatric early-warning system (pDEWS) and aimed to validate its clinical performance. This is a retrospective multicenter cohort study including five tertiary-care academic children's hospitals. All pediatric patients younger than 19 years admitted to the general ward from January 2019 to December 2019 were included. Using patient electronic medical records, we evaluated the clinical performance of the pDEWS for identifying deterioration events defined as in-hospital cardiac arrest (IHCA) and unexpected general ward-to-pediatric intensive care unit transfer (UIT) within 24 hours before event occurrence. We also compared pDEWS performance to those of the modified pediatric early-warning score (PEWS) and prediction models using logistic regression (LR) and random forest (RF). The study population consisted of 28,758 patients with 34 cases of IHCA and 291 cases of UIT. pDEWS showed better performance for predicting deterioration events with a larger area under the receiver operating characteristic curve, fewer false alarms, a lower mean alarm count per day, and a smaller number of cases needed to examine than the modified PEWS, LR, or RF models regardless of site, event occurrence time, age group, or sex. The pDEWS outperformed modified PEWS, LR, and RF models for early and accurate prediction of deterioration events regardless of clinical situation. This study demonstrated the potential of pDEWS as an efficient screening tool for efferent operation of rapid response teams.

Development and validation of interpretable machine learning models for inpatient fall events and electronic medical record integration.

Falls are one of the most frequently occurring adverse events among hospitalized patients. The Morse Fall Scale, which has been widely used for fall risk assessment, has the two limitations of low specificity and difficulty in practical implementation. The aim of this study was to develop and validate an interpretable machine learning model for prediction of falls to be integrated in an electronic medical record (EMR) system. This was a retrospective study involving a tertiary teaching hospital in Seoul, Korea. Based on the literature, 83 known predictors were grouped into seven categories. Interpretable fall event prediction models were developed using multiple machine learning models including gradient boosting and Shapley values. Overall, 191,778 cases with 272 fall events (0.1%) were included in the analysis. With the validation cohort of 2020, the area under the receiver operating curve (AUROC) of the gradient boosting model was 0.817 (95% confidence interval [CI], 0.720-0.904), better performance than random forest (AUROC, 0.801; 95% CI, 0.708-0.890), logistic regression (AUROC, 0.802; 95% CI, 0.721-0.878), artificial neural net (AUROC, 0.736; 95% CI, 0.650-0.821), and conventional Morse fall score (AUROC, 0.652; 95% CI, 0.570-0.715). The model's interpretability was enhanced at both the population and patient levels. The algorithm was later integrated into the current EMR system. We developed an interpretable machine learning prediction model for inpatient fall events using EMR integration formats.