key: cord-0335866-scjag8yx
authors: McCreary, E. K.; Bariola, J. R.; Minnier, T.; Wadas, R. J.; Shovel, J. A.; Albin, D. L.; Marroquin, O. C.; Kip, K. E.; Collins, K.; Schmidhofer, M.; Wisniewski, M. K.; Nace, D. A.; Sullivan, C.; Axe, M.; Meyer, R.; Weissman, A.; Garrard, W.; Peck-Palmer, O. M.; Wells, A.; Bart, R. D.; Yang, A.; Berry, L.; Berry, S.; McGlothin, A.; Crawford, A.; Khadem, T.; Linstrum, K.; Montgomery, S. K.; Ricketts, D.; Kennedy, J. N.; Pidro, C. J.; Haidar, G.; Snyder, G. M.; McVerry, B. J.; Seymour, C. W.; Angus, D. C.; Kip, P. L.; Huang, D. T.
title: A Learning Health System Randomized Trial of Monoclonal Antibodies for Covid-19
date: 2021-09-09
journal: nan
DOI: 10.1101/2021.09.03.21262551
sha: 42a2035244c757a61049cb41540e9c90f07e61d2
doc_id: 335866
cord_uid: scjag8yx

ABSTRACT Background: Neutralizing monoclonal antibodies (mAb) targeting SARS CoV2 decrease hospitalization and death in patients with mild to moderate Covid 19. Yet, their clinical use is limited, and comparative effectiveness is unknown. Methods: We present the first results of an ongoing, learning health system adaptive platform trial to expand mAb treatment to all eligible patients and evaluate the comparative effectiveness of available mAbs. The trial launched March 10, 2021. Results are reported as of June 25, 2021 due to the U.S. federal decision to pause distribution of bamlanivimab etesevimab; patient follow-up concluded on July 23, 2021. Patients referred for mAb who met Emergency Use Authorization criteria were provided a random mAb allocation of bamlanivimab, bamlanivimab etesevimab, or casirivimab imdevimab with a therapeutic interchange policy. The primary outcome was hospital-free days (days alive and free of hospital) within 28 days, where patients who died were assigned -1 day. The primary analysis was a Bayesian cumulative logistic model of all patients treated at an infusion center or emergency department, adjusting for treatment location, age, sex, and time. Inferiority was defined as a 99% posterior probability of an odds ratio < 1. Equivalence was defined as a 95% posterior probability that the odds ratio is within a given bound. Results: Prior to trial launch, 3.1% (502) of 16,345 patients who were potentially eligible by an automated electronic health record (EHR) screen received mAb. During the trial period, 23.2% (1,201) of 5,173 EHR-screen eligible patients were treated, a 7.5-fold increase. After including additional referred patients from outside the health system, a total of 1,935 study patients received mAb therapy (128 bamlanivimab, 885 bamlanivimab etesevimab, 922 casirivimab imdevimab). Mean age ranged from 55 to 57 years, half were female (range, 53% to 54%), and 17% were Black (range, 12% to 19%). Median hospital free days were 28 (IQR, 28 to 28) for each mAb group. Hospitalization varied between groups (bamlanivimab, 12.5%; bamlanivimab etesevimab, 14.7%, casirivimab imdevimab, 14.3%). Relative to casirivimab-imdevimab, the median adjusted odds ratios were 0.58 (95% credible interval (CI), 0.30 to 1.16) and 0.94 (95% CI, 0.72 to 1.24) for the bamlanivimab and bamlanivimab-etesevimab groups, respectively. These odds ratios yielded 91% and 94% probabilities of inferiority of bamlanivimab versus bamlanivimab etesevimab and casirivimab imdevimab respectively, and an 86% probability of equivalence between bamlanivimab etesevimab and casirivimab imdevimab, at the prespecified odds ratio bound of 0.25. Twenty one infusion related adverse events occurred in 0% (0/128), 1.4% (12/885), and 1.0% (9/922) of patients treated with bamlanivimab, bamlanivimab etesevimab, and casirivimab imdevimab, respectively. Conclusion: In non-hospitalized patients with mild to moderate Covid-19, bamlanivimab, compared to bamlanivimab etesevimab and casirivimab imdevimab, resulted in 91% and 94% probabilities of inferiority with regards to odds of improvement in hospital free days within 28 days. There was an 86% probability of equivalence between bamlanivimab etesevimab and casirivimab imdevimab at an odds ratio bound of 0.25. However, the trial was unblinded early due to federal distribution decisions, and no mAb met prespecified criteria for statistical inferiority or equivalence. (ClinicalTrials.gov, NCT04790786).

Neutralizing monoclonal antibodies (mAb) targeting SARS-CoV-2 were granted U.S. Food and Drug Administration (FDA) Emergency Use Authorization (EUA) for treatment of mild to 48 moderate Covid-19, and significantly decrease hospitalization and death in this patient 49 population. [1] [2] [3] [4] [5] [6] [7] [8] [9] However, clinical use is limited due to lack of access, logistical challenges of 50 administration, and incomplete clinician and patient awareness. 10 To evaluate clinical mAb use expansion, we screened for potential eligibility based on EHR 114 identification of all outpatients with a positive SARS-CoV-2 polymerase chain reaction or 115 antigen test performed within the health system and an EUA-defined risk factor for progression 116 to severe disease. Race was derived from registration system data using fixed categories 117 consistent with the Centers for Medicare & Medicaid Services EHR meaningful use dataset and 118 the AMA Manual of Style. 19, 20 Pre-specified categories included non-Hispanic Black, non-119

Hispanic White, and Other. Individuals were considered Other due to small sample sizes for 120

Hispanic, American Indian, and other races and ethnicities. Geographic distribution of mAb 121 treatment was illustrated using zip code of patient residence. 21,22 122 123 To evaluate comparative effectiveness, the primary outcome was hospital-free days up to day 28 124 after mAb treatment. This outcome is an ordinal endpoint with death up to day 28 as the worst 125 outcome (labeled -1), then the length of time alive and free of hospital, such that the best 126 outcome would be 28 hospital-free days. If a patient had intervening days free of hospital and 127 was then re-hospitalized, the patient was given credit for the intervening days as free of the 128 hospital. Secondary outcomes included mortality at 28 days. We evaluated rates of 129 hospitalization by infusion location, and incidence of adverse events (Supplement). We assessed 130 SARS-CoV-2 variant prevalence in our Pennsylvania catchment over time using Global 131 9 133 Data Collection 134 OPTIMISE-C19 was embedded in the EHR to access routine patient care data and was 135 augmented by manual review and data collection. We ascertained the primary outcome of 136 hospital-free days by linking inpatient (Cerner, Kansas City, Missouri) and outpatient (Epic, 137 Madison, Wisconsin) EHRs, as shown in prior work. 18 We conducted patient-directed phone 138 calls at day 28 to ascertain health care encounters outside our health system, and Social Security 139

Administration Death Master File queries for vital status. 24 We collected adverse events in i.) a 140 secure, electronic file sharing application completed by infusion center nurses on the day of mAb 141 treatment, and ii.) an internal patient safety reporting system for adverse reactions completed by 142 nursing and physician staff in infusion centers and EDs. Adverse event severity was adjudicated 143 blinded to mAb type. 144 145

To determine the epidemiology and equitability of mAb infusions in our region, we measured the 147

proportion of EHR-screen eligible patients treated with any mAb stratified by demographics, 148 geography, and prior to or after trial launch. Treatment "prior to the trial" was from December 9, 149 2020 until March 9, 2021, and "during the trial" from March 10 to June 25, 2021. 150

To analyze comparative effectiveness, the trial statistical analysis plan was written by blinded 152 investigators prior to data lock and analysis (Supplement) and applied to patients treated with 153 mAb from March 10, 2021 to June 25, 2021. The platform is designed to continuously evaluate 154 multiple mAb, with randomization continuing until pre-determined statistical thresholds are met. 155

The trial launched with equal allocation randomization and planned interim analyses for adaptive 156 randomization where mAb performing better would be given higher randomization probabilities. 157

The mAb arm at the first adaptive analysis with the largest sample size was specified as the 158 referent arm, as there was no non-mAb control and all patients received active treatment. All 159 mAb were compared to each other. The methods and results are reported in accordance with the 160 CONSORT Pragmatic Extension checklist (Supplement). 25 Officer who functioned in a data and safety monitor role for the study. 164

The primary analysis population was the "as-infused" population which consisted of patients 166 randomly allocated mAb and subsequently treated. As all trial arms included mAb, there was no 167 anticipated relationship between lack of infusion and the assigned arm. Baseline patient 168 characteristics of those infused were compared to those allocated but not infused using mean 169 (SD) or median (IQR) for continuous variables, and proportions for categorical variables. for an arm to a comparator greater than 1 implies improved outcomes on the ordinal scale. A 176 sliding scale of equivalence was set according to odds ratio bounds of 0.25, 0.20, 0.15, 0.10, and 177 0.05. Equivalence between two arms was defined as a 95% posterior probability that the odds 178 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint ratio is within a given odds ratio bound. Inferiority of one arm compared to another was defined 179 as a 99% posterior probability of an odds ratio less than 1. UPMC were EHR-screen eligible, of whom 1,382 were referred and underwent random mAb 205 allocation. 1,084 EUA-eligible patients with a positive SARS-CoV-2 test from outside of UPMC 206 were also referred, yielding a total of 2,466 patients who were assigned a random mAb 207 allocation. Of these, 1,935 (78%) were infused and comprised the primary analysis cohort 208 Eighty-eight patients (17%) received mAb at a location without available EHR data or while 213 hospitalized for non-Covid-19 reasons. No patient or provider requested a specific mAb different 214 than randomized assignment. 215 216 Baseline characteristics were similar across groups ( Table 1) . The mean age for the three groups 217 was 55 -57 years, half were female (range, 53% -54%), 18% were Black (range 12% -19%), 218 and the most common risk factors were advanced age, high body mass index, and hypertension. 219

Of the 241 patients for whom vaccine status was known, 57 (24%) reported being unvaccinated, 220 120 (49%) partly vaccinated, and 64 (27%) fully vaccinated. Mean duration of symptom onset to 221 referral was 5 (2.1) days. 222 223 224 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

The median hospital-free days were 28 (IQR, 28-28) for each mAb group (Table 2, Figure 3) . 226

Relative to the casirivimab-imdevimab group, the posterior median adjusted odds ratios from the 227 primary model were 0.58 (95% credible interval, 0.30 to 1.16) and 0.94 (95% credible interval, 228 0.72 to 1.24) for the bamlanivimab and bamlanivimab-etesevimab groups, respectively. The 229 probabilities of inferiority for bamlanivimab versus bamlanivimab-etesevimab and casirivimab-230 imdevimab were 91% and 94% respectively. The probability of equivalence between 231 bamlanivimab-etesevimab and casirivimab-imdevimab with a bound of 0.25 for the odds ratio 232 was 86%. No comparison met prespecified criteria for statistical inferiority or equivalence. 233 234

The 28-day mortality rates were 0.8% (1/128), 0.8% (7/885), and 0.7% (6/922) and 236 hospitalization rates were 12.5% (16/128), 14.7% (130/885), and 14.3% (132/922), in the 237 bamlanivimab, bamlanivimab-etesevimab, and casirivimab-imdevimab groups, respectively 238 ( Table 2) . For patients receiving mAb in an outpatient infusion center, rates of hospitalization 239 after treatment were 7.8% (bamlanivimab), 6.5% (bamlanivimab-etesevimab), and 6.9% 240 (casirivimab-imdevimab). For patients receiving mAb in an ED, rates of hospitalization after 241 treatment were 32% (bamlanivimab), 23.7% (bamlanivimab-etesevimab), and 21.7% 242

Twenty-one infusion-related adverse events occurred in 0% (0/128), 1.4% (12/885), and 1.0% 246 (9/922) of patients treated with bamlanivimab, bamlanivimab-etesevimab, and casirivimab-247 imdevimab, respectively. Five events (0 bamlanivimab, 1 bamlanivimab-etesevimab, 4 248 casirivimab-imdevimab) were adjudicated as serious (Supplement). 249 250

During the trial the Alpha SARS-CoV-2 variant was the dominant variant of concern, while the 252 Delta variant became more prevalent in the final time epoch (Supplement). We found no 253 relative difference in mAb treatment effects over time and no comparisons reached a pre-254 specified statistical threshold (Supplement). 255

In a learning healthcare system trial of mAb for Covid-19, treatment of EHR-screen eligible 258 patients increased 7.5-fold, particularly among historically and geographically underserved 259 populations. We found a 91% and 94% probability of inferiority of bamlanivimab respectively to 260 bamlanivimab-etesevimab and casirivimab-imdevimab, and an 86% probability of equivalence 261 between bamlanivimab-etesevimab and casirivimab-imdevimab at the first pre-specified bound, 262 with regard to the odds of improvement in hospital-free days by 28 days. However, the 263 bamlanivimab and bamlanivimab-etesevimab arms were stopped early, and the identified 264 probabilities did not meet prespecified statistical triggers for trial conclusions of inferiority or 265 equivalence. 266 267 OPTIMISE-C19 is designed to continuously compare all available mAb for Covid-19 and can 268 stop mAb arms based on internal statistical triggers or external factors, such as U.S. federal 269 decisions limiting mAb availability or re-introduction of mAb into distribution networks. Had 270 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. screening of all patients within the system with a positive SARS-CoV-2 followed by direct-to-279 patient outreach phone calls, 3) paper referral form for patients without access to an in-system 280 provider, 4) community leader and existing outreach network collaboration for patient 281 engagement, 5) phone line for patients and community members to call and speak with a 282 healthcare professional, 6) providing infusions at home for patients without transportation. 283

The finding of potential inferiority of bamlanivimab is similar to mechanistic studies that suggest 285 a waning efficacy of bamlanivimab in the face of certain SARS-CoV-2 variants. It supports the 286 FDA decision to revoke the bamlanivimab EUA. 26 A recent observational study reached a 287 different conclusion and found similar effectiveness between bamlanivimab and casirivimab-288 imdevimab. 31 However, this study analyzed patients treated between November 2020 and 289

February 2021, prior to widespread emergence of variants and U.S. federal decisions to halt 290 bamlanivimab distribution. 291 292 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. where variant resistance to this mAb is < 5%, based on in vitro data of activity against the Delta 296 variant, and lack of activity against the Beta, Gamma, Delta plus, and B.1.621 variants. 31,32 The 297 similar effectiveness of bamlanivimab-etesevimab and casirivimab-imdevimab in the current 298 trial supports this decision. Also aligning with these data is a recent observational study of 165 299 patients that found no difference in hospitalization or death between bamlanivimab-etesevimab 300 and casirivimab-imdevimab in patients infected with Alpha, but worse outcomes with 301 bamlanivimab-etesevimab in patients infected with Gamma. 33 302 303

The strengths of this report include capture of nearly all patients infused with mAb from 49 sites 304 across a large geographic region, enhancing the generalizability of the results. In addition, an 305 advantage of the Bayesian design is that any data, including data following unplanned cessation 306 in enrollment into a trial arm, can be analyzed and quantified as posterior probabilities, which is 307 potentially more useful and is more quantitative than a frequentist conclusion of failure to reject 308 a null hypothesis possibly because of lack of power. 34 Third, the trial was embedded into usual 309 care which enhanced patient and provider engagement. 12 The trial also has limitations. First, the 310 results are presented before any prespecified internal trigger was reached. Nonetheless, to our 311 knowledge, this trial represents the largest randomized comparative effectiveness data of mAb 312

for Covid-19. Second, the absence of patient-level variant data limited ability to directly assess 313 comparative effectiveness relative to variant strains. Alpha was also the dominant variant during 314 the majority of the trial. Using regional data as a surrogate for variant data in the Pennsylvania 315 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint population over time, we found no difference in treatment effect over time. Third, we primarily 316 relied on UPMC EHR data to capture death and hospitalization, and patients may have accessed 317 care outside our health system after mAb treatment. We conducted direct-to patient calls and 318 national death registry queries to address this concern. Fourth, the EHR eligibility screen 319 identified most, but not all EUA risk factors, and could not identify if a patient was symptomatic. 320 321

In non-hospitalized patients with mild to moderate Covid-19, bamlanivimab, compared to 323 bamlanivimab-etesevimab and casirivimab-imdevimab, resulted in 91% and 94% probabilities of 324 inferiority with regards to odds of improvement in hospital-free days within 28 days. There was 325 an 86% probability of equivalence between bamlanivimab-etesevimab and casirivimab-326 imdevimab at an odds ratio bound of 0.25. However, the trial was unblinded early to these mAb 327 due to federal distribution decisions, and no mAb met prespecified criteria for statistical 328 inferiority or equivalence. 329 330 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint * IQR, interquartile range; mAb, monoclonal antibodies; SD, standard deviation. 480 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. Due to pharmacy logistics, 5 patients who received bamlanivimab-etesevimab had been 497 randomly assigned to casirivimab-imdevimab, and 7 patients who received casirivimab-498 imdevimab had been randomly assigned to bamlanivimab-etesevimab. All infused patients who 499 received bamlanivimab monotherapy had been randomly assigned to bamlanivimab 500 monotherapy. The FDA mAb policies changed over time, resulting in varying mAb availability 501 and EUA eligibility criteria over time (Supplement). 502 503 504 505 506 507 508 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. strategy to address many of these problems by gaining economies of scale from a common 581 platform, which allows for broad enrollment but retaining the ability to examine for 582 heterogeneity of treatment effects between defined subgroups. A REMAP focuses 583 predominantly on the evaluation of treatment options for the disease of interest that are 584 variations within the spectrum of standard care (although testing of novel or experimental 585 therapies is not precluded) and does so by embedding the trial within routine healthcare 586 delivery. In this regard, the REMAP seeks to replace random variation in treatment with 587 randomized variation in treatment allowing causal inference to be generated about the 588 comparative effectiveness of different existing treatment options. The use of response 589 adaptive randomization (RAR), which allows the allocation ratios to change over time based 590 on accruing outcomes data, maximises the chance of good outcomes for trial participants. The 591 embedding of such a platform within the day-to-day activities facilitates the translation of 592 outcomes to clinical practice as a "self-learning" system. As such, it also functions as an 593 embedded and automated continuous quality-improvement program. A final advantage of a 594 REMAP for optimizing monoclonal antibody treatment outcomes is the ability to rapidly 595 adapt to generate evidence if as new interventions emerge, avoiding the inevitable delays 596 associated with conventional trials. 597 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint A REMAP applies novel and innovative trial adaptive design and statistical methods to 598 evaluate a range of treatment options as efficiently as possible. The broad objective of a 599 REMAP is, over time, to determine and continuously update the optimal set of treatments for 600 the disease of interest. The set of treatments that may be tested within a REMAP comprise the 601 set of all treatments that are used currently or may be developed in the future and used or 602 considered for use in patients. The design maximizes the efficiency with which available 603 sample size is applied to evaluate treatment options as rapidly as possible. A REMAP has the 604 capacity to identify differential treatment effects in defined sub-groups (termed strata), These design features are: 618

• frequent adaptive analyses using Bayesian statistical methods 619

• response adaptive randomization 620 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint • evaluation of differential treatment effects in pre-specified sub-groups (strata) 621

• evaluation of specified intervention-intervention interactions 622

• testing of multiple interventions in parallel and, subsequently, in series 623

This creates a 'perpetual trial' with no pre-defined sample size, the objective of which is to 624 define and continuously update best treatment over the lifetime of the REMAP. The design 625 aspects, including the risk of type I and type II errors, are optimized prior to the 626 interventions with respect to the primary endpoint. Every participant will be assigned an 638 intervention. Inference in this REMAP is determined by analyses using pre-specified 639 statistical models that incorporate time periods, age, and disease severity to adjust for 640 heterogeneity of enrolled participants that might influence risk of death. These models 641 incorporate variables that represent each intervention assigned to participants. The efficacy of 642 each intervention is modeled as possibly varying in the different stratum in the REMAP. 643 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint Whenever a model hits a predefined threshold for any of superiority, inferiority, or 644 equivalence for an intervention with respect to the primary endpoint, this is termed a 645 statistical trigger. At any given adaptive analysis, a statistical trigger may be reached for all 646 participants or for one or more strata, further described in the statistical analysis plan 647 (Protocol Appendix, page 26) . 648 649 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Start incorporating bamlanivimabetesevimab into most sites. There was a 1-week delay to use etesevimab at six sites due to low initial supply.

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint disease, cystic fibrosis, and pulmonary hypertension) • Sickle cell disease • Neurodevelopmental disorders (for example, cerebral palsy) or other conditions that confer medical complexity (for example, genetic or metabolic syndromes and severe congenital anomalies) • Having a medical-related technological dependence (e.g., tracheostomy, gastrostomy, or positive pressure ventilation [not related to COVID-19])

Sotrovimab is a new mAb treatment.

Casirivimab 600mg and imdevimab 600 mg

Casirivimab-imdevimab dose decreased from 2400mg to 1200mg. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

This trial protocol has been provided by the authors to give readers additional information about 690 their work.

Protocol for: A learning health system randomized trial of monoclonal antibodies for Covid-19

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. • References …………………………………………………………………31 711 712 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. While COVID-19 vaccination will reduce COVID-19-related morbidity and mortality, the learned 771 immune response may vary between individuals. This means interventions such as monoclonal 772 antibodies (mAB) will still be needed to prevent progression of COVID-19 illness. Monoclonal 773 antibodies seek to mimic or enhance the natural immune system response against a pathogen and 774 are often used in the care of patients with cancer or infection. 775 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint For viral infections, mABs are created by exposing a white blood cell to a particular viral protein, 776 which is then cloned to mass produce antibodies to target that virus. For SARS-CoV-2, the virus 777 that causes COVID-19, IgG1 mABs target the spike protein of SARS-CoV-2 and block viral 778 attachment and entry into cells. Despite the EUAs, the clinical use of mABs is low due in part to lack of patient access, 804 complexities in drug allocation, and lack of knowledge among providers are contributing factors. 805

Further, the comparative effectiveness of different mABs is unknown and not yet directly studied. 806

The National Academies of Sciences, Engineering, and Medicine recently called for expanded 807 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint access and clinical use of mABs, noting it is "critical to collect data and evaluate whether they are 808 working as predicted". 809

This evaluation seeks to expand access to mABs at UPMC and determine their relative effects 810 versus each other, starting with those governed by EUAs. The primary objective is to evaluate the clinical and biological effect of multiple monoclonal 817

The primary hypothesis is clinical and biological effect will vary between mABs, by SARS-CoV-819 2 variants, and patient characteristics. 820

METRICS 821 822

The primary evaluation metric is total hospital free days (HFD) at 28 days after mAB receipt 823 calculated as 28 minus the number of days during the index stay minus the number of days 824 readmitted during the 28 days after treatment. Death within 28 days is recorded as -1 HFD. 825

Secondary evaluation metrics include: 826

• All-cause and all-location mortality at 28 and 90 days 827

• Emergency department visits at 28 days 828

• Organ-support free days at day 28 829

• Where feasible: 830

• SARS-CoV-2 nasopharyngeal and plasma viral loads among participants from baseline and 831 longitudinally through day 28 832

• SARS-CoV-2 antibody titers, antibody neutralization, and other immune responses at 833 baseline and longitudinally through day 28 834

• Detection of SARS-CoV-2 variants through next-generation sequencing at baseline and 835 longitudinally through day 28 836

• Determining the duration of SAR-CoV-2 infectivity and non-culture surrogates for SARS-837

CoV-2 infectivity among patients with persistent nasopharyngeal swab viral shedding 838 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

We will conduct a pragmatic evaluation of participants with COVID-19 illness under existing 841 UPMC processes for the clinical care of COVID-19 positive patients, including EUA requirements 842 for mAB administration. A patient who presents to a UPMC facility and tests positive for COVID-843 19 will, as per current common care, be offered monoclonal antibodies. Data that are already 844 collected according to UPMC procedures and EUA requirements are used for analysis. 845 846 9.

POPULATION 847 848

We will evaluate patients that present to UPMC Emergency Departments, urgent care sites, 849

infusions centers and other facilities that can or do provide mABs for COVID-19. As of February 850 24, 2021, there are 3 EUAs, with common inclusion and exclusion criteria, and we will evaluate 851 patients that meet these criteria. As other antibodies become available, we will modify this 852 evaluation submission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. As per the current EUA criteria, the following are excluded: 884

• Are hospitalized for the treatment of COVID-19 885

• Require oxygen therapy for the treatment of COVID-19 886

• Require an increase in baseline oxygen flow rate due to COVID-19 in those on chronic 887 oxygen therapy due to underlying non-COVID-19 related comorbidity 888

• Have a known hypersensitivity to any antibody ingredient 889

Patients will receive COVID-19 mABs governed by FDA EUAs, when their treating physician 892 orders a mAB and they meet EUA criteria. Currently and under our examination, the treating 893 physician do not choose a specific mAB product. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Bamlanivimab is a human immunoglobulin G-1 (IgG1 variant) monoclonal antibody consisting of The EUAs require that healthcare facilities and providers report therapeutic information and 933 utilization data through HHS Protect, Teletracking, or National Healthcare Safety Network as 934 directed by the US Department of Health and Human Services. 935 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

We will collect data including baseline demographics and underlying conditions, results of SARS-936 COV-2 PCR or antibody testing, and initial care including mAB infusion completion. We will 937 collect post-randomization healthcare encounters, including hospitalization, emergency 938 department visits, ICU care, and other measures of healthcare utilization. We will use an electronic 939 health record data collection process to augment existing UPMC data collection processes as 940

All data will be handled and secured as per University of Pittsburgh and UPMC data guidelines. 942

There will be no research activities involving direct interaction with subjects performed as part of 943 this evaluation. 944

In addition to the primary and secondary outcome data referenced in this submission, data collected 945 will include the below areas. All data will be abstracted directly from the electronic health record 946 and handled anonymously. 947

• Which mAB was administered, including date, time, and infusion completion as well as 948 the location of the infusion 949 Antibodies will be administered as per the EUAs and UPMC Pharmacy and Therapeutics policies. 965

Providers will explain mAB risks and benefits and provide the EUA Fact Sheets for Patients, 966 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The COVID-19 mABs are currently routinely used at UPMC. Once any order for mAB infusion 971 is approved by the UPMC system oversight group, the pharmacy provides whichever EUA-972 governed mAB is available under a therapeutic interchange approach. Ordering physicians review 973 with the patient the EUA Fact Sheet for each mAB and explain that the patient could receive any 974 of the mABs governed by FDA EUAs. 975

If demand for mAB exceeds supply, UPMC has a lottery system to allot who receives the therapy 976 once requested by a physician. 977

Our current proposal is a UPMC system quality improvement initiative, embracing and extending 978 the current lottery system and therapeutic interchange policy for EUA-governed mABs for 979 COVID-19 as follows: 980

a. If scarcity present and lottery system allow provision, proceed. 982 2. The Pharmacy fills order with one of the EUA-governed COVID-19 mABs using an embedded 983 assignment system akin to current mAb provision. This system will allow a comparative 984 effectiveness evaluation of the multiple mABs by effectively ensuring random allocation. 985

3. The Physician can agree to the assigned mAB or can request a specific mAB. 986

It is the treating physician's choice to accept the assigned mAB or not, and therefore patient 987 consent for the mAB assignment is not required. Patients will be told which mAB they are 988 receiving, along with an EUA Fact Sheet, as per EUA requirements. Predefined strata will include patients discharged home after infusion, patients admitted to hospital 996 after infusion, prior vaccination, and if known, presence of virus variants of concern at baseline 997 and presence of neutralizing antibodies to SARS-CoV-2 at baseline. 998 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. • Are hospitalized for the treatment of COVID-19 • Require oxygen therapy for the treatment of COVID-19 • Require an increase in baseline oxygen flow rate due to COVID-19 in those on chronic oxygen therapy due to underlying non-COVID-19 related comorbidity • Have a known hypersensitivity to any antibody ingredient Primary evaluation metric Total hospital free days at 28 days

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. While COVID-19 vaccination will reduce COVID-19-related morbidity and mortality, the learned 1109 immune response may vary between individuals. This means interventions such as monoclonal 1110 antibodies (mAB) will still be needed to prevent progression of COVID-19 illness. Monoclonal 1111 antibodies seek to mimic or enhance the natural immune system response against a pathogen and 1112 are often used in the care of patients with cancer or infection. 1113 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Despite the EUAs, the clinical use of mABs is low due in part to lack of patient access, 1148 complexities in drug allocation, and lack of knowledge among providers are contributing factors. 1149

Further, the comparative effectiveness of different mABs is unknown and not yet directly studied. 1150

The National Academies of Sciences, Engineering, and Medicine recently called for expanded 1151 access and clinical use of mABs, noting it is "critical to collect data and evaluate whether they are 1152 working as predicted". 1153

This evaluation seeks to expand access to mABs at UPMC and determine their relative effects 1154 versus each other, starting with those governed by EUAs. The primary objective is to evaluate the clinical and biological effect of multiple monoclonal 1161 antibodies (mABs) in patients with COVID-19. 1162

The primary hypothesis is clinical and biological effect will vary between mABs, by SARS-CoV-1163 2 variants, and patient characteristics. 1164

METRICS 1165 1166

The primary evaluation metric is total hospital free days (HFD) at 28 days after mAB receipt 1167 calculated as 28 minus the number of days during the index stay minus the number of days 1168 readmitted during the 28 days after treatment. Death within 28 days is recorded as -1 HFD. 1169

Secondary evaluation metrics include: 1170

• All-cause and all-location mortality at 28 and 90 days 1171

• Emergency department visits at 28 days 1172

• Organ-support free days at day 28 1173

• Where feasible: 1174

• SARS-CoV-2 nasopharyngeal and plasma viral loads among participants from baseline and 1175 longitudinally through day 28 1176 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. do not occur. If this happens at analysis, the cells will be combined to achieve model 1493

convergence. For example, if the 4 hospital-free day outcome value does not occur it will be 1494 combined with 3, and so on, until every cell has occurred. Additional model stability conventions 1495

will be taken to preserve the model stability. 1496

Missing and Partial Data 1497

If there are missing covariates for a patient in the as-infused patient population, the following 1498

conventions will be used. 1499

1. If the treatment arm is missing the patient will be ignored. 1500

2. If a baseline covariate is missing the referent value for that covariate will be used 1501 1502

For all model analyses, only patients who have achieved 28-days of follow-up from the date of 1503

the index infusion will be used in the analysis. No use or imputation of patient data for patients 1504

with less than 28 days will be conducted.

Given the HER-based data summaries there will be no missing outcome data. If there is deemed 1507

to be a corrupted outcome that patient will be ignored. Some patients may have 28 1508

hospitalization-free days that at subsequent analyses are found to have out of system 1509 hospitalizations. The data will be updated at future analyses. 1510

Trial Inferences 1511

For the primary analysis, there is no "control" treatment and so all inferences are made 1513

comparing the individual treatment arms to each other. The main quantity of interest will be the 1514 relative odds ratio between any two treatments arms 1515 1516 = exp( )/exp ( ).

The posterior probability that the odds ratio for arm i compared to arm j is greater than 1 1519

(signifying that treatment i is superior to treatment j) is used as a comparison between arms. 1520

Additionally, the posterior mean and 95% confidence interval between arms will be used to 1521

summarize relative treatment effects.

Arm Inferiority: If one of the arms has a 99% chance of being inferior to any of the other 1524 available arms then the inferior arm will be declared inferior and may be removed from the trial. 1525

There may be conditions of the pandemic (variation frequency, new variations) or drug supply 1526

concerns that an arm is retained. 1527 1528

Equivalence: Any two arms in the trial may reach a declaration of equivalence. It is anticipated 1529 that no actions would take when equivalence is reached but a declaration and public disclosure 1530 may be made. There is a sliding scale of equivalence with different levels of equivalence bounds. 1531 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

The following posterior probabilities will be reported for each time epoch All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

Title and abstract 1 How participants were allocated to interventions (e.g., "random allocation," "randomized," or "randomly assigned") (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. Method used to implement the random allocation sequence (e.g., numbered containers or central telephone), clarifying whether the sequence was concealed until interventions were assigned 6 Randomizationimplementation 10 Who generated the allocation sequence, who enrolled participants, and who assigned participants to their groups __ Blinding (masking) 11 Whether participants, those administering the interventions, and those assessing the outcomes were blinded to group assignment If blinding was not done, or was not possible, explain why 9

Statistical methods 12 Statistical methods used to compare groups for primary outcomes; methods for additional analyses, such as subgroup analyses and adjusted analyses 9-11

Flow of participants through each stage (a diagram is strongly recommended)-specifically, for each group, report the numbers of participants randomly assigned, receiving intended treatment, completing the study protocol, and analyzed for the primary outcome; describe deviations from planned study protocol, together with reasons in each group included in each analysis and whether analysis was by "intention-to-treat"; state the results in absolute numbers when feasible (e.g., 10/20, not 50%)

Outcomes and estimation 17

For each primary and secondary outcome, a summary of results for each 13 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted September 9, 2021. ; https://doi.org/10.1101/2021.09.03.21262551 doi: medRxiv preprint

Age (with categories of <30, 30-39, 40-49, 50-59, 60-69, 70-79, and ≥80; 60-69 will be 1447 used as the referent

Sex (sex at birth, male is the referent)

Covid variant is not modeled in this primary analysis 1450 5. Time (two-week epochs of time are used for adjustments; the most current 4-week period 3. Stacked bar plots and cumulative distributions of HFDs by sex 1590

No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted September 9

Fact sheet for health care providers emergency use authorization (EUA) of bamlanivimab

Fact sheet for health care providers emergency use authorization (EUA) of bamlanivimab and etesevimab

Fact sheet for health care providers emergency use authorization (EUA) of REGEN-COVTM

Fusing Randomized trials with big data: the key to self-learning health care systems?

Department of Health & Human Services. Monoclonal Antibodies: Updates from the Federal COVID Response

No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted September 9

The authors thank the clinical staff of the UPMC monoclonal antibody

NUMBER of PARTICIPANTS 999 1000Sample size is determined by case volume throughout the course of the pandemic. 1001

STATISTICAL ANALYSIS 1002 1003The primary evaluation metric is the number of days free from hospitalization to day 28. We will 1004 finalize a statistical analysis plan which will consider mAB assignment, heterogeneity of treatment 1005 effect by patient characteristics and virus variants, and interaction with other treatments. Due to 1006 uncertainty in sample size, we will use a Bayesian adaptive design to ensure ability to provide 1007 statistical inference despite variable sample size. 1008

27. DATA MONITORING 1011 1012Evaluation center leadership will regularly monitor monthly reports on enrollment, patient 1013 characteristics, and outcomes. 1014

CONSENT 1015As per EUA requirements, physicians will discuss the risks and benefits of mABs and patients will 1016 consent to receive a mAB as part of usual care, should they desire mAB treatment. As per UPMC 1017 policy, the ordering physician reviews with patients the EUA Fact Sheet for each mAB and explain 1018 that the patient could receive any of the mABs governed by FDA EUAs. 1019

ADVERSE EVENTS and SERIOUS ADVERSE EVENTS 1020 1021The EUAs require providers and/or their designees report all medication errors and serious adverse 1022 events potentially related to the antibodies within seven calendar days from the onset of the event. 1023Serious adverse events are defined as death, life-threatening event, inpatient hospitalization or 1024 prolongation of existing hospitalization, substantial disruption of ability to conduct normal life 1025 functions, a congenital anomaly/birth defect, or an intervention to prevent death, a life-threatening 1026 event, hospitalization, disability, or congenital anomaly. 1027The EUAs require adverse event reports be submitted to FDA MedWatch via one of multiple 1028 methods. Copies of all FDA MedWatch forms are also to be sent to the antibody manufacturer. 1029Thus, there already exist reporting requirements for UPMC associated with mAB prescription. We 1030 will track and record these reported data and adverse events by mAB assignment. 1031

SAFETY and RISK MITIGATION 1032 1033The EUAs stipulate warnings including hypersensitivity, clinical worsening, and side effects. As 1034 per EUA requirements, warnings will be communicated by providers to patients, adverse events 1035 will be reported as above, and post-infusion clinical monitoring will be done. We will evaluate patients that present to UPMC Emergency Departments, urgent care sites, 1194infusions centers and other facilities that can or do provide mABs for COVID-19. As of June 30, 1195 2021, there are 2 EUAs for COVID-19 mABs, with common inclusion and exclusion criteria, and 1196we will evaluate patients that meet these criteria. As FDA antibody decisions change (E.g., FDA 1197 revokes or grants EUAs, or changes eligibility criteria), eligibility criteria will change. 1198

INCLUSION CRITERIA 1199 1200As per the current EUA criteria (June 2021), the following patients are included: 1201• Adult (> 18 years old) 1202• Children > 12 years old weighing at least 40 kg 1203• With a positive SARS-CoV-2 antigen or PCR test and within 10 days of symptom onset 1204

High risk is defined as patients who meet at least one of the following criteria: 1206 • neurodevelopmental disorders (e.g., cerebral palsy) or other conditions that confer medical 1217 complexity (e.g., genetic, or metabolic syndromes and severe congenital anomalies) 1218• a medical-related technological dependence, for example, tracheostomy or gastrostomy) 1219The EUAs note that other medical conditions or factors (for example, race or ethnicity) may also 1220 place individual patients at high risk for progression and authorization of mAB treatment under 1221 the EUA is not limited to the medical conditions or factors listed above. 1222

EXCLUSION CRITERIA 1223 1224As per the current EUA criteria (June 2021), the following are excluded: 1225• Are hospitalized for the treatment of COVID-19 1226• Require oxygen therapy for the treatment of COVID-19 1227• Require an increase in baseline oxygen flow rate due to COVID-19 in those on chronic 1228 oxygen therapy due to underlying non-COVID-19 related comorbidity 1229• Have a known hypersensitivity to any antibody ingredient 1230

We will evaluate mABs governed by FDA EUAs. Patients will receive COVID-19 mABs 1233 governed by FDA EUAs, when their treating physician orders a mAB and they meet EUA criteria. 1234As FDA antibody decisions change (E.g., FDA revokes or grants EUAs, provides full approval, or 1235 changes eligibility criteria), available evaluated treatments will change. In April 2021, FDA 1236 revoked the EUA for bamlanivimab monotherapy and in June 2021 FDA recommended 1237 bamlanivimab and etesevimab not be used. 1238As of June 30, 2021, the EUA-approved mABs are as listed below. 1239

CASIRIVIMAB and IMDEVIMAB 1240 1241Casirivimab, a human immunoglobulin G-1 (IgG1) monoclonal antibody (mAb), is covalent 1242 The EUAs require that healthcare facilities and providers report therapeutic information and 1262 utilization data through HHS Protect, Teletracking, or National Healthcare Safety Network as 1263 directed by the US Department of Health and Human Services. 1264We will collect data including baseline demographics and underlying conditions, results of SARS-1265 COV-2 PCR or antibody testing, and initial care including mAB infusion completion. We will 1266 collect post-randomization healthcare encounters, including hospitalization, emergency 1267 department visits, ICU care, and other measures of healthcare utilization. We will use an electronic 1268 health record data collection process to augment existing UPMC data collection processes as 1269

All data will be handled and secured as per University of Pittsburgh and UPMC data guidelines. 1271There will be no research activities involving direct interaction with subjects performed as part of 1272 this evaluation. 1273In addition to the primary and secondary outcome data referenced in this submission, data collected 1274 will include the below areas. All data will be abstracted directly from the electronic health record 1275 and handled anonymously. 1276• mAB was administered, including date, time, and infusion completion as well as the 1277 location of the infusion 1278 Where feasible, we will collect discarded remnant blood samples and nasal/oropharyngeal swab 1288 samples to quantify the viral load and host response to the virus. As noted under data collection, 1289 we will record laboratory and microbiology data performed for clinical purposes. 1290 129119. ANTIBODY ADMINISTRATION 1292 1293Antibodies will be administered as per the EUAs, UPMC Pharmacy and Therapeutics policies and 1294 the respective Pharmacy Manuals (as generated by the pharmaceutical companies), if applicable. 1295Providers will explain mAB risks and benefits and provide the EUA Fact Sheets for Patients, 1296Parents and Caregivers 1297 as per EUA requirements. 1298 20.mAB assignment 1299 1300The COVID-19 mABs are currently routinely used at UPMC. Once any order for mAB infusion 1301 is approved by the UPMC system oversight group, the pharmacy provides whichever EUA-1302 governed mAB is available under a therapeutic interchange approach. Ordering physicians review 1303 with the patient the EUA Fact Sheet for each mAB and explain that the patient could receive any 1304 of the mABs governed by FDA EUAs. 1305If demand for mAB exceeds supply, UPMC has a lottery system to allot who receives the therapy 1306 once requested by a physician. 1307Our current proposal is a UPMC system quality improvement initiative, embracing and extending 1308 the current lottery system and therapeutic interchange policy for EUA-governed mABs for 1309 COVID-19 as follows: 13101. The Physician orders mAB. 1311 a. If scarcity present and lottery system allow provision, proceed. 1312 2. The Pharmacy fills order with one of the EUA-governed COVID-19 mABs using an embedded 1313 assignment system akin to current mAb provision. This system will allow a comparative 1314 effectiveness evaluation of the multiple mABs by effectively ensuring random allocation. 1315

It is the treating physician's choice to accept the assigned mAB or not, and therefore patient 1317 consent for the mAB assignment is not required. Patients will be told which mAB they are 1318 receiving, along with an EUA Fact Sheet, as per EUA requirements. Predefined strata will include patients discharged home after infusion, patients admitted to hospital 1327 after infusion, prior vaccination, and if known, presence of virus variants of concern at baseline 1328 and presence of neutralizing antibodies to SARS-CoV-2 at baseline. 1329

Sample size is determined by case volume throughout the course of the pandemic. 1332

The primary evaluation metric is the number of days free from hospitalization to day 28. We will 1335 finalize a statistical analysis plan which will consider mAB assignment, heterogeneity of 1336 treatment effect by patient characteristics and virus variants, and interaction with other 1337 treatments. Due to uncertainty in sample size, we will use a Bayesian adaptive design to ensure 1338 ability to provide statistical inference despite variable sample size. UPMC clinical leadership will regularly monitor monthly reports on enrollment, patient 1344 characteristics, and outcomes. Leadership will also receive regular interim analyses from the 1345 adaptive statistical model to inform UPMC clinical policy. 1346

As per EUA requirements, physicians will discuss the risks and benefits of mABs and patients will 1348 consent to receive a mAB as part of usual care, should they desire mAB treatment. As per UPMC 1349 policy, the ordering physician reviews with patients the EUA Fact Sheet for each mAB and explain 1350 that the patient could receive any of the mABs governed by FDA EUAs. 1351

The EUAs require providers and/or their designees report all medication errors and serious adverse 1354 events potentially related to the antibodies within seven calendar days from the onset of the event. 1355Serious adverse events are defined as death, life-threatening event, inpatient hospitalization or 1356 prolongation of existing hospitalization, substantial disruption of ability to conduct normal life 1357 functions, a congenital anomaly/birth defect, or an intervention to prevent death, a life-threatening 1358 event, hospitalization, disability, or congenital anomaly. 1359The EUAs require adverse event reports be submitted to FDA MedWatch via one of multiple 1360 methods. Copies of all FDA MedWatch forms are also to be sent to the antibody manufacturer. 1361Thus, there already exist reporting requirements for UPMC associated with mAB prescription. We 1362 will track and record these reported data and adverse events by mAB assignment.

The EUAs stipulate warnings including hypersensitivity, clinical worsening, and side effects. As 1366 per EUA requirements, warnings will be communicated by providers to patients, adverse events 1367 will be reported as above, and post-infusion clinical monitoring will be done. Administration of 1368 mABs for patients with COVID-19 is routine care at UPMC, and their administration is not a 1369 research procedure. 1370

As per the EUAs, all participants should be monitored closely, as there is a risk of infusion reaction 1372 and hypersensitivity (including anaphylaxis) with any biological agent. Symptoms and signs that 1373 may occur as part of an infusion reaction include, but are not limited to fever, chills, nausea, 1374 

This document describes the detail of the analysis read-out from July 26, 2021. This document is 1413 an appendix to the Statistical Analysis Plan for the "UPMC Antibody Treatment and Evaluation 1414Center" with the details for this analysis read-out. 1415 1416• We will analyze and report three treatment arms (below) as the first two have been 1417administratively closed to enrollment due to FDA decisions. 1418 1419• Unblinded data from this and prior interim analyses can be shared with investigators after 1420 the last randomized allocations (June 25) to the two mAB treatment arms closed by FDA. 1421 1422• This unblinding is appropriate as future analyses of patients randomized to the third arm 1423(C+I) will not be compared to the first two arms, there are no "control" arms as all 1424patients receive mAB treatment, and future comparisons will be of C+I vs newer mABs. 1425 1426• Enrollment continues in the currently available treatment arms (C+I and S). "S" refers to 1427 sotrovimab, produced by GSK and Vir. 1428

There are three treatment arms that will be included in this analysis. The treatment arms are 14301. B (Bamlanivimab) 14312. B+E (Bamlaniviman/Etesevimab combination) 14323. C+I (casirivimab/imdevimab combination) 1433

The primary endpoint for this read-out is hospital-free days. 1435

The primary analysis population for this read-out is the "As-Infused" population. This analysis 1437will include each patient randomized from March 10, 2021 until June 25, 2021. The date of the 1438 data set snapshot is on July 26, 2021. 1439

The primary analysis model is as described in the trial SAP. The appropriate coefficients will be set to 0 within each covariate for identifiability (the goal will 1469be the largest category set to 0).

The effects of time are adjusted within the model using two-week epochs and a smoothing model The prior distributions for the mAB treatment effects are weak: 1484 1485 ~(0,10), = 1,2,3. 1486

The treatment arms with the largest sample size should be selected as the referent 1487 arm (label a=1)_for treatment effects and assigned a treatment effect of 1 The following levels are pre-defined: 15371. The first level of equivalence occurs when there is 95% posterior probability that the 1538 odds-ratio is within a bound of d=0.25 15392. The second level of equivalence occurs when there is a 95% posterior probability that the 1540 odds-ratio is within a bound of d=0.20 15413. The third level of equivalence occurs when there is a 95% posterior probability that the 1542 odds-ratio is within a bound of d=0.15 1543 4. The fourth level of equivalence occurs when there is a 95% posterior probability that the 1544 odds-ratio is within a bound of d=0.10 1545 5. The fifth level of equivalence occurs when there is a 95% posterior probability that the 1546 odds-ratio is within a bound of d=0.05 1547 1548Combination Futility: For comparing B+E to B the combination (B+E) will be compared to the 1549 individual component arm (B). If there is more than a 95% probability that the effect of the 1550 combination B+E is no better than a 20% improvement in the odds ratio compared to B, then the 1551 combination will be declared not clinically relevantly superior (combination futile) to N. 1552 1553

For this read out, the following time epochs will be modeled with different treatment effects 1556using the Treatment Heterogeneity analysis model.

The treatment effect, , will be modeled separately within each epoch, = (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. PrimaryAs-Infused to B, B+I, or C+I on or before June 25, 2021 HFD 2

As-Infused to B, B+I, or C+I on or before June 25, 2021 HFD Differential efficacy by time epoch specified

The primary analysis 1575 1576The following posterior probabilities will be reported 1577Quantity