Patient population and data elements. For this retrospective study, we utilized data from the Columbia University Irving Medical Center (CUIMC) and Allen Hospital sites of the NewYork-Presbyterian Hospital (NYPH). Adult patients (≥18 years of age) who were hospitalized between February 1st through May 12th, 2020, and tested positive for SARS-CoV-2 reverse transcriptase-polymerase chain reaction (RT-PCR) testing of nasopharyngeal or oropharyngeal specimens were included in the present analysis. All testing was conducted either by NYPH laboratories or the New York State Department of Health (in the period of time prior to when internal testing capabilities were available). Patients who were admitted for less than 24 hours were excluded from this analysis. The study period ended on June 11th, 2020, allowing for a follow-up period of at least 30 days in all patients. The CUIMC Institutional Review Board approved this study and waived the requirement for obtaining informed consent. De-identified data will be made available on request to the corresponding authors.
Patient data were identified in the electronic medical record by using the institution’s clinical data warehouse, which includes outpatient and inpatient information on individuals who receive care at our institution. No manual chart abstraction was performed. Follow-up for each patient continued until patients were discharged, died in-hospital, or the end of the study period was reached. Data analysis was limited to the index hospitalization in the event of readmissions. Baseline information including age, gender, race and ethnicity, insurance, New York City borough of residence, body mass index (BMI), comorbidities, and outpatient medications were recorded. Clinical comorbidities, including hypertension, diabetes, coronary artery disease, heart failure, stroke or transient ischemic attack, atrial arrhythmias (atrial fibrillation, atrial flutter and supraventricular tachycardia), chronic lung disease, chronic kidney disease, and chronic liver disease were identified using ICD-10 medical billing codes (Supplemental Table 1). Outpatient medications, including statins, angiotensin-converting enzyme inhibitors (ACEi), angiotensin receptor blockers (ARB), beta-blockers, oral anticoagulants, and P2Y12 inhibitors were extracted from medication reconciliation fields in the electronic medical record, which are entries that are updated at the time of hospital admission.
We included features of the clinical presentation, including vital signs at presentation (i.e. temperature, heart rate, blood pressure, respiratory rate, and peripheral oxygen saturation). Several laboratory parameters at presentation were also collected from the electronic medical record, including white blood cell count, platelet count, creatinine, hepatic panel tests including aspartate aminotransferase (AST), alanine aminotransferase (ALT) and albumin, lactate, high-sensitivity troponin T, D-dimer, ferritin, erythrocyte sedimentation rate (ESR), and C-reactive protein (CRP). Given that statins may act by lowering lipid levels, we collected lipid values for patients from inpatient and outpatient records at any dates after January 1st 2018, and averaged them for each patient over the study period. As some patients did not have all laboratory studies of interest collected as part of clinical care, data are presented for only the patients in whom these were available. Details for missing laboratory values are provided in the Supplemental Table 2.
Study exposure. The exposure in this study was antecedent statin use. Antecedent statin use was defined as record of statins as a home medication in the electronic medical record.
Study outcomes. The principal outcome was in-hospital mortality within 30 days of admission. The secondary outcome was a composite of in-hospital mortality or invasive mechanical ventilation within 30 days of admission. Other outcomes included in-hospital mortality at any time and invasive mechanical ventilation at any time, as some patients had a length of stay longer than 30 days. We also examined hospital length of stay (days), duration of invasive mechanical ventilation (days), renal replacement therapy with continuous veno-venous hemofiltration, and use of vasopressors.
Statistical analysis. We examined differences in socio-demographic, baseline clinical characteristics, and outpatient medications by antecedent statin use. Summary statistics are presented as numbers and percentages for categorical variables and medians and interquartile ranges for continuous variables. Differences between groups were examined using the 2-sided independent t-test and chi-squared test, as appropriate.
To address confounding by indication, we constructed a multivariable logistic regression model to predict the propensity of antecedent statin administration, adjusted for the following variables: age, sex, first BMI assessment, race and ethnicity, insurance, New York City borough of residence, history of hypertension, diabetes, coronary artery disease, heart failure, stroke or transient ischemic attack, atrial arrhythmias, chronic lung disease, chronic kidney disease, and liver disease; outpatient use of beta-blockers, ACEi, ARBs, oral anticoagulants, and P2Y12 receptor inhibitors. Propensity-score matching was implemented with the use of a nearest-neighbor strategy with specification of caliper width equal to 0.1 of the standard deviation of the logit of the propensity score. Descriptive analyses were performed for all baseline variables in the propensity-matched cohort.
For the primary and secondary endpoints, we performed logistic regression on the propensity-matched cohort. In addition, to examine whether the effect estimate remained consistent in the overall cohort, we performed logistic regression with multivariable adjustment on the overall cohort. We adjusted the multivariable models for variables that have been previously studied in association with mortality in COVID-19 including baseline socio-demographic and clinical characteristics and outpatient medications.21,22
Sensitivity Analyses. We performed sensitivity analyses by defining any recent statin use as either antecedent statin or inpatient statin use. Using this definition, we evaluated the association of any recent statin use with the primary endpoint using multivariable logistic regression. We also performed subgroup analyses to assess the association of antecedent statin use with primary endpoint in a subset of patients with history of hypertension, coronary artery disease and stroke, conditions for which statins are usually prescribed.
Missing data. BMI and insurance information were missing in 19% and 15% of the patients, respectively, and multiple imputation with predictive mean matching was utilized to adjust the models for BMI and insurance. We imputed one hundred datasets, fitted the logistic regression models for the primary and secondary endpoints for each imputed dataset, estimated the hazard ratios on each imputed dataset, and then averaged the one hundred estimated values to obtain the pooled estimates. Model estimates and standard errors were calculated with Rubin’s rules11. Race and ethnicity were missing in 30% of the patients and were classified as ‘others/missing’ while adjusting in the models. Lipid levels were available for only 32% of the cohort. As such, we have presented them only at baseline. The remaining variables were missing in fewer than 5% of the study cohort.
P-values <0.05 were considered significant for the analysis. We did not adjust for multiple comparisons as this was an exploratory analysis. All analyses were performed using version 3.5.1 of the R programming language (R Project for Statistical Computing; R Foundation, mice, MatchIt, cobalt, rms packages).