Ischemic Stroke Occurs Less Frequently in Patients With COVID-19

A Multicenter Cross-Sectional Study

Kimon Bekelis, MD; Symeon Missios, MD; Javaad Ahmad, MD; Nicos Labropoulos, PhD; Clemens M. Schirmer, MD, PhD; Daniel R. Calnan, MD, PhD; Jonathan Skinner, PhD; Todd A. MacKenzie, PhD

Disclosures

Stroke. 2020;51(12):3570-3576. 

In This Article

Methods

Data and analytic methods will be made available by contacting the corresponding author.

Study Design

We conducted a cross-sectional study of the association of ischemic stroke and COVID-19 infection using hospital discharges from 6 hospitals.

Patient Population

This study was approved by the Institutional Review Board of Catholic Health Services of Long Island. The need for informed consent was waived by the Institutional Review Board. All patients discharged from any of the six hospitals of Catholic Health Services of Long Island in Suffolk (Good Samaritan Hospital Medical Center, Saint Joseph's Hospital, Saint Catherine's of Siena Hospital, Saint Charles Hospital) and Nassau County, New York (Saint Francis Hospital, Mercy Medical Center) between January and April 2020 were included in the analysis. We used discharge data generated through our electronic medical record system. These hospitals are comprised of a Comprehensive Stroke Center and 5 Advanced Primary Stroke Centers (one with thrombectomy capabilities), all certified by the NY Department of Health and The Joint Commission. More information about Catholic Health Services of Long Island is available at https://www.chsli.org.

Outcome Variables

We used International Classification of Disease, Tenth Revision (ICD-10) codes to identify outcomes in the database. The primary outcome variable was the occurrence of new-onset stroke (ICD-10 code I63.xx, G45.xx) in our cross-section of patients. The diagnosis and coding of stroke was limited to patients who presented with symptoms of an acute ischemic stroke and were confirmed to have acute ischemic change on magnetic resonance imaging. Patients with negative magnetic resonance imaging were not coded as having a stroke. Secondary outcomes were case-fatality and discharge to rehabilitation for patients presenting with acute ischemic stroke.

Exposure Variables

The primary exposure variable was infection with SARS-CoV-2, resulting in COVID-19 (ICD-10 code U07.1). All patients of Catholic Health Systems of Long Island received a nasopharyngeal swab using the COVID-19 RNA polymerase chain reaction testing method on arrival to emergency department during the dates of this study.

Covariates used for risk adjustment were age, sex, race (Black, Asian, White, other), insurance status (Medicare, Medicaid, private, self pay, other), mechanical thrombectomy (ICD-10 code 03CG3ZZ), and administration of IV tPA (intravenous tissue-type plasminogen activator; ICD-10 code Z92.82).

The comorbidities used for risk adjustment were diabetes, smoking, chronic obstructive lung disease, hypertension, hypercholesterolemia, congestive heart failure, coronary artery disease, alcohol abuse, peripheral vascular disease, and chronic renal failure. Only variables that were defined as present on admission were considered part of the patient's preadmission comorbidity profile (See Variable Definitions in the Data Supplement).

Statistical Analysis

We examined the association of COVID-19 with our primary outcome (acute ischemic stroke) using crude and 3 methods of adjustment for confounders. Our primary analysis was based on a logistic regression model controlling for age, sex, race, insurance status, and all the comorbidities mentioned previously, with hospital fixed effects to control for clustering at the hospital level. To demonstrate the robustness of our data in sensitivity analysis, we used additional techniques to account for measured confounding while accounting for clustering at the hospital level. These were comprised of inverse-weighted propensity estimation of the absolute difference and odds ratio (OR) and binned propensity scores. Inverse-weighted propensities were trimmed at their 99% percentile. The inverse-weighted propensity approach would yield estimates of the Average Treatment Effect, for example, comparison of the counterfactuals if everyone versus no one had COVID-19. The results of the inverse-weighted propensity model are only presented in the tables. The propensity score bins were utilized as follows; we controlled for the 100 level factor created from percentiles of the propensity score in a logistic regression with COVID-19 and this factor. The propensity scores were derived using a logistic regression of COVID-19 as a function of the covariates mentioned above and all their pairwise interactions. To further test the sensitivity of our results, we repeated the above analyses separately for ischemic stroke (ICD-10 code I63.xx) and transient ischemic attack (ICD-10 code G45.xx). The consistency of our results was also examined by considering acute myocardial infarction (MI), a disease closely correlated with ischemic stroke, as an outcome (ICD-10 code I21.9). Observing a similar trend in cardiovascular disease would validate our analysis.

For our secondary outcomes, given the limited df (relatively small number of observations), we used a step-wise logistic regression model controlling for age, sex, hypertension, hypercholesterolemia, diabetes, and peripheral vascular disease. In a sensitivity analysis, we repeated the models for our secondary outcomes forcing IV tPA administration and mechanical thrombectomy as covariates in our regression models to account for more severe strokes. The direction of the observed associations did not change and is therefore not reported any further.

All results are based on 2-sided tests, and the level of statistical significance was set at 0.05. Post hoc power calculation: This study, based on 24 808 patients, has sufficient power (80%) at a 5% type I error rate to detect an increase in stroke frequency from 2% to 2.9% of patients (or a decrease from 2% to 1.2%). Statistical analyses were performed using R version 3.3.1 (R Foundation for Statistical Computing).

processing....