3
views
0
recommends
+1 Recommend
1 collections
    0
    shares
      • Record: found
      • Abstract: found
      • Article: found
      Is Open Access

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

      research-article

      , MD 1 , 2 , , BSc 1 , 2 , , PhD 3 , , BSc 1 , 2 , , BSc 1 , 2 , , BSc 1 , 2 , , BSc 1 , 2 , , BSc 1 , 2 , , BSc 1 , 2 , 4 , , BSc 1 , 5 , 6 , , PhD 1 , 2 , , PhD 1 , 2 , 4 , , MD 7 , , MSc 8 , , PhD 8 , , MD, PhD 1 , 9 , , PhD 1 , 4 , , MD, PhD 2 , 4 , 10 , 11 , , MD 1 , 2 , 12 , , PhD 13 , 14 , , MPH, MD 1 , 2 , 15 , 16 , , PhD 3 , , PhD , 1 , 2 , 4

      (Reviewer)

      JMIR Medical Informatics

      JMIR Publications

      federated learning, COVID-19, machine learning, electronic health records

      Read this article at

      Bookmark
          There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

          Abstract

          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 LASSO federated model outperformed the LASSO local model at 3 hospitals, and the MLP federated model performed better than the MLP local model at all 5 hospitals, as determined by the area under the receiver operating characteristic curve. The LASSO pooled model outperformed the LASSO federated model at all hospitals, and the MLP federated model outperformed the MLP pooled 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.

          Related collections

          Most cited references 17

          • Record: found
          • Abstract: found
          • Article: found

          Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China: Summary of a Report of 72 314 Cases From the Chinese Center for Disease Control and Prevention

            Bookmark
            • Record: found
            • Abstract: found
            • Article: not found

            Pathological findings of COVID-19 associated with acute respiratory distress syndrome

             Zhe Xu,  Lei Shi,  Yijin Wang (2020)
            Since late December, 2019, an outbreak of a novel coronavirus disease (COVID-19; previously known as 2019-nCoV)1, 2 was reported in Wuhan, China, 2 which has subsequently affected 26 countries worldwide. In general, COVID-19 is an acute resolved disease but it can also be deadly, with a 2% case fatality rate. Severe disease onset might result in death due to massive alveolar damage and progressive respiratory failure.2, 3 As of Feb 15, about 66 580 cases have been confirmed and over 1524 deaths. However, no pathology has been reported due to barely accessible autopsy or biopsy.2, 3 Here, we investigated the pathological characteristics of a patient who died from severe infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by postmortem biopsies. This study is in accordance with regulations issued by the National Health Commission of China and the Helsinki Declaration. Our findings will facilitate understanding of the pathogenesis of COVID-19 and improve clinical strategies against the disease. A 50-year-old man was admitted to a fever clinic on Jan 21, 2020, with symptoms of fever, chills, cough, fatigue and shortness of breath. He reported a travel history to Wuhan Jan 8–12, and that he had initial symptoms of mild chills and dry cough on Jan 14 (day 1 of illness) but did not see a doctor and kept working until Jan 21 (figure 1 ). Chest x-ray showed multiple patchy shadows in both lungs (appendix p 2), and a throat swab sample was taken. On Jan 22 (day 9 of illness), the Beijing Centers for Disease Control (CDC) confirmed by reverse real-time PCR assay that the patient had COVID-19. Figure 1 Timeline of disease course according to days from initial presentation of illness and days from hospital admission, from Jan 8–27, 2020 SARS-CoV-2=severe acute respiratory syndrome coronavirus 2. He was immediately admitted to the isolation ward and received supplemental oxygen through a face mask. He was given interferon alfa-2b (5 million units twice daily, atomisation inhalation) and lopinavir plus ritonavir (500 mg twice daily, orally) as antiviral therapy, and moxifloxacin (0·4 g once daily, intravenously) to prevent secondary infection. Given the serious shortness of breath and hypoxaemia, methylprednisolone (80 mg twice daily, intravenously) was administered to attenuate lung inflammation. Laboratory tests results are listed in the appendix (p 4). After receiving medication, his body temperature reduced from 39·0 to 36·4 °C. However, his cough, dyspnoea, and fatigue did not improve. On day 12 of illness, after initial presentation, chest x-ray showed progressive infiltrate and diffuse gridding shadow in both lungs. He refused ventilator support in the intensive care unit repeatedly because he suffered from claustrophobia; therefore, he received high-flow nasal cannula (HFNC) oxygen therapy (60% concentration, flow rate 40 L/min). On day 13 of illness, the patient's symptoms had still not improved, but oxygen saturation remained above 95%. In the afternoon of day 14 of illness, his hypoxaemia and shortness of breath worsened. Despite receiving HFNC oxygen therapy (100% concentration, flow rate 40 L/min), oxygen saturation values decreased to 60%, and the patient had sudden cardiac arrest. He was immediately given invasive ventilation, chest compression, and adrenaline injection. Unfortunately, the rescue was not successful, and he died at 18:31 (Beijing time). Biopsy samples were taken from lung, liver, and heart tissue of the patient. Histological examination showed bilateral diffuse alveolar damage with cellular fibromyxoid exudates (figure 2A, B ). The right lung showed evident desquamation of pneumocytes and hyaline membrane formation, indicating acute respiratory distress syndrome (ARDS; figure 2A). The left lung tissue displayed pulmonary oedema with hyaline membrane formation, suggestive of early-phase ARDS (figure 2B). Interstitial mononuclear inflammatory infiltrates, dominated by lymphocytes, were seen in both lungs. Multinucleated syncytial cells with atypical enlarged pneumocytes characterised by large nuclei, amphophilic granular cytoplasm, and prominent nucleoli were identified in the intra-alveolar spaces, showing viral cytopathic-like changes. No obvious intranuclear or intracytoplasmic viral inclusions were identified. Figure 2 Pathological manifestations of right (A) and left (B) lung tissue, liver tissue (C), and heart tissue (D) in a patient with severe pneumonia caused by SARS-CoV-2 SARS-CoV-2=severe acute respiratory syndrome coronavirus 2. The pathological features of COVID-19 greatly resemble those seen in SARS and Middle Eastern respiratory syndrome (MERS) coronavirus infection.4, 5 In addition, the liver biopsy specimens of the patient with COVID-19 showed moderate microvesicular steatosis and mild lobular and portal activity (figure 2C), indicating the injury could have been caused by either SARS-CoV-2 infection or drug-induced liver injury. There were a few interstitial mononuclear inflammatory infiltrates, but no other substantial damage in the heart tissue (figure 2D). Peripheral blood was prepared for flow cytometric analysis. We found that the counts of peripheral CD4 and CD8 T cells were substantially reduced, while their status was hyperactivated, as evidenced by the high proportions of HLA-DR (CD4 3·47%) and CD38 (CD8 39·4%) double-positive fractions (appendix p 3). Moreover, there was an increased concentration of highly proinflammatory CCR6+ Th17 in CD4 T cells (appendix p 3). Additionally, CD8 T cells were found to harbour high concentrations of cytotoxic granules, in which 31·6% cells were perforin positive, 64·2% cells were granulysin positive, and 30·5% cells were granulysin and perforin double-positive (appendix p 3). Our results imply that overactivation of T cells, manifested by increase of Th17 and high cytotoxicity of CD8 T cells, accounts for, in part, the severe immune injury in this patient. X-ray images showed rapid progression of pneumonia and some differences between the left and right lung. In addition, the liver tissue showed moderate microvesicular steatosis and mild lobular activity, but there was no conclusive evidence to support SARS-CoV-2 infection or drug-induced liver injury as the cause. There were no obvious histological changes seen in heart tissue, suggesting that SARS-CoV-2 infection might not directly impair the heart. Although corticosteroid treatment is not routinely recommended to be used for SARS-CoV-2 pneumonia, 1 according to our pathological findings of pulmonary oedema and hyaline membrane formation, timely and appropriate use of corticosteroids together with ventilator support should be considered for the severe patients to prevent ARDS development. Lymphopenia is a common feature in the patients with COVID-19 and might be a critical factor associated with disease severity and mortality. 3 Our clinical and pathological findings in this severe case of COVID-19 can not only help to identify a cause of death, but also provide new insights into the pathogenesis of SARS-CoV-2-related pneumonia, which might help physicians to formulate a timely therapeutic strategy for similar severe patients and reduce mortality. This online publication has been corrected. The corrected version first appeared at thelancet.com/respiratory on February 25, 2020
              Bookmark
              • Record: found
              • Abstract: found
              • Article: not found

              An interactive web-based dashboard to track COVID-19 in real time

              In December, 2019, a local outbreak of pneumonia of initially unknown cause was detected in Wuhan (Hubei, China), and was quickly determined to be caused by a novel coronavirus, 1 namely severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The outbreak has since spread to every province of mainland China as well as 27 other countries and regions, with more than 70 000 confirmed cases as of Feb 17, 2020. 2 In response to this ongoing public health emergency, we developed an online interactive dashboard, hosted by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University, Baltimore, MD, USA, to visualise and track reported cases of coronavirus disease 2019 (COVID-19) in real time. The dashboard, first shared publicly on Jan 22, illustrates the location and number of confirmed COVID-19 cases, deaths, and recoveries for all affected countries. It was developed to provide researchers, public health authorities, and the general public with a user-friendly tool to track the outbreak as it unfolds. All data collected and displayed are made freely available, initially through Google Sheets and now through a GitHub repository, along with the feature layers of the dashboard, which are now included in the Esri Living Atlas. The dashboard reports cases at the province level in China; at the city level in the USA, Australia, and Canada; and at the country level otherwise. During Jan 22–31, all data collection and processing were done manually, and updates were typically done twice a day, morning and night (US Eastern Time). As the outbreak evolved, the manual reporting process became unsustainable; therefore, on Feb 1, we adopted a semi-automated living data stream strategy. Our primary data source is DXY, an online platform run by members of the Chinese medical community, which aggregates local media and government reports to provide cumulative totals of COVID-19 cases in near real time at the province level in China and at the country level otherwise. Every 15 min, the cumulative case counts are updated from DXY for all provinces in China and for other affected countries and regions. For countries and regions outside mainland China (including Hong Kong, Macau, and Taiwan), we found DXY cumulative case counts to frequently lag behind other sources; we therefore manually update these case numbers throughout the day when new cases are identified. To identify new cases, we monitor various Twitter feeds, online news services, and direct communication sent through the dashboard. Before manually updating the dashboard, we confirm the case numbers with regional and local health departments, including the respective centres for disease control and prevention (CDC) of China, Taiwan, and Europe, the Hong Kong Department of Health, the Macau Government, and WHO, as well as city-level and state-level health authorities. For city-level case reports in the USA, Australia, and Canada, which we began reporting on Feb 1, we rely on the US CDC, the government of Canada, the Australian Government Department of Health, and various state or territory health authorities. All manual updates (for countries and regions outside mainland China) are coordinated by a team at Johns Hopkins University. The case data reported on the dashboard aligns with the daily Chinese CDC 3 and WHO situation reports 2 for within and outside of mainland China, respectively (figure ). Furthermore, the dashboard is particularly effective at capturing the timing of the first reported case of COVID-19 in new countries or regions (appendix). With the exception of Australia, Hong Kong, and Italy, the CSSE at Johns Hopkins University has reported newly infected countries ahead of WHO, with Hong Kong and Italy reported within hours of the corresponding WHO situation report. Figure Comparison of COVID-19 case reporting from different sources Daily cumulative case numbers (starting Jan 22, 2020) reported by the Johns Hopkins University Center for Systems Science and Engineering (CSSE), WHO situation reports, and the Chinese Center for Disease Control and Prevention (Chinese CDC) for within (A) and outside (B) mainland China. Given the popularity and impact of the dashboard to date, we plan to continue hosting and managing the tool throughout the entirety of the COVID-19 outbreak and to build out its capabilities to establish a standing tool to monitor and report on future outbreaks. We believe our efforts are crucial to help inform modelling efforts and control measures during the earliest stages of the outbreak.
                Bookmark

                Author and article information

                Contributors
                Journal
                JMIR Med Inform
                JMIR Med Inform
                JMI
                JMIR Medical Informatics
                JMIR Publications (Toronto, Canada )
                2291-9694
                January 2021
                27 January 2021
                27 January 2021
                : 9
                : 1
                Affiliations
                [1 ] The Hasso Plattner Institute for Digital Health at Mount Sinai Icahn School of Medicine at Mount Sinai New York, NY United States
                [2 ] The Mount Sinai Clinical Intelligence Center New York, NY United States
                [3 ] Department of Population Health Sciences Weill Cornell Medicine New York, NY United States
                [4 ] Department of Genetics and Genomic Sciences Icahn School of Medicine at Mount Sinai New York, NY United States
                [5 ] Intelligent System Engineering Indiana University Bloomington, IN United States
                [6 ] School of Information University of Texas Austin Austin, TX United States
                [7 ] Institute for Healthcare Delivery Science Department of Population Health Science and Policy Icahn School of Medicine at Mount Sinai New York, NY United States
                [8 ] Department of Neurological Surgery Icahn School of Medicine at Mount Sinai New York, NY United States
                [9 ] Department of Anesthesiology Perioperative and Pain Medicine Icahn School of Medicine at Mount Sinai New York, NY United States
                [10 ] The Pamela Sklar Division of Psychiatric Genomics Icahn School of Medicine at Mount Sinai New York, NY United States
                [11 ] Department of Psychiatry Icahn School of Medicine at Mount Sinai New York, NY United States
                [12 ] Digital Health Center Hasso Plattner Institute University of Potsdam Potsdam Germany
                [13 ] The BioMedical Engineering and Imaging Institute Icahn School of Medicine at Mount Sinai New York, NY United States
                [14 ] Department of Radiology Icahn School of Medicine at Mount Sinai New York, NY United States
                [15 ] Department of Medicine Icahn School of Medicine at Mount Sinai New York, NY United States
                [16 ] The Charles Bronfman Institute for Personalized Medicine Icahn School of Medicine at Mount Sinai New York, NY United States
                Author notes
                Corresponding Author: Benjamin S Glicksberg benjamin.glicksberg@ 123456mssm.edu
                Article
                v9i1e24207
                10.2196/24207
                7842859
                33400679
                ©Akhil Vaid, Suraj K Jaladanki, Jie Xu, Shelly Teng, Arvind Kumar, Samuel Lee, Sulaiman Somani, Ishan Paranjpe, Jessica K De Freitas, Tingyi Wanyan, Kipp W Johnson, Mesude Bicak, Eyal Klang, Young Joon Kwon, Anthony Costa, Shan Zhao, Riccardo Miotto, Alexander W Charney, Erwin Böttinger, Zahi A Fayad, Girish N Nadkarni, Fei Wang, Benjamin S Glicksberg. Originally published in JMIR Medical Informatics (http://medinform.jmir.org), 27.01.2021.

                This is an open-access article distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on http://medinform.jmir.org/, as well as this copyright and license information must be included.

                Categories
                Original Paper
                Original Paper

                Comments

                Comment on this article