Effectiveness of casirivimab and imdevimab, and sotrovimab during Delta variant surge: a prospective cohort study and comparative effectiveness randomized trial

Study setting and data sources

UPMC is a 40-hospital integrated health system principally in central and western Pennsylvania. After an EUA was granted for bamlanivimab on November 9, 2020, ⁸ UPMC developed a mAb treatment infrastructure.^{9, 10} On November 21, 2020, the UPMC Pharmacy and Therapeutics Committee wrote a therapeutic interchange policy in response to the issuance of an EUA for casirivimab and imdevimab and unpredictable mAb supply. The policy considered all available mAbs equivalent; a patient could receive any mAb based on local inventory. The policy was updated to include sotrovimab on July 9, 2021. All pharmacies supplying all infusion sites had equal opportunity to order any available mAb from a central supply facility. Prescribers were required to provide and review all mAb EUA Fact Sheets with the patient at time of referral and explain the patient could receive any EUA-governed mAb under the therapeutic interchange policy. EUA-eligible patients were referred via the electronic medical record (EMR) systems for UPMC providers and by paper order for non-UPMC prescribers. A centralized team with nurses, administrators, pharmacists, and physicians reviewed orders daily to confirm criteria for receipt. Decentralized nursing teams then contacted and scheduled eligible patients for infusions.

We used the EMR to access all key clinical data including detailed sociodemographic and medical history data, and hospitalization after mAb treatment, and augmented by manual review and data collection. We linked the deidentified primary data sources using common variables within the UPMC data systems aggregated in its Clinical Data Warehouse. We conducted Social Security Administration Death Master File queries for vital status.¹¹

Exposure

Both mAb were provided as per their EUAs and were provided intravenously for all patients in this report. Patients who received casirivimab and imdevimab via subcutaneous injection were excluded from the analysis.

Propensity-matched observational study of mAb treatment versus no mAb treatment

Statistical analysis

We performed three analyses. We used propensity matching to compared mAb-eligible patients to, i) all mAb treated patients, ii.) casirivimab and imdevimab treated patients, and iii.) sotrovimab treated patients, respectively. Propensity scores were derived from multivariable logistic regression models fit from variables with mAb treatment as the response variable and forward stepwise selection of pre-treatment explanatory variables at p < 0.15. We included variables deemed biologically relevant (e.g., age) prior to stepwise selection. We used a matching ratio (treated/untreated) of 1:2 with a maximum propensity score probability difference of 0.01. We compared characteristics of treated versus nontreated patients using Student t tests for continuous variables and χ 2 tests for categorical variables, prior to and after matching by propensity score. The primary outcome analysis consisted of the 1:2 matched comparisons of treated patients to non-treated control patients, with study outcomes expressed as an effect estimate of treatment (risk ratio, 95% confidence interval). Because of the similarity of casirivimab and imdevimab and sotrovimab treated patients (including nearly identical distributions of propensity scores), and to insure the same reference group (untreated patients) for comparison, the full set of matched controls was used in all analyses.

Comparative effectiveness trial of casirivimab and imdevimab versus sotrovimab

The Bayesian adaptive platform trial provided the mAb therapeutic interchange via random allocation. The UPMC Quality Improvement Committee approved the study, including the random therapeutic interchange. The University of Pittsburgh IRB considered the randomized therapeutic interchange to be quality improvement and approved the additional data collection and analyses. A custom application built into the EMR linked local inventory to patient encounters and provided a random mAb allocation at time of mAb referral. Patients provided verbal consent to receive mAb therapy as part of routine care within the EUA. The prescriber and/or patient could request a specific mAb if desired.

Patient cohort

The primary analysis population was the “as-infused” population of all patients allocated to sotrovimab or casirivimab and imdevimab, and who received mAb treatment at UPMC infusion centers and EDs during the study period. As all arms included mAb, there was no anticipated relationship between lack of infusion and the assigned arm. Due to episodic mAb shortages, some patients were treated at sites with only one available mAb at the time of treatment. We included these patients in the primary analysis as prescribing physicians and patients were blinded to drug availability at time of randomization and patients could be treated at a site different than the randomization site, pending scheduling availability.

Outcomes

The primary outcome was hospital-free days up to day 28 after mAb treatment. This outcome is an ordinal endpoint with death as the worst outcome (labeled -1), then the length of time alive and free of hospitalization, such that the best outcome would be 28 hospital-free days. If a patient had intervening days free of hospital and was then re-hospitalized, the patient was given credit for the intervening days as free of the hospital. The secondary outcome was mortality at 28 days. We evaluated outcomes stratified by infusion location (ED versus infusion center), and the frequency of adverse events. We assessed SARS-CoV-2 variant prevalence in a random subset of enrolled patients and in our Pennsylvania catchment using Global Initiative on Sharing All Influenza Data (GISAID).⁷

Statistical Analysis

The trial statistical analysis plan was written by blinded investigators prior to data lock and analysis. The platform is designed to continuously evaluate multiple mAb, with randomization continuing until pre-determined statistical thresholds are met. The trial launched with equal allocation randomization and planned interim analyses for adaptive randomization where mAb performing better would be given higher randomization probabilities. The mAb arm at the first adaptive analysis with the largest sample size was specified as the referent arm, as there was no non-mAb control and all patients received active treatment. Methods and results are reported as per the CONSORT Pragmatic Extension checklist (Supplement 2).¹² An unblinded statistical analysis committee conducted interim and final analyses with R version 4.0.5 using the RStan package version 2.21.0 (R Foundation, Vienna, Austria) and reported results to the UPMC Chief Medical Officer who functioned in a data and safety monitor role for the study.

The primary model was a Bayesian cumulative logistic model that adjusted for treatment location (infusion center or ED), age (<30, 30-39, 40-49, 50-59, 60-69, 70-79, and ≥80 years), sex, and time (2-week epochs). The comparison between individual mAbs were based on the relative odds ratio between a given two arms for the ordinal primary outcome. An odds ratio for an arm to a comparator greater than 1 implies improved outcomes on the ordinal scale. A sliding scale with different levels of equivalence bounds was pre-defined. Equivalence between two arms was defined as a 95% posterior probability that the odds ratio is within a given odds ratio bound. Inferiority of one arm compared to another was defined as a 99% posterior probability of an odds ratio less than 1.

Secondary analyses

We conducted a sensitivity analysis that excluded patients who received a mAb infusion at a site with only one available mAb treatment option in their “as infused” group. We conducted a priori subgroup analyses by vaccine status (full, partial, unvaccinated, unknown), symptom onset (>5 days, <5 days), and location (infusion center, ED).

Decision to publish interim results

Due to the Delta variant public health crisis, we unblinded and analyzed patients allocated through September 29, 2021, prior to having knowingly met any pre-specified statistical threshold.

Propensity-matched observational study of mAb treatment versus no mAb treatment

The mean age (SD) of the 1,028 treated patients was 53 (16) years versus 49 (21) years for the 5,171 patients who did not receive mAb (p < .001). Treated patients were less likely than no mAb patients to be of Black race, had lower mean Charlson Comorbidity Index scores, and more likely to have greater body mass index. After propensity score matching, treated and non-mAb treated patients were similar for all variables included in the propensity score model, the distribution of propensity scores, and variables not included in the model (eFigure 2, eTable 1).

Of the 1,023 propensity-matched patients who received mAb at an infusion center, 35 (3%) were hospitalized or died, and of the 2,046 propensity-matched patients who did not receive mAb, 174 (8.5%) were hospitalized or died. The mortality rate was 2.9% (n=60). In the propensity matched analysis, patients receiving mAb had lower adjusted risk of hospitalization or death (risk ratio 0.40, 95% CI: 0.28– 0.57), compared to no mAb treatment. Of the 712 propensity-matched patients who received casirivimab and imdevimab, 19 (2.7%) were hospitalized and one died (0.1%), with lower adjusted risk of hospitalization or death (risk ratio 0.31, 95% CI: 0.20–0.50), compared to no mAb treatment. Of the 311 propensity-matched patients who received sotrovimab, 16 (5.1%) were hospitalized and none died, yielding lower adjusted risk of hospitalization or death (risk ratio 0.60, 95% CI: 0.37–1.00), compared to no mAb treatment. Unmatched analyses had similar results (eTable 2).

Comparative effectiveness trial of casirivimab and imdevimab versus sotrovimab

Among 4,530 referred and allocated patients, 3,558 (79%) were infused (casirivimab and imdevimab, n=2454, sotrovimab, n=1104, Figure 1). The mean age across groups was 54 years, half were female (54%), 12% were Black, and the most common risk factors were body mass index > 25, age > 65 years, and hypertension. Mean duration of symptom onset to infusion was 5.9 days. Of those allocated to a mAb but excluded (N=972), most were not infused due to declining treatment or inability to contact (N=437, 45%), undetermined reasons (N=183, 19%), or were scheduled but not infused (N=116, 12%). One patient requested a specific mAb different than randomized assignment. All patients (N=79) tested for variant type had Delta, consistent with Pennsylvania GSAID data for this time period which showed nearly all cases were Delta (eFigure 1). Baseline characteristics were similar across groups (Table 2).

Primary outcome

The median hospital-free days were 28 (IQR 28–28) (Table 3, Figure 2). Relative to the casirivimab and imdevimab group, the posterior median adjusted odds ratios from the primary model was 0.88 (95% credible interval, 0.70–1.11) for the sotrovimab group. This odds ratio resulted in an 86% probability of inferiority for sotrovimab versus casirivimab and imdevimab. The probability of equivalence between sotrovimab and casirivimab and imdevimab at the first pre-specified bound was 79%. No comparison met prespecified criteria for statistical inferiority or equivalence.

• Download figure
• Open in new tab

Figure 1. CONSORT Diagram

• Download figure
• Open in new tab

Figure 2. Hospital-Free Days to Day 28 in comparative effectiveness trial

Primary outcome is displayed as horizontally stacked proportions by monoclonal antibody type. Red represents worse values and blue represents better values. The median adjusted odds ratio from the primary analysis, using a Bayesian cumulative logistic model, was 0.88 (95% credible interval, 0.70–1.11) for sotrovimab versus casirivimab and imdevimab. This odds ratio yielded 86% probability of inferiority for sotrovimab versus casirivimab and imdevimab, and 79% probability of equivalence between the two at the first pre-specified bound.

Secondary outcomes

The 28-day mortality rates were 0.5% (12/2,454) and 0.6% (7/1,104), and hospitalization rates were 11.9% (291/2,454) and 12.7% (140/1,104), in the casirivimab and imdevimab, and sotrovimab groups, respectively (Table 3). For patients receiving mAb in an ED, rates of hospitalization after treatment were 18.7% (262/1,399) for casirivimab and imdevimab and 17.4% (113/648) for sotrovimab. For patients receiving mAb in an infusion center, rates of hospitalization after treatment were 2.7% (29 of 1,055) for casirivimab and imdevimab and 5.9% (27 of 456) for sotrovimab.

Sensitivity and subgroup analyses

We performed multiple sensitivity analyses to determine the robustness of our results. First, we excluded patients (n=1,162) who received casirivimab and imdevimab at a location where only casirivimab and imdevimab was available and found similar results (81% probability of sotrovimab inferiority to casirivimab and imdevimab, 79% equivalence). No subgroup analysis except location met the pre-specified thresholds for statistical inferiority or equivalence. Among the 1,511 patients treated at an infusion center, the median odds ratio of sotrovimab compared to casirivimab and imdevimab was 0.48 (95% credible interval, 0.28–0.84), resulting in a 99.5% probability of inferiority.

Adverse events

Among the 2,454 patients treated with casirivimab and imdevimab, adverse events (AE) and serious adverse events (SAE) were rare, reported in 17 (0.5%) and 7 (0.2%) of patients, respectively. Similarly, for the 1,104 sotrovimab treated patients, respective rates were 0.5% (6 patients) and 0.4% (4 patients). The most commonly reported AE were flushing, itching, breathing difficulties, and chest tightness/pain.