A roadmap to artificial intelligence (AI): Methods for designing and building AI ready data to promote fairness.

OBJECTIVES: We evaluated methods for preparing electronic health record data to reduce bias before applying artificial intelligence (AI). METHODS: We created methods for transforming raw data into a data framework for applying machine learning and natural language processing techniques for predicting falls and fractures. Strategies such as inclusion and reporting for multiple races, mixed data sources such as outpatient, inpatient, structured codes, and unstructured notes, and addressing missingness were applied to raw data to promote a reduction in bias. The raw data was carefully curated using validated definitions to create data variables such as age, race, gender, and healthcare utilization. For the formation of these variables, clinical, statistical, and data expertise were used. The research team included a variety of experts with diverse professional and demographic backgrounds to include diverse perspectives. RESULTS: For the prediction of falls, information extracted from radiology reports was converted to a matrix for applying machine learning. The processing of the data resulted in an input of 5,377,673 reports to the machine learning algorithm, out of which 45,304 were flagged as positive and 5,332,369 as negative for falls. Processed data resulted in lower missingness and a better representation of race and diagnosis codes. For fractures, specialized algorithms extracted snippets of text around keywork "femoral" from dual x-ray absorptiometry (DXA) scans to identify femoral neck T-scores that are important for predicting fracture risk. The natural language processing algorithms yielded 98% accuracy and 2% error rate The methods to prepare data for input to artificial intelligence processes are reproducible and can be applied to other studies. CONCLUSION: The life cycle of data from raw to analytic form includes data governance, cleaning, management, and analysis. When applying artificial intelligence methods, input data must be prepared optimally to reduce algorithmic bias, as biased output is harmful. Building AI-ready data frameworks that improve efficiency can contribute to transparency and reproducibility. The roadmap for the application of AI involves applying specialized techniques to input data, some of which are suggested here. This study highlights data curation aspects to be considered when preparing data for the application of artificial intelligence to reduce bias.

Patterns of COVID-19 testing and mortality by race and ethnicity among United States veterans: A nationwide cohort study.

BACKGROUND: There is growing concern that racial and ethnic minority communities around the world are experiencing a disproportionate burden of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and coronavirus disease 2019 (COVID-19). We investigated racial and ethnic disparities in patterns of COVID-19 testing (i.e., who received testing and who tested positive) and subsequent mortality in the largest integrated healthcare system in the United States. METHODS AND FINDINGS: This retrospective cohort study included 5,834,543 individuals receiving care in the US Department of Veterans Affairs; most (91%) were men, 74% were non-Hispanic White (White), 19% were non-Hispanic Black (Black), and 7% were Hispanic. We evaluated associations between race/ethnicity and receipt of COVID-19 testing, a positive test result, and 30-day mortality, with multivariable adjustment for a wide range of demographic and clinical characteristics including comorbid conditions, health behaviors, medication history, site of care, and urban versus rural residence. Between February 8 and July 22, 2020, 254,595 individuals were tested for COVID-19, of whom 16,317 tested positive and 1,057 died. Black individuals were more likely to be tested (rate per 1,000 individuals: 60.0, 95% CI 59.6-60.5) than Hispanic (52.7, 95% CI 52.1-53.4) and White individuals (38.6, 95% CI 38.4-38.7). While individuals from minority backgrounds were more likely to test positive (Black versus White: odds ratio [OR] 1.93, 95% CI 1.85-2.01, p < 0.001; Hispanic versus White: OR 1.84, 95% CI 1.74-1.94, p < 0.001), 30-day mortality did not differ by race/ethnicity (Black versus White: OR 0.97, 95% CI 0.80-1.17, p = 0.74; Hispanic versus White: OR 0.99, 95% CI 0.73-1.34, p = 0.94). The disparity between Black and White individuals in testing positive for COVID-19 was stronger in the Midwest (OR 2.66, 95% CI 2.41-2.95, p < 0.001) than the West (OR 1.24, 95% CI 1.11-1.39, p < 0.001). The disparity in testing positive for COVID-19 between Hispanic and White individuals was consistent across region, calendar time, and outbreak pattern. Study limitations include underrepresentation of women and a lack of detailed information on social determinants of health. CONCLUSIONS: In this nationwide study, we found that Black and Hispanic individuals are experiencing an excess burden of SARS-CoV-2 infection not entirely explained by underlying medical conditions or where they live or receive care. There is an urgent need to proactively tailor strategies to contain and prevent further outbreaks in racial and ethnic minority communities.

Identification of hepatic steatosis among persons with and without HIV using natural language processing.

BACKGROUND: Steatotic liver disease (SLD) is a growing phenomenon, and our understanding of its determinants has been limited by our ability to identify it clinically. Natural language processing (NLP) can potentially identify hepatic steatosis systematically within large clinical repositories of imaging reports. We validated the performance of an NLP algorithm for the identification of SLD in clinical imaging reports and applied this tool to a large population of people with and without HIV. METHODS: Patients were included in the analysis if they enrolled in the Veterans Aging Cohort Study between 2001 and 2017, had an imaging report inclusive of the liver, and had ≥2 years of observation before the imaging study. SLD was considered present when reports contained the terms "fatty," "steatosis," "steatotic," or "steatohepatitis." The performance of the SLD NLP algorithm was compared to a clinical review of 800 reports. We then applied the NLP algorithm to the first eligible imaging study and compared patient characteristics by SLD and HIV status. RESULTS: NLP achieved 100% sensitivity and 88.5% positive predictive value for the identification of SLD. When applied to 26,706 eligible Veterans Aging Cohort Study patient imaging reports, SLD was identified in 72.2% and did not significantly differ by HIV status. SLD was associated with a higher prevalence of metabolic comorbidities, alcohol use disorder, and hepatitis B and C, but not HIV infection. CONCLUSIONS: While limited to those undergoing radiologic study, the NLP algorithm accurately identified SLD in people with and without HIV and offers a valuable tool to evaluate the determinants and consequences of hepatic steatosis.

A Roadmap to Artificial Intelligence (AI): Methods for Designing and Building AI ready Data for Women's Health Studies.

Objectives: Evaluating methods for building data frameworks for application of AI in large scale datasets for women's health studies. Methods: We created methods for transforming raw data to a data framework for applying machine learning (ML) and natural language processing (NLP) techniques for predicting falls and fractures. Results: Prediction of falls was higher in women compared to men. Information extracted from radiology reports was converted to a matrix for applying machine learning. For fractures, by applying specialized algorithms, we extracted snippets from dual x-ray absorptiometry (DXA) scans for meaningful terms usable for predicting fracture risk. Discussion: Life cycle of data from raw to analytic form includes data governance, cleaning, management, and analysis. For applying AI, data must be prepared optimally to reduce algorithmic bias. Conclusion: Algorithmic bias is harmful for research using AI methods. Building AI ready data frameworks that improve efficiency can be especially valuable for women's health.

Performance of a modified fracture risk assessment tool for fragility fracture prediction among older veterans living with HIV.

OBJECTIVE: Fragility fractures (fractures) are a critical outcome for persons aging with HIV (PAH). Research suggests that the fracture risk assessment tool (FRAX) only modestly estimates fracture risk among PAH. We provide an updated evaluation of how well a 'modified FRAX' identifies PAH at risk for fractures in a contemporary HIV cohort. DESIGN: Cohort study. METHODS: We used data from the Veterans Aging Cohort Study to evaluate veterans living with HIV, aged 50+ years, for the occurrence of fractures from 1 January 2010 through 31 December 2019. Data from 2009 were used to evaluate the eight FRAX predictors available to us: age, sex, BMI, history of previous fracture, glucocorticoid use, rheumatoid arthritis, alcohol use, and smoking status. These predictor values were then used to estimate participant risk for each of two types of fractures (major osteoporotic and hip) over the subsequent 10 years in strata defined by race/ethnicity using multivariable logistic regression. RESULTS: Discrimination for major osteoporotic fracture was modest [Blacks: area under the curve (AUC) 0.62; 95% confidence interval (CI) 0.62, 0.63; Whites: AUC 0.61; 95% CI 0.60, 0.61; Hispanic: AUC 0.63; 95% CI 0.62, 0.65]. For hip fractures, discrimination was modest to good (Blacks: AUC 0.70; 95% CI 0.69, 0.71; Whites: AUC 0.68; 95% CI 0.67, 0.69]. Calibration was good in all models across all racial/ethnic groups. CONCLUSION: Our 'modified FRAX' exhibited modest discrimination for predicting major osteoporotic fracture and slightly better discrimination for hip fracture. Future studies should explore whether augmentation of this subset of FRAX predictors results in enhanced prediction of fractures among PAH.

Pharmacogenomics driven decision support prototype with machine learning: A framework for improving patient care.

Introduction: A growing number of healthcare providers make complex treatment decisions guided by electronic health record (EHR) software interfaces. Many interfaces integrate multiple sources of data (e.g., labs, pharmacy, diagnoses) successfully, though relatively few have incorporated genetic data. Method: This study utilizes informatics methods with predictive modeling to create and validate algorithms to enable informed pharmacogenomic decision-making at the point of care in near real-time. The proposed framework integrates EHR and genetic data relevant to the patient's current medications including decision support mechanisms based on predictive modeling. We created a prototype with EHR and linked genetic data from the Department of Veterans Affairs (VA), the largest integrated healthcare system in the US. The EHR data included diagnoses, medication fills, and outpatient clinic visits for 2,600 people with HIV and matched uninfected controls linked to prototypic genetic data (variations in single or multiple positions in the DNA sequence). We then mapped the medications that patients were prescribed to medications defined in the drug-gene interaction mapping of the Clinical Pharmacogenomics Implementation Consortium's (CPIC) level A (i.e., sufficient evidence for at least one prescribing action) guidelines that predict adverse events. CPIC is a National Institute of Health funded group of experts who develop evidence based pharmacogenomic guidelines. Preventable adverse events (PAE) can be defined as a harmful outcome from an intervention that could have been prevented. For this study, we focused on potential PAEs resulting from a medication-gene interaction. Results: The final model showed AUC scores of 0.972 with an F1 score of 0.97 with genetic data as compared to 0.766 and 0.73 respectively, without genetic data integration. Discussion: Over 98% of people in the cohort were on at least one medication with CPIC level a guideline in their lifetime. We compared predictive power of machine learning models to detect a PAE between five modeling methods: Random Forest, Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), K Nearest neighbors (KNN), and Decision Tree. We found that XGBoost performed best for the prototype when genetic data was added to the framework and improved prediction of PAE. We compared area under the curve (AUC) between the models in the testing dataset.

HIV care using differentiated service delivery during the COVID-19 pandemic: a nationwide cohort study in the US Department of Veterans Affairs.

INTRODUCTION: The Department of Veterans Affairs (VA) is the largest provider of HIV care in the United States. Changes in healthcare delivery became necessary with the COVID-19 pandemic. We compared HIV healthcare delivery during the first year of the COVID-19 pandemic to a prior similar calendar period. METHODS: We included 27,674 people with HIV (PWH) enrolled in the Veterans Aging Cohort Study prior to 1 March 2019, with ≥1 healthcare encounter from 1 March 2019 to 29 February 2020 (2019) and/or 1 March 2020 to 28 February 2021 (2020). We counted monthly general medicine/infectious disease (GM/ID) clinic visits and HIV-1 RNA viral load (VL) tests. We determined the percentage with ≥1 clinic visit (in-person vs. telephone/video [virtual]) and ≥1 VL test (detectable vs. suppressed) for 2019 and 2020. Using pharmacy records, we summarized antiretroviral (ARV) medication refill length (<90 vs. ≥90 days) and monthly ARV coverage. RESULTS: Most patients had ≥1 GM/ID visit in 2019 (96%) and 2020 (95%). For 2019, 27% of visits were virtual compared to 64% in 2020. In 2019, 82% had VL measured compared to 74% in 2020. Of those with VL measured, 92% and 91% had suppressed VL in 2019 and 2020. ARV refills for ≥90 days increased from 39% in 2019 to 51% in 2020. ARV coverage was similar for all months of 2019 and 2020 ranging from 76% to 80% except for March 2019 (72%). Women were less likely than men to be on ARVs or to have a VL test in both years. CONCLUSIONS: During the COVID-19 pandemic, the VA increased the use of virtual visits and longer ARV refills, while maintaining a high percentage of patients with suppressed VL among those with VL measured. Despite decreased in-person services during the pandemic, access to ARVs was not disrupted. More follow-up time is needed to determine whether overall health was impacted by the use of differentiated service delivery and to evaluate whether a long-term shift to increased virtual healthcare could be beneficial, particularly for PWH in rural areas or with transportation barriers. Programmes to increase ARV use and VL testing for women are needed.

Early initiation of prophylactic anticoagulation for prevention of coronavirus disease 2019 mortality in patients admitted to hospital in the United States: cohort study.

OBJECTIVE: To evaluate whether early initiation of prophylactic anticoagulation compared with no anticoagulation was associated with decreased risk of death among patients admitted to hospital with coronavirus disease 2019 (covid-19) in the United States. DESIGN: Observational cohort study. SETTING: Nationwide cohort of patients receiving care in the Department of Veterans Affairs, a large integrated national healthcare system. PARTICIPANTS: All 4297 patients admitted to hospital from 1 March to 31 July 2020 with laboratory confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and without a history of anticoagulation. MAIN OUTCOME MEASURES: The main outcome was 30 day mortality. Secondary outcomes were inpatient mortality, initiating therapeutic anticoagulation (a proxy for clinical deterioration, including thromboembolic events), and bleeding that required transfusion. RESULTS: Of 4297 patients admitted to hospital with covid-19, 3627 (84.4%) received prophylactic anticoagulation within 24 hours of admission. More than 99% (n=3600) of treated patients received subcutaneous heparin or enoxaparin. 622 deaths occurred within 30 days of hospital admission, 513 among those who received prophylactic anticoagulation. Most deaths (510/622, 82%) occurred during hospital stay. Using inverse probability of treatment weighted analyses, the cumulative incidence of mortality at 30 days was 14.3% (95% confidence interval 13.1% to 15.5%) among those who received prophylactic anticoagulation and 18.7% (15.1% to 22.9%) among those who did not. Compared with patients who did not receive prophylactic anticoagulation, those who did had a 27% decreased risk for 30 day mortality (hazard ratio 0.73, 95% confidence interval 0.66 to 0.81). Similar associations were found for inpatient mortality and initiation of therapeutic anticoagulation. Receipt of prophylactic anticoagulation was not associated with increased risk of bleeding that required transfusion (hazard ratio 0.87, 0.71 to 1.05). Quantitative bias analysis showed that results were robust to unmeasured confounding (e-value lower 95% confidence interval 1.77 for 30 day mortality). Results persisted in several sensitivity analyses. CONCLUSIONS: Early initiation of prophylactic anticoagulation compared with no anticoagulation among patients admitted to hospital with covid-19 was associated with a decreased risk of 30 day mortality and no increased risk of serious bleeding events. These findings provide strong real world evidence to support guidelines recommending the use of prophylactic anticoagulation as initial treatment for patients with covid-19 on hospital admission.

Extracting lung function measurements to enhance phenotyping of chronic obstructive pulmonary disease (COPD) in an electronic health record using automated tools.

BACKGROUND: Chronic obstructive pulmonary disease (COPD) is associated with poor quality of life, hospitalization and mortality. COPD phenotype includes using pulmonary function tests to determine airflow obstruction from the forced expiratory volume in one second (FEV1):forced vital capacity. FEV1 is a commonly used value for severity but is difficult to identify in structured electronic health record (EHR) data. DATA SOURCE AND METHODS: Using the Microsoft SQL Server's full-text search feature and string functions supporting regular-expression-like operations, we developed an automated tool to extract FEV1 values from progress notes to improve ascertainment of FEV1 in EHR in the Veterans Aging Cohort Study (VACS). RESULTS: The automated tool increased quantifiable FEV1 values from 12,425 to 16,274 (24% increase in numeric FEV1). Using chart review as the reference, positive predictive value of the tool was 99% (95% Confidence interval: 98.2-100.0%) for identifying quantifiable FEV1 values and a recall value of 100%, yielding an F-measure of 0.99. The tool correctly identified FEV1 measurements in 95% of cases. CONCLUSION: A SQL-based full text search of clinical notes for quantifiable FEV1 is efficient and improves the number of values available in VA data. Future work will examine how these methods can improve phenotyping of patients with COPD in the VA.

Development and validation of a 30-day mortality index based on pre-existing medical administrative data from 13,323 COVID-19 patients: The Veterans Health Administration COVID-19 (VACO) Index.

BACKGROUND: Available COVID-19 mortality indices are limited to acute inpatient data. Using nationwide medical administrative data available prior to SARS-CoV-2 infection from the US Veterans Health Administration (VA), we developed the VA COVID-19 (VACO) 30-day mortality index and validated the index in two independent, prospective samples. METHODS AND FINDINGS: We reviewed SARS-CoV-2 testing results within the VA between February 8 and August 18, 2020. The sample was split into a development cohort (test positive between March 2 and April 15, 2020), an early validation cohort (test positive between April 16 and May 18, 2020), and a late validation cohort (test positive between May 19 and July 19, 2020). Our logistic regression model in the development cohort considered demographics (age, sex, race/ethnicity), and pre-existing medical conditions and the Charlson Comorbidity Index (CCI) derived from ICD-10 diagnosis codes. Weights were fixed to create the VACO Index that was then validated by comparing area under receiver operating characteristic curves (AUC) in the early and late validation cohorts and among important validation cohort subgroups defined by sex, race/ethnicity, and geographic region. We also evaluated calibration curves and the range of predictions generated within age categories. 13,323 individuals tested positive for SARS-CoV-2 (median age: 63 years; 91% male; 42% non-Hispanic Black). We observed 480/3,681 (13%) deaths in development, 253/2,151 (12%) deaths in the early validation cohort, and 403/7,491 (5%) deaths in the late validation cohort. Age, multimorbidity described with CCI, and a history of myocardial infarction or peripheral vascular disease were independently associated with mortality-no other individual comorbid diagnosis provided additional information. The VACO Index discriminated mortality in development (AUC = 0.79, 95% CI: 0.77-0.81), and in early (AUC = 0.81 95% CI: 0.78-0.83) and late (AUC = 0.84, 95% CI: 0.78-0.86) validation. The VACO Index allows personalized estimates of 30-day mortality after COVID-19 infection. For example, among those aged 60-64 years, overall mortality was estimated at 9% (95% CI: 6-11%). The Index further discriminated risk in this age stratum from 4% (95% CI: 3-7%) to 21% (95% CI: 12-31%), depending on sex and comorbid disease. CONCLUSION: Prior to infection, demographics and comorbid conditions can discriminate COVID-19 mortality risk overall and within age strata. The VACO Index reproducibly identified individuals at substantial risk of COVID-19 mortality who might consider continuing social distancing, despite relaxed state and local guidelines.

Differences in Pathology, Staging, and Treatment between HIV+ and Uninfected Patients with Microscopically Confirmed Hepatocellular Carcinoma.

BACKGROUND: The incidence of hepatocellular carcinoma (HCC) is substantially higher among HIV-infected (HIV+) than uninfected persons. It remains unclear if HCC in the setting of HIV infection is morphologically distinct or more aggressive. METHODS: We evaluated differences in tumor pathology in a cohort of HIV+ and uninfected patients with microscopically confirmed HCC in the Veterans Aging Cohort Study from 2000 to 2015. We reviewed pathology reports and medical records to determine Barcelona Clinic Liver Cancer stage (BCLC), HCC treatment, and survival by HIV status. Multivariable Cox regression was used to determine the hazard ratio [HR; 95% confidence interval (CI)] of death associated with HIV infection after microscopic confirmation. RESULTS: Among 873 patients with HCC (399 HIV+), 140 HIV+ and 178 uninfected persons underwent liver tissue sampling and had microscopically confirmed HCC. There were no differences in histologic features of the tumor between HIV+ and uninfected patients, including tumor differentiation (well differentiated, 19% vs. 28%, P = 0.16) and lymphovascular invasion (6% vs. 7%, P = 0.17) or presence of advanced hepatic fibrosis (40% vs. 39%, P = 0.90). There were no differences in BCLC stage (P = 0.06) or treatment (P = 0.29) by HIV status. After adjustment for risk factors, risk of death was higher among HIV-infected than uninfected patients (HR = 1.37; 95% CI, 1.02-1.85). CONCLUSIONS: We found no differences in HCC tumor characteristics or background hepatic parenchyma by HIV status, yet HIV was associated with poorer survival. Of note, pathology reports often omitted these characteristics. IMPACT: Systematic evaluation of HCC pathology by HIV status is needed to understand tumor characteristics associated with improved survival.

Covid-19 by Race and Ethnicity: A National Cohort Study of 6 Million United States Veterans.

BACKGROUND: There is growing concern that racial and ethnic minority communities around the world are experiencing a disproportionate burden of morbidity and mortality from symptomatic SARS-Cov-2 infection or coronavirus disease 2019 (Covid-19). Most studies investigating racial and ethnic disparities to date have focused on hospitalized patients or have not characterized who received testing or those who tested positive for Covid-19. OBJECTIVE: To compare patterns of testing and test results for coronavirus 2019 (Covid-19) and subsequent mortality by race and ethnicity in the largest integrated healthcare system in the United States. DESIGN: Retrospective cohort study. SETTING: United States Department of Veterans Affairs (VA). PARTICIPANTS: 5,834,543 individuals in care, among whom 62,098 were tested and 5,630 tested positive for Covid-19 between February 8 and May 4, 2020. Exposures: Self-reported race/ethnicity. MAIN OUTCOME MEASURES: We evaluated associations between race/ethnicity and receipt of Covid-19 testing, a positive test result, and 30-day mortality, accounting for a wide range of demographic and clinical risk factors including comorbid conditions, site of care, and urban versus rural residence. RESULTS: Among all individuals in care, 74% were non-Hispanic white (white), 19% non-Hispanic black (black), and 7% Hispanic. Compared with white individuals, black and Hispanic individuals were more likely to be tested for Covid-19 (tests per 1000: white=9.0, [95% CI 8.9 to 9.1]; black=16.4, [16.2 to 16.7]; and Hispanic=12.2, [11.9 to 12.5]). While individuals from minority backgrounds were more likely to test positive (black vs white: OR 1.96, 95% CI 1.81 to 2.12; Hispanic vs white: OR 1.73, 95% CI 1.53 to 1.96), 30-day mortality did not differ by race/ethnicity (black vs white: OR 0.93, 95% CI 0.64 to 1.33; Hispanic vs white: OR 1.07, 95% CI 0.61 to 1.87). CONCLUSIONS: Black and Hispanic individuals are experiencing an excess burden of Covid-19 not entirely explained by underlying medical conditions or where they live or receive care. While there was no observed difference in mortality by race or ethnicity, our findings may underestimate risk in the broader US population as health disparities tend to be reduced in VA.

Covid-19 Testing, Hospital Admission, and Intensive Care Among 2,026,227 United States Veterans Aged 54-75 Years.

IMPORTANCE: Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection causes coronavirus disease 2019 (Covid-19), an evolving pandemic. Limited data are available characterizing SARS-Cov-2 infection in the United States. OBJECTIVE: To determine associations between demographic and clinical factors and testing positive for coronavirus 2019 (Covid-19+), and among Covid-19+ subsequent hospitalization and intensive care. DESIGN, SETTING, AND PARTICIPANTS: Retrospective cohort study including all patients tested for Covid-19 between February 8 and March 30, 2020, inclusive. We extracted electronic health record data from the national Veterans Affairs Healthcare System, the largest integrated healthcare system in the United States, on 2,026,227 patients born between 1945 and 1965 and active in care. Exposures: Demographic data, comorbidities, medication history, substance use, vital signs, and laboratory measures. Laboratory tests were analyzed first individually and then grouped into a validated summary measure of physiologic injury (VACS Index). Main Outcomes and Measures: We evaluated which factors were associated with Covid-19+ among all who tested. Among Covid-19+ we identified factors associated with hospitalization or intensive care. We identified independent associations using multivariable and conditional multivariable logistic regression with multiple imputation of missing values. RESULTS: Among Veterans aged 54-75 years, 585/3,789 (15.4%) tested Covid-19+. In adjusted analysis (C-statistic=0.806) black race was associated with Covid-19+ (OR 4.68, 95% CI 3.79-5.78) and the association remained in analyses conditional on site (OR 2.56, 95% CI 1.89-3.46). In adjusted models, laboratory abnormalities (especially fibrosis-4 score [FIB-4] >3.25 OR 8.73, 95% CI 4.11-18.56), and VACS Index (per 5-point increase OR 1.62, 95% CI 1.43-1.84) were strongly associated with hospitalization. Associations were similar for intensive care. Although significant in unadjusted analyses, associations with comorbid conditions and medications were substantially reduced and, in most cases, no longer significant after adjustment. CONCLUSIONS AND RELEVANCE: Black race was strongly associated with Covid-19+, but not with hospitalization or intensive care. Among Covid-19+, risk of hospitalization and intensive care may be better characterized by laboratory measures and vital signs than by comorbid conditions or prior medication exposure.

Early initiation of prophylactic anticoagulation for prevention of COVID-19 mortality: a nationwide cohort study of hospitalized patients in the United States.

IMPORTANCE: Deaths among patients with coronavirus disease 2019 (COVID-19) are partially attributed to venous thromboembolism and arterial thromboses. Anticoagulants prevent thrombosis formation, possess anti-inflammatory and anti-viral properties, and may be particularly effective for treating patients with COVID-19. OBJECTIVE: To evaluate whether initiation of prophylactic anticoagulation within 24 hours of admission is associated with decreased risk of death among patients hospitalized with COVID-19. DESIGN: Observational cohort study. SETTING: Nationwide cohort of patients receiving care in the Department of Veterans Affairs, the largest integrated healthcare system in the United States. PARTICIPANTS: All patients hospitalized with laboratory-confirmed SARS-CoV-2 infection March 1 to July 31, 2020, without a history of therapeutic anticoagulation. EXPOSURES: Prophylactic doses of subcutaneous heparin, low-molecular-weight heparin, or direct oral anticoagulants. MAIN OUTCOMES AND MEASURES: 30-day mortality. Secondary outcomes: inpatient mortality and initiating therapeutic anticoagulation. RESULTS: Of 4,297 patients hospitalized with COVID-19, 3,627 (84.4%) received prophylactic anticoagulation within 24 hours of admission. More than 99% (n=3,600) received subcutaneous heparin or enoxaparin. We observed 622 deaths within 30 days of admission, 513 among those who received prophylactic anticoagulation. Most deaths (510/622, 82%) occurred during hospitalization. In inverse probability of treatment weighted analyses, cumulative adjusted incidence of mortality at 30 days was 14.3% (95% CI 13.1-15.5) among those receiving prophylactic anticoagulation and 18.7% (95% CI 15.1-22.9) among those who did not. Compared to patients who did not receive prophylactic anticoagulation, those who did had a 27% decreased risk for 30-day mortality (HR 0.73, 95% CI 0.66-0.81). Similar associations were found for inpatient mortality and initiating therapeutic anticoagulation. Quantitative bias analysis demonstrated that results were robust to unmeasured confounding (e-value lower 95% CI 1.77). Results persisted in a number of sensitivity analyses. CONCLUSIONS AND RELEVANCE: Early initiation of prophylactic anticoagulation among patients hospitalized with COVID-19 was associated with a decreased risk of mortality. These findings provide strong real-world evidence to support guidelines recommending the use of prophylactic anticoagulation as initial therapy for COVID-19 patients upon hospital admission.