key: cord-1009325-dmro8spg
authors: Satterfield, Benjamin A.; Dikilitas, Ozan; Kullo, Iftikhar J.
title: Leveraging the Electronic Health Record to Address the COVID-19 Pandemic
date: 2021-04-21
journal: Mayo Clin Proc
DOI: 10.1016/j.mayocp.2021.04.008
sha: 6c844357897ddde26fd8f04a0a61aae1e4c6c5a5
doc_id: 1009325
cord_uid: dmro8spg

The coronavirus disease 2019 (COVID-19) pandemic continues its global spread. Coordinated effort on a vast scale is required to halt its progression and save lives. Electronic health record (EHR) data is a valuable resource to mitigate the COVID-19 pandemic. We review how the EHR could be utilized for disease surveillance and contact tracing. When linked to ‘omic’ data, the EHR could facilitate identification of genetic susceptibility variants, leading to insights into risk factors, disease complications, and drug repurposing. Real-time monitoring of patients could enable early detection of potential complications, informing appropriate interventions and therapy. We reviewed relevant articles from PubMED, MEDLINE, and Google Scholar searches as well as preprint servers given the rapidly evolving understanding of the COVID-19 pandemic.

Coronavirus disease 2019 (COVID-19), an ongoing worldwide pandemic disease caused by infection with Severe Acute Respiratory Syndrome Coronavirus-2 (SARS-CoV-2), has a wide spectrum of disease severity with diverse presentation and organ system involvement. More than 120 million cases have been documented and over 2.6 million deaths have resulted worldwide as of March 2021. 1 A coordinated scientific effort on a vast scale is necessary to mitigate the widespread misery, morbidity, and mortality inflicted by the pandemic.

Many similarities have been drawn between the current pandemic and the influenza pandemic of 1918. However, one major difference is the availability of several data sources to track and mitigate the effects of the pandemic. In this review we examine use of the electronic health record (EHR) to address the COVID-19 pandemic. EHR data have the potential to: (1) facilitate disease surveillance and contact tracing; (2) enable risk stratification and real-time monitoring for early detection and management of severe disease; (3) identify risk factors and disease complications, including long-term sequelae; (4) provide insights into pathophysiology using 'omics' approaches; and (5) serve as a platform for innovations related to artificial intelligence, remote monitoring, and early detection of pandemics.

Traditionally, outbreaks of infectious diseases are monitored by active surveillance and contact tracing. Contact tracing is a time-consuming process where the index case detected by surveillance is asked to provide details about other people who were in close contact during the timeframe that puts them at risk of acquiring infection. Public health officials then trace these contacts and inform them about the possible exposure, leading to quarantine and testing. 2 Given the extent of the COVID-19 pandemic and its rapid spread, conventional contact tracing is not feasible in most regions; however mobile device data linked to the EHR is a potential alternative.

Various apps for mobile devices have been developed to collect and share COVID-19 related tracking information. 3 App users can enter information when they test positive with COVID-19, and other users are then alerted if they were within close proximity to an infected individual during a pre-specified amount of time prior to testing positive. 4 This information can also be shared with public health officials depending on the permissions and capabilities of the app 5 and it could be linked to the EHR of health systems, triggering an alert in the system when an individual tests positive for COVID-19 (Figure 1) . The publicprivate collaboration Sync for Science is an example of how individuals could share such EHR data with researchers (http://healthit.gov/topic/sync-science). 6 These approaches may improve modeling of COVID-19 transmission as shown in studies from Brazil 7 and the United States 8 which used mobile phone geolocation data. However, it is important to balance public good with personal privacy when employing such techniques. 9 Tracking of cases and data on infection J o u r n a l P r e -p r o o f rates, deaths, hospitalization, and patient recovery metrics at the local/county, state/district, national, and global levels is critical to inform policy and strategy related to the pandemic. This has led to creation of publicly available online tracking tools such as the COVID-19 Dashboard developed by the Johns Hopkins University (https://coronavirus.jhu.edu/). 1

Risk stratification. Phenotyping algorithms can be deployed across EHR systems for rapid ascertainment of comorbidities, multiorgan complication, and disease severity in patients with COVID-19. Predictive electronic algorithms could identify patients heading towards invasive ventilator support and allow for early interventions to reduce risk of progression. These algorithms leverage rule-based strategies and/or machine learning to help identify patients with a particular clinical profile by querying a wide spectrum of EHR data elements, such as diagnoses and procedures derived using billing codes and/or clinical notes via natural language processing (NLP), laboratory tests, medications, and imaging studies. 10 As case numbers continue to rise throughout the world, substantial EHR data are being generated on patients who have tested positive for COVID-19 and subsequently developed related complications. One study 11 used data from the health system in Ontario, Canada to develop a logistic regression-based model to predict mortality which included age and the presence of certain comorbidities (diabetes, renal disease, and immunocompromised state). In UK Biobank participants, a model including clinical risk factors and 64 single-nucleotide variants was developed to predict risk of severe COVID-

Phenotyping and common data models. The eMERGE network has created Phenotype KnowledgeBase (PheKB, http://phekb.org) 13 to catalog and share electronic phenotyping algorithms across different institutions and conduct large-scale genomic studies. 14 To date, 94 phenotyping algorithms have been validated and 53 of those are finalized and publicly available.

While many of these are relevant to COVID-19, the eMERGE network is currently developing phenotyping algorithms specific for COVID-19-related complications including acute myocardial injury, arrhythmias, ischemic stroke, thromboembolic disease, bleeding diathesis, acute kidney injury and progression to chronic kidney disease, and interstitial lung disease among others. A challenge is to develop algorithms that detect the above complications within a narrow window temporally related to acute COVID-19 infection.

Portability of phenotyping algorithms across platforms can be challenging due to heterogeneity of data representations in different institutions and biorepositories. [15] [16] [17] Recently, PhenX 18, 19 (Phenotypes and eXposures) -a catalog of measurement protocols and bioinformatics tools to promote unified study design, data integration, and analyses among researchers -added a COVID-19 page (https://www.phenxtoolkit.org/covid19) 20 for compiling and distributing COVID-19 protocols currently in use. Common data models, 21 such as the informatics for integrating biology and the bedside (i2b2) 22 or Observational Medical Outcomes Partnership (OMOP), 23 may help harmonize data from diverse EHR systems. Tools to standardize the execution of computable phenotype representations, 24, 25 to assess and quantify the portability of phenotyping algorithms, 10 and machine learning-based methodologies 26 for constructing and sharing phenotype classifiers across sites, are being developed to facilitate COVID-19 research.

Social determinants of health. Numerous studies have demonstrated that COVID-19 is not evenly distributed among the population, with higher rates of COVID-19 positivity, hospitalizations, and deaths in individuals with any of the following characteristics: male sex, older age, non-white ethnicity, higher body mass index, lower income, and smokers (discussed in detail elsewhere). 27-31 Reasons for these disparities likely include factors such as increased number and severity of comorbid conditions, inability to work from home, and lack of access to healthcare among other reasons. Indeed, mobile phone geolocation data 8 reveals that increased rates of infection among disadvantaged socioeconomic groups is in part due to the inability of these groups to reduce mobility to the degree of other groups. Documenting social determinants of health in the EHR can increase our understanding of how these factors are associated with risk of infection and severity of illness.

Comorbidities associated with complications of severe COVID-19. The EHR can aid in the identification of comorbidities related to disease severity and associated complications. Elderly patients with multiple comorbidities make up the majority of severely ill patients with COVID-19. 32-34 These comorbid conditions include obesity, hypertension, diabetes mellitus, and cardiovascular disease, 35 and their presence increases risk of myocardial injury and myocardial infarction, malignant arrhythmias, thromboembolic disease including pulmonary embolism and stroke, acute kidney injury, acute respiratory distress syndrome (ARDS), cytokine storm, and death. 33, 34 Table 1 summarizes these complications and highlights how electronic algorithms linked to clinical decision support can be used for their early detection and management. Laboratory findings associated with poor outcomes include an increasing white-cell count with lymphopenia, a prolonged prothrombin time, and elevated levels of liver enzymes, lactate dehydrogenase, D-dimer, interleukin-6, C-reactive protein (CRP), and procalcitonin. 7, 67, 68 Markers of cardiac, 69 immune, 70 coagulation, 71 muscle, 72 hepatic, 32 and renal 73 injury/dysfunction are also associated with severe disease. Electronic algorithms incorporating real-time laboratory data, especially cardiac troponins and inflammatory markers, could be useful in early detection of complications and linked to clinical decision support for appropriate escalation of care to decrease morbidity and mortality.

Target organ damage and 'Long COVID'. Although most individuals appear to recover completely after infection with SARS-CoV-2, some develop target organ damage and/or longterm symptoms (Long COVID). While there is significant data on acute and short-term complications, not much is known about long-term effects sequelae in COVID-19 survivors. EHR data will be important for tracking and understanding the of COVID-19. Target organ damage can include cardiovascular (eg myocarditis, pericarditis, microvascular angina, arrhythmias), pulmonary (eg interstitial lung disease, chronic pulmonary emboli), neurological (eg myelopathy, neuropathy, neurocognitive disorders), renal (eg chronic kidney disease), and others (eg multisystem inflammatory syndrome in children). 74 For example, in a study 47 of patients who had recovered from COVID-19 with initial diagnosis occurring between 64-92 days previously, the majority continued to have elevated high-sensitivity troponin levels, lower left ventricular systolic function, and imaging evidence of ongoing myocardial inflammation.

Long COVID. Many individuals experience persistent symptoms and a decline in health-related quality of life after COVID-19. 75, 76 These can include a variety of nonspecific symptoms such as chest pain, shortness of breath, fatigue, headache, loss of taste and smell, and brain fog. 74, 75, 77 Whether COVID-19 survivors will completely recover these persistent symptoms is not yet clear. 78 Congress recently allocated $1.15 billion for the National Institutes of Health to support research into Long COVID. 79 This includes efforts to develop a EHR based registry detailing symptoms that is linked to blood, tissue and other samples from patients. EHR data will be valuable to further characterize Long COVID. to patient genotype/sequence data from DNA biorepositories allowed rapid assembly of large national/international consortia to better understand the contribution of common DNA sequence variants to COVID-19 disease severity 91, 92 (Table 2, Figure 3) . 13, 14, 31, 66, 93-106

The first GWAS 91 of COVID-19 severity included ~2,000 COVID-19 patients with respiratory failure from seven hospitals in Europe. Genetic variants in loci involved in inflammation pathways, that functionally interact with the SARS-CoV-2 cellular receptor angiotensinconverting enzyme-2 (ACE2), and the ABO blood groups were associated with severe disease; risk was higher in patients with blood group A than other blood groups, and lower risk in patients with blood group O compared to other blood groups, consistent with previous studies implicating the ABO locus in susceptibility to infection with SARS-CoV-2. 91, 107, 108 An additional GWAS and transcriptome-wide association study (TWAS) from the Genetics of Mortality in Critical Care (GenOMICC) consortium in the UK 109 included 2,244 patients with severe disease and identified variants in multiple genes, many of which are known to be involved in innate immunity, to be associated with disease severity. Some of these variants hint at specific targets for novel drug development. A GWAS 102 meta-analysis of 46 studies including 49,562 cases and over 2,000,000 controls with diverse ancestries found 15 variants (13 of them novel) associated with COVID-19. This study differed from the prior GWAS in that it examined infection susceptibility in addition to hospitalization and severe illness. The 25 genetic susceptibility variants discovered to date are listed in Supplemental Table 1 . As additional genetic susceptibility loci are discovered, it may be possible to combine these into a polygenic risk score for severe disease which could be integrated in the EHR for risk stratification. 110, 111 Rare variant analyses have been enabled by exome/genome sequencing data. The COVID Human Genetics Effort discovered 24 rare, deleterious, genetic variants in 8 genes that mediate type I interferon antiviral immunity that are enriched in patients who develop life-threatening COVID-19 compared to patients with mild or asymptomatic presentations of COVID-19. 38 within the previous 5 years had decreased rates of COVID-19. 126 The EHR-based 'omics' studies mentioned above may provide useful insights into new drug development or drug repurposing.

Targeted medical therapy. Variants can be associated with either improved therapeutic outcomes or adverse reactions for drugs being trialed for COVID-19; these variants are being tracked by the Pharmacogeneomics Knowledgebase (https://www.pharmgkb.org/disease/PA166197121/overview). 127 The presence of such variants can be placed in the EHR to inform decisions related to selecting drug therapy. 128 In addition, EHR-based algorithms and clinical decision support can be used to notify medical providers if patients meet pre-set criteria for use of certain medical therapies to improve patient care; these might include dexamethasone -if supplemental oxygen is being provided, monoclonal antibody therapy -based on age and presence of certain comorbidities, or potential eligibility for experimental therapeutics and clinical trials.

Artificial intelligence and the EHR. Artificial intelligence is being integrated into dashboards within the EHR to assist clinicians in real-time monitoring. Wagner et al 129 131 Additionally, the use of artificial intelligence-enhanced ECG for diagnosis of COVID-19, 132 possibly even in pre-symptomatic patients, is being explored for potential use in portable devices to rapidly screen individuals.

Remote monitoring. Infectious pandemics such as COVID-19 can overwhelm medical facilities, and there is a need for hospital-at-home models to care for less severely ill patients at home, thus decreasing transmission risk and freeing up hospital beds for more critically ill patients. 133 

Telemedicine can be used to manage not only mild disease but also severe disease by providing expanded access to critical care. 134 

To mitigate global pandemics such as COVID-19, international collaborations that share EHR data using common data models are key. Phenotyping algorithms based on billing codes, laboratory and medications data, and NLP of clinical notes are useful for tracking disease epidemiology and severity. Such algorithms enable rapid automated phenotyping for genomic studies and drug repurposing efforts. Data sharing between EHRs of health systems, public health entities, and patients through mobile apps can improve disease tracking and contact tracing to limit spread. EHR surveillance for early detection of complications can improve patient outcomes by linkage to clinical decision support tools for point-of-care patient management. Such surveillance can identify risk stratification to detect high risk individuals who may benefit from increased monitoring either at home or through early hospitalization. Together, various EHR-centered approaches can improve patient care and advance the scientific understanding needed to combat and end the COVID-19 pandemic. Proposed causes include down-regulation of ACE2 leading to unopposed angiotensin II in the setting of direct viral infection, genetic variation in inflammatory cascades, and antibody dependent enhancement from prior exposure to other coronaviruses. Leads to multiorgan damage including myocardial injury and ARDS. May be involved in multisystem inflammatory syndrome.

Elevated CRP, IL-6, leukocyte count.

Baseline leukocyte count and CRP.

Electronic algorithms incorporating vital signs, laboratory values of inflammatory markers, and supplemental oxygen requirements/decline in oxygen saturation could prompt anti-inflammatory therapies (ie dexamethasone, colchicine, or canakinumab) or cardioprotective therapies (ie angiotensin receptor blockers) if trials prove these to be effective.

Defined by elevated troponin levels. Presumably due to hostile inflammatory milieu in many cases, however, may be due direct viral infection, myocardial infarction, or malignant arrhythmias. Associated with adverse outcomes and death in COVID-19 patients.

Elevated levels of troponin, NT-pro BNP, CRP.

Baseline troponin and NT-pro BNP in all individuals at time of hospitalization/ emergency department evaluation. Could be used to identify high risk individuals that should be admitted for close observation who otherwise might be discharged home.

Periodic monitoring of troponin, NT-pro BNP, and/or CRP during hospitalization in high risk individuals could inform decisions on length of stay and need for higher/lower level of care. Electronic algorithms incorporating multiple markers will provide useful prognostication and risk stratification information. 

Higher incidence of stroke in patients with severe COVID-19 may result from hypercoagulable and proinflammatory state.

Brain MRI, head CT. Elevated levels of Ddimer, CRP, IL-6.

Evidence of hypercoagulability or elevated inflammatory markers could prompt more frequent neuro-checks in hospitalized patients, or admission of patients for close monitoring who otherwise might be dismissed from emergency departments.

Real-time NLP of nursing notes for neuro-checks could prompt stroke pager activation for early imaging and subsequent treatment with fibrinolytic therapy. Occurs frequently when patients are in sepsis and/or shock and usually is due acute tubular injury. Associated with poor outcome and may require dialysis.

creatinine or cystatin C. Urinalysis/microscopy. cystatin C. Closely monitoring urine output in hospitalized patients. These are especially important in patients with underlying chronic kidney disease or who present with sepsis.

creatinine and/or cystatin C trends can alert providers of likely acute renal failure and highlight any nephrotoxic medications that could be held.

Likely multifactorial including direct infection of alveolar cells, inflammation including cytokine storm, pulmonary embolism, and neurologic involvement. This can lead to dyspnea with adequate oxygenation, hypoxemia requiring supplemental oxygen or PAP therapy, or ARDS requiring mechanical ventilation, prone positioning, and ECMO.

Chest x-ray, chest CT, ultrasonography. Arterial/venous blood gases with pH.

Oxygen saturation at home, physician's office, emergency department, or hospital.

Real-time electronic monitoring of oxygen saturation, supplemental oxygen requirements, and arterial/venous blood gases with pH with reported respiratory rate can alert providers of impending respiratory failure and need for escalation in management.

62-64 a ACE2, angiotensin-converting enzyme-2; AI, artificial intelligence; aPTT, activated partial thromboplastin time; ARDS, acute respiratory distress syndrome; COVID-19, Coronavirus disease 2019; CRP, C-reactive protein; CT, computed tomography; DIC, disseminated intravascular coagulation; DVT, deep venous thrombosis; ECG, electrocardiogram; ECMO, extracorporeal membrane oxygenation; EHR, electronic health record; IL-6, interleukin 6; MRI, magnetic resonance imaging; NLP, natural language processing; NT-pro BNP, N-terminal pro-B type natriuretic peptide; PAP, positive airway pressure; PE, pulmonary embolism; PT, prothrombin time

COVID-19 Dashboard

Impact of delays on effectiveness of contact tracing strategies for COVID-19: a modelling study

The Role of Health Technology and Informatics in a Global Public Health Emergency: Practices and Implications From the COVID-19 Pandemic

COVID-19 contact tracing and data protection can go together

Global Preparedness Against COVID-19: We Must Leverage the Power of Digital Health

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Mobility network models of COVID-19 explain inequities and inform reopening

Contact Tracing to Manage COVID-19 Spread-Balancing Personal Privacy and Public Health

Omics-Driven Systems Interrogation of Metabolic Dysregulation in COVID-19 Pathogenesis

Metabolomics to Predict Antiviral Drug Efficacy in COVID-19

Proteomic and Metabolomic Characterization of COVID-19

Antihypertensive Drugs and COVID-19 Risk: A Cohort Study of 2 Million Hypertensive Patients

Association between antecedent statin use and decreased mortality in hospitalized patients with COVID-19

Identifying common pharmacotherapies associated with reduced COVID-19 morbidity using electronic health records

Exploratory analysis of immunization records highlights decreased SARS-CoV-2 rates in individuals with recent non-COVID-19 vaccinations

COVID-19: a defining moment for clinical pharmacology?: Wiley Online Library

Augmented curation of clinical notes from a massive EHR system reveals symptoms of impending COVID-19 diagnosis

Emergency use of the ELECT during the COVID-19

Artificial intelligence ECG to detect left ventricular dysfunction in COVID-19: a case series

Artificial intelligence-enhanced electrocardiography in cardiovascular disease management

Hospital-at-Home as an Alternative to Release the Overload of Healthcare Systems During the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) Pandemic

Rapidly Deploying Critical Care Telemedicine Across States and Health Systems During the Covid-19 Pandemic. NEJM Catalyst Innovations in Care Delivery

Tele-ICU: experience to date

Big Data and the Global Public Health Intelligence Network (GPHIN)

Meaningful Big Data Integration for a Global COVID-19 Strategy

A consortium created to foster the sharing of resources, such as protocols to facilitate COVID-19 host genetics research, organize and coordinate analyses across participating sites, and provide a platform to share summary-level and individual-level data among the scientific community. This initiative currently includes ~190 individual studies, a number that will grow as more biobanks/EHR-linked biorepositories contribute COVID-19 related data. 

A repository of patient-level data from many clinical centers within the United States that will use common data models to reveal patterns of risk factors, comorbidities, and testing among COVID-19 patients. This collaborative currently includes over 120,000 positive COVID-19 patients and hundreds of thousands of negative controls.