Deep Learning with Heterogeneous Graph Embeddings for Mortality Prediction from Electronic Health Records

Computational prediction of in-hospital mortality in the setting of an intensive care unit can help clinical practitioners to guide care and make early decisions for interventions. As clinical data are complex and varied in their structure and components, continued innovation of modelling strategies is required to identify architectures that can best model outcomes. In this work, we trained a Heterogeneous Graph Model (HGM) on electronic health record (EHR) data and used the resulting embedding vector as additional information added to a Convolutional Neural Network (CNN) model for predicting in-hospital mortality. We show that the additional information provided by including time as a vector in the embedding captured the relationships between medical concepts, lab tests, and diagnoses, which enhanced predictive performance. We found that adding HGM to a CNN model increased the mortality prediction accuracy up to 4%. This framework served as a foundation for future experiments involving different EHR data types on important healthcare prediction tasks.

Prognostic value of polygenic risk scores for adults with psychosis

Polygenic risk scores (PRS) summarize genetic liability to a disease at the individual level, and the aim is to use them as biomarkers of disease and poor outcomes in real-world clinical practice. To date, few studies have assessed the prognostic value of PRS relative to standards of care. Schizophrenia (SCZ), the archetypal psychotic illness, is an ideal test case for this because the predictive power of the SCZ PRS exceeds that of most other common diseases. Here, we analyzed clinical and genetic data from two multi-ethnic cohorts totaling 8,541 adults with SCZ and related psychotic disorders, to assess whether the SCZ PRS improves the prediction of poor outcomes relative to clinical features captured in a standard psychiatric interview. For all outcomes investigated, the SCZ PRS did not improve the performance of predictive models, an observation that was generally robust to divergent case ascertainment strategies and the ancestral background of the study participants.

Evaluation and management of COVID-19-related severity in people with type 2 diabetes

INTRODUCTION

People with type 2 diabetes (T2D) have an increased rate of hospitalization and mortality related to COVID-19. To identify ahead of time those who are at risk of developing severe diseases and potentially in need of intensive care, we investigated the independent associations between longitudinal glycated hemoglobin (HbA1c), the impact of common medications (metformin, insulin, ACE inhibitors (ACEIs), angiotensin receptor blockers (ARBs), and corticosteroids) and COVID-19 severity in people with T2D.

RESEARCH DESIGN AND METHODS

Retrospective cohort study was conducted using deidentified claims and electronic health record data from the OptumLabs Data Warehouse across the USA between January 2017 and November 2020, including 16 504 individuals with T2D and COVID-19. A univariate model and a multivariate model were applied to evaluate the association between 2 and 3-year HbA1c average, medication use between COVID-19 diagnosis and intensive care unit admission (if applicable), and risk of intensive care related to COVID-19.

RESULTS

With covariates adjusted, the HR of longitudinal HbA1c for risk of intensive care was 1.12 (per 1% increase, p<0.001) and 1.48 (comparing group with poor (HbA1c ≥9%) and adequate glycemic control (HbA1c 6%-9%), p<0.001). The use of corticosteroids and the combined use of insulin and metformin were associated with significant reduction of intensive care risk, while ACEIs and ARBs were not associated with reduced risk of intensive care.

CONCLUSIONS

Two to three-year longitudinal glycemic level is independently associated with COVID-19-related severity in people with T2D. Here, we present a potential method to use HbA1c history, which presented a stronger association with COVID-19 severity than single-point HbA1c, to identify in advance those more at risk of intensive care due to COVID-19 in the T2D population. The combined use of metformin and insulin and the use of corticosteroids might be significant to prevent patients with T2D from becoming critically ill from COVID-19.

The association between obesity and peak antibody titer response in COVID-19 infection

OBJECTIVE

Obesity is associated with severe coronavirus disease 2019 (COVID-19) infection. Disease severity is associated with a higher COVID-19 antibody titer. The COVID-19 antibody titer response of patients with obesity versus patients without obesity was compared.

METHODS

The data of individuals tested for COVID-19 serology at the Mount Sinai Health System in New York City between March 1, 2020, and December 14, 2021, were retrospectively retrieved. The primary outcome was peak antibody titer, assessed as a binary variable (1:2,880, which was the highest detected titer, versus lower than 1:2,880). In patients with a positive serology test, peak titer rates were compared between BMI groups (<18.5, 18.5 to 25, 25 to 30, 30 to 40, and ≥40 kg/m² ). A multivariable logistic regression model was used to analyze the independent association between different BMI groups and peak titer.

RESULTS

Overall, 39,342 individuals underwent serology testing and had BMI measurements. A positive serology test was present in 12,314 patients. Peak titer rates were associated with obesity (BMI < 18.5 [34.5%], 18.5 to 25 [29.2%], 25 to 30 [37.7%], 30 to 40 [44.7%], ≥40 [52.0%]; p < 0.001). In a multivariable analysis, severe obesity had the highest adjusted odds ratio for peak titer (95% CI: 2.1-3.0).

CONCLUSION

COVID-19 neutralizing antibody titer is associated with obesity. This has implications on the understanding of the role of obesity in COVID-19 severity.

Development and validation of techniques for phenotyping ST-elevation myocardial infarction encounters from electronic health records

Objectives

Classifying hospital admissions into various acute myocardial infarction phenotypes in electronic health records (EHRs) is a challenging task with strong research implications that remains unsolved. To our knowledge, this study is the first study to design and validate phenotyping algorithms using cardiac catheterizations to identify not only patients with a ST-elevation myocardial infarction (STEMI), but the specific encounter when it occurred.

Materials and Methods

We design and validate multi-modal algorithms to phenotype STEMI on a multicenter EHR containing 5.1 million patients and 115 million patient encounters by using discharge summaries, diagnosis codes, electrocardiography readings, and the presence of cardiac catheterizations on the encounter.

Results

We demonstrate that robustly phenotyping STEMIs by selecting discharge summaries containing “STEM” has the potential to capture the most number of STEMIs (positive predictive value [PPV] = 0.36, N = 2110), but that addition of a STEMI-related International Classification of Disease (ICD) code and cardiac catheterizations to these summaries yields the highest precision (PPV = 0.94, N = 952).

Discussion and Conclusion

In this study, we demonstrate that the incorporation of percutaneous coronary intervention increases the PPV for detecting STEMI-related patient encounters from the EHR.

Downregulation of exhausted cytotoxic T cells in gene expression networks of multisystem inflammatory syndrome in children

Multisystem inflammatory syndrome in children (MIS-C) presents with fever, inflammation and pathology of multiple organs in individuals under 21 years of age in the weeks following severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection. Although an autoimmune pathogenesis has been proposed, the genes, pathways and cell types causal to this new disease remain unknown. Here we perform RNA sequencing of blood from patients with MIS-C and controls to find disease-associated genes clustered in a co-expression module annotated to CD56dimCD57+ natural killer (NK) cells and exhausted CD8+ T cells. A similar transcriptome signature is replicated in an independent cohort of Kawasaki disease (KD), the related condition after which MIS-C was initially named. Probing a probabilistic causal network previously constructed from over 1,000 blood transcriptomes both validates the structure of this module and reveals nine key regulators, including TBX21, a central coordinator of exhausted CD8+ T cell differentiation. Together, this unbiased, transcriptome-wide survey implicates downregulation of NK cells and cytotoxic T cell exhaustion in the pathogenesis of MIS-C.

FIBER: enabling flexible retrieval of electronic health records data for clinical predictive modeling

Objectives

The development of clinical predictive models hinges upon the availability of comprehensive clinical data. Tapping into such resources requires considerable effort from clinicians, data scientists, and engineers. Specifically, these efforts are focused on data extraction and preprocessing steps required prior to modeling, including complex database queries. A handful of software libraries exist that can reduce this complexity by building upon data standards. However, a gap remains concerning electronic health records (EHRs) stored in star schema clinical data warehouses, an approach often adopted in practice. In this article, we introduce the FlexIBle EHR Retrieval (FIBER) tool: a Python library built on top of a star schema (i2b2) clinical data warehouse that enables flexible generation of modeling-ready cohorts as data frames.

Materials and Methods

FIBER was developed on top of a large-scale star schema EHR database which contains data from 8 million patients and over 120 million encounters. To illustrate FIBER’s capabilities, we present its application by building a heart surgery patient cohort with subsequent prediction of acute kidney injury (AKI) with various machine learning models.

Results

Using FIBER, we were able to build the heart surgery cohort (n = 12 061), identify the patients that developed AKI (n = 1005), and automatically extract relevant features (n = 774). Finally, we trained machine learning models that achieved area under the curve values of up to 0.77 for this exemplary use case.

Conclusion

FIBER is an open-source Python library developed for extracting information from star schema clinical data warehouses and reduces time-to-modeling, helping to streamline the clinical modeling process.

Predictive Approaches for Acute Dialysis Requirement and Death in COVID-19

BACKGROUND AND OBJECTIVES

AKI treated with dialysis initiation is a common complication of coronavirus disease 2019 (COVID-19) among hospitalized patients. However, dialysis supplies and personnel are often limited.

DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS

Using data from adult patients hospitalized with COVID-19 from five hospitals from the Mount Sinai Health System who were admitted between March 10 and December 26, 2020, we developed and validated several models (logistic regression, Least Absolute Shrinkage and Selection Operator (LASSO), random forest, and eXtreme GradientBoosting [XGBoost; with and without imputation]) for predicting treatment with dialysis or death at various time horizons (1, 3, 5, and 7 days) after hospital admission. Patients admitted to the Mount Sinai Hospital were used for internal validation, whereas the other hospitals formed part of the external validation cohort. Features included demographics, comorbidities, and laboratory and vital signs within 12 hours of hospital admission.

RESULTS

A total of 6093 patients (2442 in training and 3651 in external validation) were included in the final cohort. Of the different modeling approaches used, XGBoost without imputation had the highest area under the receiver operating characteristic (AUROC) curve on internal validation (range of 0.93-0.98) and area under the precision-recall curve (AUPRC; range of 0.78-0.82) for all time points. XGBoost without imputation also had the highest test parameters on external validation (AUROC range of 0.85-0.87, and AUPRC range of 0.27-0.54) across all time windows. XGBoost without imputation outperformed all models with higher precision and recall (mean difference in AUROC of 0.04; mean difference in AUPRC of 0.15). Features of creatinine, BUN, and red cell distribution width were major drivers of the model's prediction.

CONCLUSIONS

An XGBoost model without imputation for prediction of a composite outcome of either death or dialysis in patients positive for COVID-19 had the best performance, as compared with standard and other machine learning models.

PODCAST

This article contains a podcast at https://www.asn-online.org/media/podcast/CJASN/2021_07_09_CJN17311120.mp3.

Development of a federated learning approach to predict acute kidney injury in adult hospitalized patients with COVID-19 in New York City

Federated learning is a technique for training predictive models without sharing patient-level data, thus maintaining data security while allowing inter-institutional collaboration. We used federated learning to predict acute kidney injury within three and seven days of admission, using demographics, comorbidities, vital signs, and laboratory values, in 4029 adults hospitalized with COVID-19 at five sociodemographically diverse New York City hospitals, between March-October 2020. Prediction performance of federated models was generally higher than single-hospital models and was comparable to pooled-data models. In the first use-case in kidney disease, federated learning improved prediction of a common complication of COVID-19, while preserving data privacy.

Abstract 174: In-depth transcriptomic comparisons provide novel insights into cancer metastasis

Comparison of the transcriptome between primary and metastatic cancer is frequently performed to study cancer metastasis; however, distilling plausible biology from the comparison results remains a big challenge. We show that in the differential expression (DE) analysis performed on bulk RNAseq data a large proportion of DE genes are identified due to technical issues instead of biological differences between primary and metastatic cancer cells. Next, we develop a computational pipeline DEBoost to minimize such artifacts and apply it to five cancer types. We perform comprehensive analysis on the retained tumor-intrinsic DE genes to explore the biological mechanisms underlying the metastasis of different cancer types. Many of the DE genes are associated with the progression of primary cancer, suggesting the differential expression may have already been established before metastasis. Clustering analysis on DE genes identifies CNVs that account for the transcriptomic change and we further confirm 19p13.12 amplification drives the metastasis of primary Basal-like breast cancer via activating NOTCH signaling pathway. We also find that cell cycling is more active in metastatic prostate cancer than in primary cancer, suggesting the difference of responses to treatment like cell cycling inhibitors. Finally, we show that some metastatic cancer cells tend to express genes whose expression is normally observed at the metastasis site and we experimentally validate this phenomenon in colorectal cancer liver metastasis. Our pan-cancer liver metastasis study provides novel biological insights into cancer metastasis and meanwhile demonstrates the feasibility of characterizing tumor-intrinsic DE genes in cancer metastasis using our computational pipeline.

Citation Format: KE LIU, Benjamin S. Glicksberg, Shreya Paithankar, Bin Chen. In-depth transcriptomic comparisons provide novel insights into cancer metastasis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 174.

Non-invasive ventilation versus mechanical ventilation in hypoxemic patients with COVID-19

Purpose

Limited mechanical ventilators (MV) during the Coronavirus disease (COVID-19) pandemic have led to the use of non-invasive ventilation (NIV) in hypoxemic patients, which has not been studied well. We aimed to assess the association of NIV versus MV with mortality and morbidity during respiratory intervention among hypoxemic patients admitted with COVID-19.

Methods

We performed a retrospective multi-center cohort study across 5 hospitals during March–April 2020. Outcomes included mortality, severe COVID-19-related symptoms, time to discharge, and final oxygen saturation (SpO2) at the conclusion of the respiratory intervention. Multivariable regression of outcomes was conducted in all hypoxemic participants, 4 subgroups, and propensity-matched analysis.

Results

Of 2381 participants with laboratory-confirmed SARS-CoV-2, 688 were included in the study who were hypoxemic upon initiation of respiratory intervention. During the study period, 299 participants died (43%), 163 were admitted to the ICU (24%), and 121 experienced severe COVID-19-related symptoms (18%). Participants on MV had increased mortality than those on NIV (128/154 [83%] versus 171/534 [32%], OR = 30, 95% CI 16–60) with a mean survival of 6 versus 15 days, respectively. The MV group experienced more severe COVID-19-related symptoms [55/154 (36%) versus 66/534 (12%), OR = 4.3, 95% CI 2.7–6.8], longer time to discharge (mean 17 versus 7.1 days), and lower final SpO2 (92 versus 94%). Across all subgroups and propensity-matched analysis, MV was associated with a greater OR of death than NIV.

Conclusions

NIV was associated with lower respiratory intervention mortality and morbidity than MV. However, findings may be liable to unmeasured confounding and further study from randomized controlled trials is needed to definitively determine the role of NIV in hypoxemic patients with COVID-19.

Supplementary Information

The online version contains supplementary material available at 10.1007/s15010-021-01633-6.

Quantification of US Food and Drug Administration Premarket Approval Statements for High-Risk Medical Devices With Pediatric Age Indications

IMPORTANCE

Medical device companies submit premarket approval (PMA) statements to the US Food and Drug Administration (FDA) for approval of the highest-risk class of devices. Devices indicated for the pediatric population that use the PMA pathway have not been well characterized or analyzed.

OBJECTIVE

To identify and characterize high-risk devices with pediatric age indications derived from PMA statements.

DESIGN, SETTING, AND PARTICIPANTS

In this cross-sectional study of PMA statements, those statements containing the words indicated or intended for medical devices listed in the FDA PMA database as of February 2020 were retrieved. Age indications were manually annotated in these approval statements via PubAnnotation. Based on the PMA identification from the PMA statements, device metadata including product codes, regulation numbers, advisory panels, and approval dates were queried.

MAIN OUTCOMES AND MEASURES

The main outcome was discernment of the distribution of devices indicated for the pediatric population (neonate, infant, child, and adolescent). Secondary measures included outlining the clinical specialties, device types, and lag time between the initial approval date and the first date of an approval statement with a pediatric indication for generic device categories.

RESULTS

A total of 297 documents for 149 unique devices were analyzed. Based on the manual age annotations, 102 devices with a pediatric indication, 10 with a neonate age indication, 32 with an infant age indication, 60 with a child age indication, and 94 with an adolescent age indication were identified. For indications for patients from age 17 to 18 years, the number of devices available nearly doubled from 42 devices to 81 devices. Although more than half of the surveyed devices had a pediatric age indication, many were available only for a limited range of the pediatric population (age 18-21 years). For indications for patients from age 0 to 17 years, the mean (SD) number of clinical specialties at each age was 7.27 (1.4), and 12 clinical specialties were represented from ages 18 to 21 years.

CONCLUSIONS AND RELEVANCE

In this cross-sectional study on device PMA statements, a gap was identified in both quantity and diversity of high-risk devices indicated for the pediatric population. Because the current scarcity of pediatric devices may limit therapeutic possibilities for children, this study represents a step toward quantifying this scarcity and identifying clinical specialties with the greatest need for pediatric device innovation and may help inform future device development efforts.

Intestinal Host Response to SARS-CoV-2 Infection and COVID-19 Outcomes in Patients With Gastrointestinal Symptoms

BACKGROUND & AIMS

Given that gastrointestinal (GI) symptoms are a prominent extrapulmonary manifestation of COVID-19, we investigated intestinal infection with SARS-CoV-2, its effect on pathogenesis, and clinical significance.

METHODS

Human intestinal biopsy tissues were obtained from patients with COVID-19 (n = 19) and uninfected control individuals (n = 10) for microscopic examination, cytometry by time of flight analyses, and RNA sequencing. Additionally, disease severity and mortality were examined in patients with and without GI symptoms in 2 large, independent cohorts of hospitalized patients in the United States (N = 634) and Europe (N = 287) using multivariate logistic regressions.

RESULTS

COVID-19 case patients and control individuals in the biopsy cohort were comparable for age, sex, rates of hospitalization, and relevant comorbid conditions. SARS-CoV-2 was detected in small intestinal epithelial cells by immunofluorescence staining or electron microscopy in 15 of 17 patients studied. High-dimensional analyses of GI tissues showed low levels of inflammation, including down-regulation of key inflammatory genes including IFNG, CXCL8, CXCL2, and IL1B and reduced frequencies of proinflammatory dendritic cells compared with control individuals. Consistent with these findings, we found a significant reduction in disease severity and mortality in patients presenting with GI symptoms that was independent of sex, age, and comorbid illnesses and despite similar nasopharyngeal SARS-CoV-2 viral loads. Furthermore, there was reduced levels of key inflammatory proteins in circulation in patients with GI symptoms.

CONCLUSIONS

These data highlight the absence of a proinflammatory response in the GI tract despite detection of SARS-CoV-2. In parallel, reduced mortality in patients with COVID-19 presenting with GI symptoms was observed. A potential role of the GI tract in attenuating SARS-CoV-2-associated inflammation needs to be further examined.

Opportunities and challenges for artificial intelligence in clinical cardiovascular genetics

A combination of emerging genomic and artificial intelligence (AI) techniques may ultimately unlock a deeper understanding of heterogeneity and biological complexities in cardiovascular diseases (CVDs), leading to advances in prognostic guidance and personalized therapies. We discuss the state of AI in cardiovascular genetics, current applications, limitations, and future directions of the field.

Toward a fine-scale population health monitoring system

Understanding population health disparities is an essential component of equitable precision health efforts. Epidemiology research often relies on definitions of race and ethnicity, but these population labels may not adequately capture disease burdens and environmental factors impacting specific sub-populations. Here, we propose a framework for repurposing data from electronic health records (EHRs) in concert with genomic data to explore the demographic ties that can impact disease burdens. Using data from a diverse biobank in New York City, we identified 17 communities sharing recent genetic ancestry. We observed 1,177 health outcomes that were statistically associated with a specific group and demonstrated significant differences in the segregation of genetic variants contributing to Mendelian diseases. We also demonstrated that fine-scale population structure can impact the prediction of complex disease risk within groups. This work reinforces the utility of linking genomic data to EHRs and provides a framework toward fine-scale monitoring of population health.

Containment of COVID-19: Simulating the impact of different policies and testing capacities for contact tracing, testing, and isolation

Efficient contact tracing and testing are fundamental tools to contain the transmission of SARS-CoV-2. We used multi-agent simulations to estimate the daily testing capacity required to find and isolate a number of infected agents sufficient to break the chain of transmission of SARS-CoV-2, so decreasing the risk of new waves of infections. Depending on the non-pharmaceutical mitigation policies in place, the size of secondary infection clusters allowed or the percentage of asymptomatic and paucisymptomatic (i.e., subclinical) infections, we estimated that the daily testing capacity required to contain the disease varies between 0.7 and 9.1 tests per thousand agents in the population. However, we also found that if contact tracing and testing efficacy dropped below 60% (e.g. due to false negatives or reduced tracing capability), the number of new daily infections did not always decrease and could even increase exponentially, irrespective of the testing capacity. Under these conditions, we show that population-level information about geographical distribution and travel behaviour could inform sampling policies to aid a successful containment, while avoiding concerns about government-controlled mass surveillance.

Relational Learning Improves Prediction of Mortality in COVID-19 in the Intensive Care Unit

Traditional Machine Learning (ML) models have had limited success in predicting Coronoavirus-19 (COVID-19) outcomes using Electronic Health Record (EHR) data partially due to not effectively capturing the inter-connectivity patterns between various data modalities. In this work, we propose a novel framework that utilizes relational learning based on a heterogeneous graph model (HGM) for predicting mortality at different time windows in COVID-19 patients within the intensive care unit (ICU). We utilize the EHRs of one of the largest and most diverse patient populations across five hospitals in major health system in New York City. In our model, we use an LSTM for processing time varying patient data and apply our proposed relational learning strategy in the final output layer along with other static features. Here, we replace the traditional softmax layer with a Skip-Gram relational learning strategy to compare the similarity between a patient and outcome embedding representation. We demonstrate that the construction of a HGM can robustly learn the patterns classifying patient representations of outcomes through leveraging patterns within the embeddings of similar patients. Our experimental results show that our relational learning-based HGM model achieves higher area under the receiver operating characteristic curve (auROC) than both comparator models in all prediction time windows, with dramatic improvements to recall.

Combining Initial Radiographs and Clinical Variables Improves Deep Learning Prognostication in Patients with COVID-19 from the Emergency Department

PURPOSE

To train a deep learning classification algorithm to predict chest radiograph severity scores and clinical outcomes in patients with coronavirus disease 2019 (COVID-19).

MATERIALS AND METHODS

In this retrospective cohort study, patients aged 21-50 years who presented to the emergency department (ED) of a multicenter urban health system from March 10 to 26, 2020, with COVID-19 confirmation at real-time reverse-transcription polymerase chain reaction screening were identified. The initial chest radiographs, clinical variables, and outcomes, including admission, intubation, and survival, were collected within 30 days (n = 338; median age, 39 years; 210 men). Two fellowship-trained cardiothoracic radiologists examined chest radiographs for opacities and assigned a clinically validated severity score. A deep learning algorithm was trained to predict outcomes on a holdout test set composed of patients with confirmed COVID-19 who presented between March 27 and 29, 2020 (n = 161; median age, 60 years; 98 men) for both younger (age range, 21-50 years; n = 51) and older (age >50 years, n = 110) populations. Bootstrapping was used to compute CIs.

RESULTS

The model trained on the chest radiograph severity score produced the following areas under the receiver operating characteristic curves (AUCs): 0.80 (95% CI: 0.73, 0.88) for the chest radiograph severity score, 0.76 (95% CI: 0.68, 0.84) for admission, 0.66 (95% CI: 0.56, 0.75) for intubation, and 0.59 (95% CI: 0.49, 0.69) for death. The model trained on clinical variables produced an AUC of 0.64 (95% CI: 0.55, 0.73) for intubation and an AUC of 0.59 (95% CI: 0.50, 0.68) for death. Combining chest radiography and clinical variables increased the AUC of intubation and death to 0.88 (95% CI: 0.79, 0.96) and 0.82 (95% CI: 0.72, 0.91), respectively.

CONCLUSION

The combination of imaging and clinical information improves outcome predictions.Supplemental material is available for this article.© RSNA, 2020.

Comparison of Approaches for Prediction of Renal Replacement Therapy-Free Survival in Patients with Acute Kidney Injury

BACKGROUND/AIMS

Acute kidney injury (AKI) in critically ill patients is common, and continuous renal replacement therapy (CRRT) is a preferred mode of renal replacement therapy (RRT) in hemodynamically unstable patients. Prediction of clinical outcomes in patients on CRRT is challenging. We utilized several approaches to predict RRT-free survival (RRTFS) in critically ill patients with AKI requiring CRRT.

METHODS

We used the Medical Information Mart for Intensive Care (MIMIC-III) database to identify patients ≥18 years old with AKI on CRRT, after excluding patients who had ESRD on chronic dialysis, and kidney transplantation. We defined RRTFS as patients who were discharged alive and did not require RRT ≥7 days prior to hospital discharge. We utilized all available biomedical data up to CRRT initiation. We evaluated 7 approaches, including logistic regression (LR), random forest (RF), support vector machine (SVM), adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost), multilayer perceptron (MLP), and MLP with long short-term memory (MLP + LSTM). We evaluated model performance by using area under the receiver operating characteristic (AUROC) curves.

RESULTS

Out of 684 patients with AKI on CRRT, 205 (30%) patients had RRTFS. The median age of patients was 63 years and their median Simplified Acute Physiology Score (SAPS) II was 67 (interquartile range 52-84). The MLP + LSTM showed the highest AUROC (95% CI) of 0.70 (0.67-0.73), followed by MLP 0.59 (0.54-0.64), LR 0.57 (0.52-0.62), SVM 0.51 (0.46-0.56), AdaBoost 0.51 (0.46-0.55), RF 0.44 (0.39-0.48), and XGBoost 0.43 (CI 0.38-0.47).

CONCLUSIONS

A MLP + LSTM model outperformed other approaches for predicting RRTFS. Performance could be further improved by incorporating other data types.

Use of Physiological Data From a Wearable Device to Identify SARS-CoV-2 Infection and Symptoms and Predict COVID-19 Diagnosis: Observational Study

BACKGROUND

Changes in autonomic nervous system function, characterized by heart rate variability (HRV), have been associated with infection and observed prior to its clinical identification.

OBJECTIVE

We performed an evaluation of HRV collected by a wearable device to identify and predict COVID-19 and its related symptoms.

METHODS

Health care workers in the Mount Sinai Health System were prospectively followed in an ongoing observational study using the custom Warrior Watch Study app, which was downloaded to their smartphones. Participants wore an Apple Watch for the duration of the study, measuring HRV throughout the follow-up period. Surveys assessing infection and symptom-related questions were obtained daily.

RESULTS

Using a mixed-effect cosinor model, the mean amplitude of the circadian pattern of the standard deviation of the interbeat interval of normal sinus beats (SDNN), an HRV metric, differed between subjects with and without COVID-19 (P=.006). The mean amplitude of this circadian pattern differed between individuals during the 7 days before and the 7 days after a COVID-19 diagnosis compared to this metric during uninfected time periods (P=.01). Significant changes in the mean and amplitude of the circadian pattern of the SDNN was observed between the first day of reporting a COVID-19-related symptom compared to all other symptom-free days (P=.01).

CONCLUSIONS

Longitudinally collected HRV metrics from a commonly worn commercial wearable device (Apple Watch) can predict the diagnosis of COVID-19 and identify COVID-19-related symptoms. Prior to the diagnosis of COVID-19 by nasal swab polymerase chain reaction testing, significant changes in HRV were observed, demonstrating the predictive ability of this metric to identify COVID-19 infection.

Deep learning and the electrocardiogram: review of the current state-of-the-art

In the recent decade, deep learning, a subset of artificial intelligence and machine learning, has been used to identify patterns in big healthcare datasets for disease phenotyping, event predictions, and complex decision making. Public datasets for electrocardiograms (ECGs) have existed since the 1980s and have been used for very specific tasks in cardiology, such as arrhythmia, ischemia, and cardiomyopathy detection. Recently, private institutions have begun curating large ECG databases that are orders of magnitude larger than the public databases for ingestion by deep learning models. These efforts have demonstrated not only improved performance and generalizability in these aforementioned tasks but also application to novel clinical scenarios. This review focuses on orienting the clinician towards fundamental tenets of deep learning, state-of-the-art prior to its use for ECG analysis, and current applications of deep learning on ECGs, as well as their limitations and future areas of improvement.

Federated Learning of Electronic Health Records to Improve Mortality Prediction in Hospitalized Patients With COVID-19: Machine Learning Approach

BACKGROUND

Machine learning models require large datasets that may be siloed across different health care institutions. Machine learning studies that focus on COVID-19 have been limited to single-hospital data, which limits model generalizability.

OBJECTIVE

We aimed to use federated learning, a machine learning technique that avoids locally aggregating raw clinical data across multiple institutions, to predict mortality in hospitalized patients with COVID-19 within 7 days.

METHODS

Patient data were collected from the electronic health records of 5 hospitals within the Mount Sinai Health System. Logistic regression with L1 regularization/least absolute shrinkage and selection operator (LASSO) and multilayer perceptron (MLP) models were trained by using local data at each site. We developed a pooled model with combined data from all 5 sites, and a federated model that only shared parameters with a central aggregator.

RESULTS

The LASSOfederated model outperformed the LASSOlocal model at 3 hospitals, and the MLPfederated model performed better than the MLPlocal model at all 5 hospitals, as determined by the area under the receiver operating characteristic curve. The LASSOpooled model outperformed the LASSOfederated model at all hospitals, and the MLPfederated model outperformed the MLPpooled model at 2 hospitals.

CONCLUSIONS

The federated learning of COVID-19 electronic health record data shows promise in developing robust predictive models without compromising patient privacy.