We included asymptomatic adults (≥18 years of age) who had a recent history of close-contact exposure to a PCR-confirmed case patient with Covid-19 (i.e., >15 minutes within 2 m, up to 7 days before enrollment), who had no Covid-19–like symptoms during the 2 weeks before enrollment, and who had an increased risk of infection (e.g., a health care worker, a household contact, a nursing-home worker, or a nursing-home resident). Trial candidates were tested by PCR assay for SARS-CoV-2 infection at baseline. We included candidates with either a negative or positive PCR test at baseline to assess the prophylactic and preemptive effect of hydroxychloroquine treatment, respectively. All eligibility criteria are listed in the Supplementary Appendix and the trial protocol, both available with the full text of this article at NEJM.org.
Trial Design and Oversight
This was an open-label, phase 3, cluster-randomized trial conducted from March 17 to April 28, 2020, during the early stages of the Covid-19 outbreak, in three of nine health administrative regions in Catalonia, Spain (total target population, 4,206,440) (Fig. S1 in the Supplementary Appendix). Trial candidates were screened with the use of the electronic registry of the national health information system.13
The trial was supported by the crowdfunding campaign YoMeCorono (https://www.yomecorono.com/), Generalitat de Catalunya, Zurich Seguros, Synlab Diagnósticos, Laboratorios Rubió, and Laboratorios Gebro Pharma. Laboratorios Rubió donated and supplied the hydroxychloroquine (Dolquine). The sponsors had no role in the conduct of the trial, the analysis, or the decision to submit the manuscript for publication. The trial protocol and subsequent amendments were approved by the institutional review board at Hospital Germans Trias i Pujol and the Spanish Agency of Medicines and Medical Devices. All the participants provided written informed consent.
We defined trial clusters (called rings) of healthy persons (contacts) who were epidemiologically linked to a PCR-positive case patient with Covid-19 (index case patient). All the contacts in a ring simultaneously underwent cluster randomization (in a 1:1 ratio) to either the hydroxychloroquine group or the usual-care group. Contacts in the former group received hydroxychloroquine (Dolquine) at a dose of 800 mg on day 1, followed by 400 mg once daily for 6 days; the dosing regimen was based on pharmacokinetic simulations. Contacts in the usual-care group received no specific therapy. After cluster randomization, we verified the selection criteria of individual candidates, obtained informed consent, and revealed the trial-group assignments. In accordance with national guidelines, all the contacts were quarantined.
All the contacts were visited at home or in the workplace on day 1 (enrollment) and day 14 (final outcome measurement) for assessment of health status and collection of nasopharyngeal swabs. Symptoms were monitored by telephone on days 3 and 7. Contacts in whom symptoms developed at any time point were visited at home within 24 hours for assessment of health status and collection of nasopharyngeal swabs. Safety (i.e., frequency and severity of adverse events), medication adherence (i.e., treatment and number of doses taken), and crossover (i.e., unplanned conversion from usual care to hydroxychloroquine) were assessed with the use of contact reports collected in telephone interviews on days 3, 7, and 28.
All testing of nasopharyngeal swabs for SARS-CoV-2 and analyses to determine viral load were performed by technicians who were unaware of previous PCR results, trial-group assignments, and response. PCR amplification was based on the 2019 Novel Coronavirus Real-Time RT [reverse transcriptase]–PCR Diagnostic Panel guidelines of the Centers for Disease Control and Prevention.14 For quantification, a standard curve was built with the use of 1:5 serial dilutions of a SARS-CoV-2 plasmid (with known concentration) and run in parallel with 300 study samples. The accuracy of the qualitative estimate (i.e., cycle threshold [Ct] values) was determined by correlation with the quantitative measure on 300 samples (Fig. S2). The coefficient of correlation between the two methods was 0.93, which permitted the use of qualitative Ct data to estimate viral load in contacts. Detection of IgM and IgG antibodies was performed by means of fingertip blood testing on the day 14 visit with the use of a rapid test (VivaDiag COVID-19).15
The primary outcome was the onset of a PCR-confirmed, symptomatic Covid-19 episode, defined as symptomatic illness (at least one of the following symptoms: fever, cough, difficulty breathing, myalgia, headache, sore throat, new olfactory or taste disorder, or diarrhea) and a positive RT-PCR test for SARS-CoV-2. The primary outcome was assessed in all asymptomatic contacts, irrespective of the baseline PCR result; in a post hoc analysis, we explored the outcome separately in contacts with a positive baseline PCR test and those with a negative baseline PCR test. The time until the primary event was defined as the number of days until the onset of symptomatic illness from the date of exposure and from the date of randomization.
The secondary outcome was the incidence of SARS-CoV-2 infection, defined as either the RT-PCR detection of SARS-CoV-2 in a nasopharyngeal specimen or the presence of any of the aforementioned symptoms compatible with Covid-19. The rationale for this outcome was to encompass definitions of Covid-19 used elsewhere.12,16 Contacts who were hospitalized or who died and whose hospital and vital records listed Covid-19 as the main diagnosis (including PCR confirmation) were also considered for the primary and secondary outcomes.
With an enrollment target of 95 clusters per trial group17 ― 15 contacts per cluster and intraclass correlation of 1.0 ― the initial design provided a power of 90% to detect a between-group difference of 10 percentage points in the incidence of PCR-confirmed, symptomatic Covid-19, with an expected incidence of 5% in the hydroxychloroquine group and 15% in the usual-care group. Owing to the limited information available by March 2020 regarding the cluster size and the incidence of Covid-19 after exposure, the protocol prespecified a sample-size reestimation at the interim analysis. Reestimation was aimed at maintaining the ability (at 80% power) to detect a between-group difference of 3.5 percentage points in the incidence of primary-outcome events (3.0% in the hydroxychloroquine group and 6.5% in the usual-care group), yielding 320 clusters per trial group with 3.5 contacts per cluster, an intraclass correlation of 1.0, and no provision for crossover.
The primary efficacy analysis was performed in the intention-to-treat population. Multiple imputation by chained equations was applied to account for missing data.18,19 The assumption that unobserved values were missing at random was deemed to be appropriate because we could not find any pattern among the missing values.20 A complete-case analysis and a per-protocol analysis were conducted as sensitivity analyses. The cumulative incidence of trial outcomes was compared at the individual level with the use of a binomial regression model with robust sandwich standard errors to account for grouping within clusters.21 We defined a generalized linear model with a binomial distribution and a log-link function to estimate the risk ratio as a measure of effect.22 The analyses were adjusted for the baseline variables of age, sex, geographic region, and time of exposure. We performed additional prespecified analyses to assess the consistency of treatment effects in subgroups defined according to the viral load of the contact at baseline, viral load of the index case patient, place of exposure, and time of exposure to the index case patient. The reported confidence intervals have not been adjusted for multiple comparisons and cannot be used to infer effects. Survival curves according to trial group for time-to-event outcomes were compared with the use of a Cox proportional-hazards model with a cluster-level frailty term to adjust for clustering.23 The significance threshold was set at a two-sided alpha value of 0.05, unless otherwise indicated. All statistical analyses were conducted with R software, version 18.104.22.168