Healthcare disparities among anticoagulation therapies for severe COVID‐19 patients in the multi‐site VIRUS registry

Abstract Here we analyze hospitalized andintensive care unit coronavirus disease 2019 (COVID‐19) patient outcomes from the international VIRUS registry (https://clinicaltrials.gov/ct2/show/NCT04323787). We find that COVID‐19 patients administered unfractionated heparin but not enoxaparin have a higher mortality‐rate (390 of 1012 = 39%) compared to patients administered enoxaparin but not unfractionated heparin (270 of 1939 = 14%), presenting a risk ratio of 2.79 (95% confidence interval [CI]: [2.42, 3.16]; p = 4.45e−52). This difference persists even after balancing on a number of covariates including demographics, comorbidities, admission diagnoses, and method of oxygenation, with an increased mortality rate on discharge from the hospital of 37% (268 of 733) for unfractionated heparin versus 22% (154 of 711) for enoxaparin, presenting a risk ratio of 1.69 (95% CI: [1.42, 2.00]; p = 1.5e−8). In these balanced cohorts, a number of complications occurred at an elevated rate for patients administered unfractionated heparin compared to patients administered enoxaparin, including acute kidney injury, acute cardiac injury, septic shock, and anemia. Furthermore, a higher percentage of Black/African American COVID patients (414 of 1294 [32%]) were noted to receive unfractionated heparin compared to White/Caucasian COVID patients (671 of 2644 [25%]), risk ratio 1.26 (95% CI: [1.14, 1.40]; p = 7.5e−5). After balancing upon available clinical covariates, this difference in anticoagulant use remained statistically significant (311 of 1047 [30%] for Black/African American vs. 263 of 1047 [25%] for White/Caucasian, p = .02, risk ratio 1.18; 95% CI: [1.03, 1.36]). While retrospective studies cannot suggest any causality, these findings motivate the need for follow‐up prospective research into the observed racial disparity in anticoagulant use and outcomes for severe COVID‐19 patients.

patients. While the availability of anticoagulant dosing information in the SCCM registry is relatively sparse, we are able to examine differential patient outcomes associated with whether a patient has or has not received a specific anticoagulant medication. We consider three categories of anticoagulant medications: (1) Unfractionated Heparin, (2) Enoxaparin, and (3) Other types of low molecular weight heparin (LMWH). First, we consider head-to-head comparisons of enoxaparin versus unfractionated heparin and enoxaparin versus other types of LMWH by constructing cohorts of hospitalized COVID patients who received one anticoagulant medication but not the other during their hospital stay for COVID-19. For each cohort comparison, we evaluate patient outcomes including: mortality at hospital discharge, 28-day mortality status, average hospital length of stay in days, average intensive care unit (ICU) length of stay in days, and complications during the 28-day follow-up period. In addition, for each comparison we repeat the analysis using propensity score matching to control for potential confounding variables including: demographics, comorbidities, evidence of infiltrates, ICU admission status, initial oxygenation method, admission diagnoses, and time in days to anticoagulant administration. Finally, we analyzed the rates of anticoagulant medication administration by race, focusing on cohorts of Black/African American and White/ Caucasian patients. Similar, we used propensity score matching to construct race-based cohorts balanced on the clinical covariates listed previously, and we report patient outcomes for both the original and the propensity-matched race-stratified cohorts. Prior studies suggest that enoxaparin may be more efficacious than unfractionated heparin in the treatment of conditions like acute coronary syndromes 6 and these are two most frequently administered anticoagulants (Table S1). Thus, we compare the outcomes of patients taking enoxaparin and heparin by constructing two cohorts: (i) patients who were administered enoxaparin but not unfractionated heparin and (ii) patients who were administered unfractionated heparin but not enoxaparin. The cohort sizes were 1814 and 887, respectively. Statistical tests were applied to 21 outcomes (with Benjamini-Hochberg procedure applied to account for the problem of multiple comparisons; details below). Mortality at hospital discharge was the primary outcome of interest. Outcomes that were compared include (1) mortality at hospital discharge, (2) mortality at 28 days, (3) admission to ICU (within 28 days of hospitalization), (4) length of stay in ICU (among alive patients), (5) length of stay in hospital (among alive patients), and the following anemia, (9) bacteremia, (10) bacterial pneumonia, (11) cardiac arrest, (12) cardiac arrhythmia, (13) co-or secondary infection, (14) congestive heart failure, (15) Table S2.
To account for potentially confounding variables, we performed propensity score matching to balance covariates between the two cohorts. The statistical tests for differences in outcomes were repeated on the matched cohorts. The covariates which were balanced include demographics, comorbidities, and various features on admission. Further detail on the procedure, including a listing of covariates used, is below. The code to process the raw data files was written in R v3.6.1. The code to perform the statistical analyses was written in Python v3.7.7, using the scikit-learn package v0.23.2 to train the logistic regression models for the propensity score matching step. In Table S3, we show the data completeness for the clinical covariates that we used for matching. Most covariates have close to full completeness (over 90%), with the exception of the "evidence of infiltrates" covariate, which has roughly 80% data completeness. For this field, missing values were imputed to be the mean of other values of the field within the treatment group.

| Statistical methods
For each of the cohort comparisons, we ran a series of statistical significance tests to compare across each of the patient outcome variables of interest. For categorical outcome variables (e.g. mortality status, complications), we report the proportion of patients in each cohort that have the outcome variable, the relative risk (ratio of proportions for each cohort), 95% confidence interval (CI) for the relative risk, and χ 2 p-value. The function stats. chi2_contingency from the SciPy package in Python was used to compute the χ 2 p-values. For continuous outcome variables (e.g., hospital/ICU length of stay), we report the mean and standard deviation of the variable in each cohort, along with the p value from a two-sided Mann-Whitney test (stats. mannwhitneyu from SciPy) comparing the two cohorts.
Finally, we apply the Benjamini-Hochberg correction to adjust p values for multiple comparisons.

| Propensity score matching
To control for potential confounding factors which may explain differences in patient outcomes between the enoxaparin and unfractionated heparin cohorts, we used propensity score matching to balance the cohorts. 7 First, propensity scores for each of the patients in the two cohorts were computed by fitting a logistic regression model as a function of the clinical covariates (listed below). Next, patients from the enoxaparin and unfractionated heparin cohorts were matched using a 1:1 matching ratio and a heuristic caliper of 0.1 x pooled standard deviation, 8 allowing for drops. Before matching, there were 2120 patients in the enoxaparin cohort (administered enoxaparin but not unfractionated heparin), and there were 1076 patients in the unfractionated heparin cohort (administered unfractionated heparin but not enoxaparin). From these two cohorts, 778 matched pairs were found, and statistical analyses were run on the final matched cohorts. Here is the full list of covariates that were considered for the propensity score matching step: • Demographics: age, gender, race, ethnicity.
• Comorbidities: pre-existing conditions, including: (1) asthma, (2) blood loss anemia, (3) cardiac arrhythmias, (4) chronic kidney disease, (5) chronic dialysis, (6) chronic pulmonary disease, (7) coagulopathy, (8) congestive heart failure, (9) coronary artery disease, (10) dementia, (11) depression, (12) diabetes, (13) dyslipidemia/hyperlipidemia, (14) HIV/AIDS or other immunosuppression, (15) hematologic malignancy, • Evidence of infiltrates via X-ray or CT scan  The same propensity score matching procedure was done with enoxaparin versus other low molecular weight heparin in place of enoxaparin versus unfractionated heparin. Propensity score matching was also applied to balance covariates between the Black/African American and White/Caucasian patient cohorts; the "outcome" compared in this case was the rate of administration of each anticoagulant. All of the same covariates (except race and day of anticoagulant administration) listed above were used in this balancing.

| RESULTS
In Figure 1, we present the mortality rate and ICU admission rate for patients in the SCCM VIRUS registry 5 with outcomes data available.  Figure 1B).
Next, we compared the average lengths of stay in the ICU and hospital for the two cohorts. Here, we restricted the analysis to only patients that were alive at discharge. Among patients with length of stay information available, the average length of stay in the hospital was shorter for the enoxaparin patients (mean hospital duration: 10.99 days; 1350 patients) compared to the unfractionated heparin patients (mean hospital duration: 13.33 days; 676 patients) ( Figure 1C). For patients admitted to the ICU, the length of stay in the ICU was also shorter for enoxaparin patients (mean ICU duration: 10.70 days; 647 patients) compared to unfractured heparin patients (mean ICU duration: 12.16 days; 446 patients) ( Figure 1D).
While the difference on average hospital length of stay is statistically significant (Mann-Whitney p = 7.9e−5), the difference on average ICU length of stay is also statistically significant (Mann-Whitney In Figure 2, we present the mortality rate and ICU admission rate for patients with different comorbidities: diabetes, hypertension, chronic kidney disease, and congestive heart failure. We observe that for the subgroups of patients with diabetes, hypertension, and congestive heart failure, patients administered enoxaparin have significantly lower rates of ICU admission and death compared to patients administered unfractionated heparin. For patients with chronic kidney disease, the difference in ICU admission rates between the unfractionated heparin and enoxaparin cohorts is statistically significant (risk ratio: 1.4; 95% CI: [1.14, 1.7]; p = 5.88e−04), however, the difference in mortality status is not statistically significant (risk ratio: 1.23; 95% CI: [0.92, 1.67], p = .18).
Next, we perform propensity score matching to control for a wide array of confounding factors simultaneously. The clinical characteristics of the matched and original unfractionated heparin and enoxaparin cohorts are shown in Table 1. Most covariates (including demographics, comorbidities, and admission diagnoses) appear wellmatched.
Of the 778 patients in the matched heparin cohort, mortality status at discharge was available for 733, of which 268 (37%) were deceased on discharge; in the matched enoxaparin cohort, information was available for 711 patients of which 154 (22%) were deceased on discharge (Table 2). This difference in mortality rates upon discharge was statistically significant (risk ratio: 1.69; 95% CI: [1.42, 2.00]; adjusted p = 1.5e−8). The mortality rates reported at 28-days for both cohorts were consistent with the mortality rates reported upon hospital discharge, and differences in rates between the two cohorts were similarly statistically significant. Differences between the two cohorts in the average hospital and ICU length of stays were not statistically significant after matching.   Table 5. We observe that the clinical covariates are well-balanced for F I G U R E 2 Comparison of outcomes between unfractionated heparin and enoxaparin patient cohorts in patients also reporting comorbidities. Bar charts show a comparison of Mortality Status at discharge from the hospital and status of admission to the ICU for two cohorts-patients receiving enoxaparin and reporting a comorbidity of interest (blue), and patients receiving heparin and reporting a comorbidity of interest (orange). Comorbidities include-hypertension, diabetes, chronic kidney disease and congestive heart failure. Statistics for these plots are included in the corresponding tables. CI, confidence interval; ICU, intemsive care unit

| DISCUSSION
Prior work has shown that anticoagulant treatments and prophylaxis are associated with improved outcomes for COVID-19 patients. 9,10 In particular, there is evidence to suggest that low molecular weight heparin can be used to effectively treat COVID-19 patients with coagulopathy. 11 This retrospective analysis suggests that enoxaparin, a particular form of low molecular weight heparin, shows promise as an anticoagulant therapy for severe COVID-19, compared to both unfractionated heparin and other low molecular weight heparin therapies. These findings are consistent with a retrospective study on electronic health records from the Mayo Clinic which has found that enoxaparin is associated with lower rates of thrombotic events, kidney injury, and mortality in comparison with unfractionated heparin. 12 However, this study goes beyond the previous analysis by leveraging the massive SCCM VIRUS data registry of hospitalized investigate differential patient outcomes for other variants of low molecular weight heparin beyond enoxaparin. Similar comparative analyses may be undertaken for other COVID-19 treatment options beyond anticoagulants, such as supplemental oxygenation methods. Important insights may also be gained from studying how varying dosing patterns of anticoagulant administration and indications driving anticoagulant administration relate to differential patient outcomes. In addition, a number of studies have been analyzing the association between race/ethnicity and clinical outcomes in COVID-19. 15,16 The finding from this study that there are race-associated differences in the administration of the anticoagulants enoxaparin and unfractionated heparin warrants further analyses into the associations between patients' race/ethnicity, comorbidities, and administration of medications in managing COVID-19.
Overall, this study demonstrates the utility of the SCCM VIRUS data registry for analyzing diverse research questions related to therapeutics for severe COVID-19 patients. 5

ACKNOWLEDGMENTS
The authors would like to thank the SCCM Discovery VIRUS data registry and the collaborative co-authors listed in Table S4

CONFLICT OF INTERESTS
The authors from nference have financial interests in the company. ADB is a consultant for Abbvie, is on scientific advisory boards for nference and Zentalis, and is founder and President of Splissen therapeutics.

DATA AVAILABILITY STATEMENT
Deidentified data will be made available for research and qualitative studies purposes with appropriate approvals from SCCM after the study completion in December 2022. Reasonable requests may be made to SCCM (discovery@sccm.org) and to the corresponding author (venky@nference.net).

ETHICS STATEMENT
The VIRUS registry was granted exempt status for human subjects research by the institutional review board at the Mayo Clinic (IRB:20-002610). The ClinicalTrials.gov number is NCT04323787 (https://clinicaltrials.gov/ct2/show/NCT04323787). Each study site submitted a proposal to their local review boards for approval and signed a data use agreement before being granted permission to extract and enter deidentified data into the registry.