Comparison of Time to Clinical Improvement With vs Without Remdesivir Treatment in Hospitalized Patients With COVID-19

Key Points Question Does time to clinical improvement or time to death differ among hospitalized patients treated with vs without remdesivir (alone or with corticosteroids) for coronavirus disease 2019 outside of a clinical trial? Findings In this multicenter comparative effectiveness research study that included 2483 consecutive admissions with a high proportion of non-White individuals, treatment with remdesivir was associated with more rapid clinical improvement than no remdesivir receipt in propensity score–matched controls. The addition of corticosteroids to remdesivir was not associated with improved time to death. Meaning Remdesivir administration is associated with more rapid clinical improvement than no remdesivir receipt among patients with coronavirus disease 2019.

This supplementary material has been provided by the authors to give readers additional information about their work.

Time-dependent Propensity Score Matching
Since the timing of initial remdesivir administration was different and the remdesivir assignment was nonrandomized, a time-dependent propensity score matching method was applied to undertake 1:1 propensity score matching for the treatment arm so that additional analyses could be performed on the matched patients. Propensity scores were calculated from a time-dependent Cox proportional hazards regression model using the time to the first receipt of remdesivir as the outcome, where the propensity score at a given hospitalization day is the hazard of exposure to remdesivir treatment at that day. The PH assumption was checked using the Schoenfeld residuals against the transformed time. [1][2] Time-dependent covariates in records before the first remdesivir administration date (for treatment group patients) and records before right-censoring or last follow-up date (for control group patients) were used as predictors to obtain parameter estimates. Time-invariant (fixed) covariates included hospital admission race, age, sex, body mass index (BMI), Charlson comorbidity index (CCI), and code status (i.e. whether the patient had a "do not resuscitate" order). Time-dependent (varying) covariates included various clinical measures of disease severity, such as the ratio of measured oxygen saturation to fraction of inspired oxygen (SpO2/FiO2), systolic blood pressure (SBP), diastolic blood pressure (DBP), pulse, temperature, respiratory rate, and supplemental oxygen device (O2 device). SpO2/FiO2 was selected since it can be calculated without obtaining an arterial blood gas (not available on all patients). Other time-dependent variables included key laboratory test results, such as C-reactive protein (CRP), absolute lymphocyte count (ALC), platelets count, white blood cell count (WBC), hemoglobin (HGB), albumin, alanine aminotransferase (ALT), estimated Glomerular Filtration Rate (eGFR), and D-dimer.
Beginning from day zero, a sequential 1:1 greedy matching without replacement was conducted. Patients were included in the matching process only if their admission dates were later than the earliest admission date (April 27, 2020) of patients in the remdesivir group. A patient who first received remdesivir at a given day t of hospitalization was matched with those who did not, based on their propensity scores (hazard components) at day t. In addition, a time constraint was imposed so that a patient in the remdesivir group with k days of treatment, was forced to match a patient in the control group who stayed at least k days (5 days maximum) in the hospital since the matched day. This constraint removed patients from the control group who were healthy enough to be discharged in one or two days from the matched day as those patients were unlikely to receive remdesivir treatment at the matched day if they were close to discharge.

Marginal Structural Cox Regression Model
In sub-analyses considering the effects of combination drug use, comparisons of patients who used combination of corticosteroids and remdesivir (n=185) with patients who used remdesivir alone (n=158) were conducted. Since the sample sizes for two groups were similar, and patients' exposure to corticosteroids was timevariant, instead of time-dependent propensity score matching, Marginal Structural Cox regression models were conducted to adjust for the non-randomized administration of corticosteroids, in the meantime, to analyze the effects of corticosteroids on outcomes of interests in patients who had exposure to remdesivir. 3 The same set of timedependent covariates and time-invariant variables as in the matching models from all remdesivir patients were included in the model. The Inverse Probability Treatment Weighting (IPTW) method was applied for parameter estimation.

Outcome of Interest Analyses
The primary outcome of interest was time to clinical improvement from the treatment start of remdesivir, defined as discharge alive from the hospital without worsening of their WHO disease severity score or at least a twopoint decrease in the WHO severity score during hospitalization within 28 days or max follow-up after the first treatment of remdesivir. Failure of clinical improvement was censored at max follow-up day or 28 days, whichever came first. Death was also censored at 28 days. The secondary outcome was time to death from the first remdesivir treatment day. Patients who were discharged alive were censored at 28 days.