Methods
Trial design
CORIMUNO-19 is a publicly supported platform, established by Assistance Publique-Hôpitaux de Paris, France, at the beginning of the covid-19 pandemic, dedicated to performing cohort, open label, randomised controlled trials of immune modulatory drugs in patients admitted to hospital with moderate or severe covid-19 disease.17 CORIPLASM (Efficacy of Convalescent Plasma to Treat Covid-19 Patients, a Nested Trial in the CORIMUNO-19 Cohort) was an embedded multicentre, open label, randomised controlled trial in patients with moderate covid-19 pneumonia conducted in French hospitals. Online supplemental appendix II and III have the full trial protocol and statistical analysis plan.
Study population and randomisation
At hospital admission, patients were evaluated for eligibility criteria: adults aged ≥18 years admitted to hospital, positive test result for the SARS-CoV-2 virus by nasopharyngeal polymerase chain reaction or computed tomography scan, or both, before randomisation, onset of symptoms <9 days, illness of mild or moderate severity according to the WHO Clinical Progression Scale (admitted to hospital, mild disease, no oxygen needed; admitted to hospital, moderate disease, oxygen <3 litres needed, online supplemental appendix I), not pregnant, no previous severe grade 3 allergic reaction to plasma transfusion, and no current bacterial infection reported.
ABO compatibility with available covid-19 convalescent plasma was verified before inclusion of patients. Written informed consent was obtained from all patients or their legal representatives at inclusion in CORIMUNO-19. Specific written informed consent was sought from eligible patients before inclusion in the CORIPLASM trial. The independent clinical research organisation compiled the computerised randomisation list, and the patient's randomisation number was accessed through a secure site by a site study team member. Randomisation was performed within two hours of enrolment. Eligible patients were randomised 1:1 to receive convalescent plasma or usual care. Usual care could include the use of dexamethasone, tocilizumab, supportive care, including supplemental oxygen, antiviral agents, and antibiotics. A data and safety monitoring board provided guidance on the trial after inclusion of every 60 patients.
Study product
Convalescent donors were eligible for plasma donation 15 days after the end of symptoms related to covid-19 disease. Apheresis plasma was collected by Etablissement Français du Sang and underwent pathogen reduction (Intercept Blood System, Cerus, Concord, CA) and standard testing according to current regulations in France. Anti-SARS-CoV-2 potency was assessed in each donation, with a requirement for a SARS-CoV-2 seroneutralisation titre ≥40, as described by Gallian et al.18 Antibody content was determined by immunoglobulin G enzyme linked immunosorbent assay (Euroimmun, Bussy-Saint-Martin, France). Covid-19 convalescent plasma with a seroneutralisation titre ≥40, made available for the trial and collected between April and June 2020, gave a mean enzyme linked immunosorbent assay ratio of 6.1 (standard deviation 2.9, range 0.4-13.0). After the first three patients received two units of ABO compatible covid-19 convalescent plasma according to the protocol, all subsequent patients randomised to the convalescent plasma group received four units of convalescent plasma (200-220 mL/unit, 2 units/day over two consecutive days) provided by different donors.
Study endpoints
As in all of the CORIMUNO-19 nested trials, an early primary endpoint was defined as a WHO Clinical Progression Scale score of ≥6 (online supplemental appendix I) on day 4 of randomisation. Higher values on the WHO Clinical Progression Scale indicate a worse outcome. The primary endpoint specific to the CORIPLASM trial was survival without the need for assisted ventilation (including non-invasive ventilation or high flow oxygen) at day 14 of randomisation (WHO Clinical Progression Scale score <6) or additional immunomodulatory treatment, with the exception of corticosteroids included within the standard of care (changes to the protocol, online supplemental file 1). Secondary endpoints were WHO Clinical Progression Scale score on days 4, 7, and 14 after randomisation, overall survival on days 14 and 28 after randomisation (ie, for the periods days 1-14 and days 1-28, respectively), time to discharge, time to end of dependence on oxygen supply, and changes to a series of biological parameters at days 4, 7, and 14 after randomisation.
Predefined subgroup analyses included immunosuppression status (underlying immunodeficiency: yes/no), duration of symptoms before randomisation (≤5 days, >5 days), and use of steroids. Safety data included all clinical and biological adverse events observed during the study follow-up. Immunodeficiency was defined as the presence of at least one of these medical conditions: active malignant neoplasm, lymphoid or myeloid neoplasms, haematopoietic stem cell or solid organ transplantation, or HIV/AIDS and not receiving highly active antiretroviral treatment.
Statistical analysis
The sample size was set at 120 participants (60 per group), with a bayesian interim analysis after 60 participants were randomised. We computed that the trial would have a frequentist power of 97.2% to detect a decrease in event proportions from 0.50 to 0.20, and 73.9% to detect a decrease in event proportions from 0.50 to 0.30. The study statisticians, who were masked to the group assignments, oversaw the interim and final analyses. Interim analysis reports were shared only with members of the data and safety monitoring board and not with the trial investigators. The trial investigators were blinded to all results during the trial.
The treatment effect was mainly expressed as an absolute risk difference for the early primary endpoint, and a hazard ratio for the longer term primary endpoint. Both were analysed in a bayesian framework. A posterior probability of absolute risk difference <0 or hazard ratio <1 but >0.99 at the interim analysis or >0.95 at the final analysis, indicated efficacy. We also computed posterior probabilities of absolute risk difference <−5.5% and hazard ratio <0.85, denoting a moderate or greater effect. At the interim analysis, a posterior probability of moderate or greater impact <0.20 defined a futility boundary. The treatment effect was summarised by the posterior median and equal tail credible intervals.
Because decision rules are one sided, consistent credible intervals would theoretically be one sided 95% credible intervals, but we chose to report two sided 90% credible intervals with the same upper boundary. For the early primary endpoint, the posterior distribution of absolute risk difference was computed analytically, with a beta prior distribution, with parameters one and one for the proportion in each group. An odds ratio adjusted for age and centre (centre being treated as a random effect) was also estimated with a bayesian logistic regression model. For the longer term primary endpoint, the posterior hazard ratio distribution adjusted for age and centre was computed with Markov chain Monte Carlo with normal prior distributions, with mean 0 and variance 106 for the log hazard ratio. Different prior distributions were used as sensitivity analyses.
Secondary outcomes were analysed in a frequentist framework, except for WHO Clinical Progression Scale scores, analysed as an ordinal variable with a bayesian proportional odds model. Analyses of secondary outcomes were also adjusted for age and centre. For time to discharge and time to end of dependence on oxygen supply, we estimated adjusted subdistribution hazard ratios with Fine-Grey models, death being the competing event. Estimating subdistribution hazard ratios was preferred over cause specific hazard ratios because subdistribution hazard ratios have a one-to-one relation with the cumulative incidence (ie, the proportion) of events, and we considered that subdistribution hazard ratios would therefore be more relevant than the ratio of rates at which these events occur in time. We used interaction tests between the treatment group and subgroups to test for treatment effect heterogeneity between the subgroups, with similar regression models as the main adjusted analyses. The statistical analysis plan has full details of the statistical analyses (online supplemental appendix III).
Analyses were done on an intention-to-treat basis. The original protocol specified a modified intention-to-treat analysis excluding patients declining the intervention and those who could not receive the planned plasma treatment because ABO compatible covid-19 convalescent plasma was not available. Because those situations did not occur, no modified intention-to-treat analysis was performed. No correction for multiplicity was done for secondary outcomes, and corresponding results should be regarded as exploratory. Two interim analyses were conducted (online supplemental table S1). Statistical analyses were done with SAS (version 9.4, SAS Institute) and R (version 4.0.5, R Foundation) statistical software.
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research, because the trial has been designed in an emergency setting. However, according to the French law, the results have been sent to all participants or to their families.