ABSTRACT
BACKGROUNDThe objective is to determine the impact of the Bacillus Calmette-Guerin (BCG) vaccine compared to placebo or no vaccine on COVID-19 infections and hospitalisations in healthcare workers. We are using a living and prospective approach to Individual-Participant-Data (IPD) meta-analysis of ongoing studies based on the Anytime Live and Leading Interim (ALL-IN) meta-analysis statistical methodology. METHODSPlanned and ongoing randomised controlled trials were identified from trial registries and by snowballing (final elicitation: Oct 3 2022). The methodology was specified prospectively - with no trial results available - for trial inclusion as well as statistical analysis. Inclusion decisions were made collaboratively based on a risk-of-bias assessment by an external protocol review committee (Cochrane risk-of-bias tool adjusted for use on protocols), expected homogeneity in treatment effect, and agreement with the predetermined event definitions. The co-primary endpoints were incidence of COVID-19 infection and COVID-19-related hospital admission. Accumulating IPD from included trials was analysed sequentially using the exact e-value logrank test (at level = 0.5% for infections and level = 4.5% for hospitalisations) and anytime-valid 95%-confidence intervals (CIs) for the hazard ratio (HR) for a predetermined fixed-effects approach to meta-analysis (no measures of statistical heterogeneity). Infections were included if demonstrated by PCR tests, antigen tests or suggestive lung CTs. Participants were censored at date of first COVID-19-specific vaccination and two-stage analyses were performed in calendar time, with a stratification factor per trial. RESULTSSix trials were included in the primary analysis with 4 433 participants in total. The e-values showed no evidence of a favourable effect of minimal clinically relevance (HR < 0.8) in comparison to the null (HR = 1) for COVID-19 infections, nor for COVID-19 hospitalisations (HR < 0.7 vs HR = 1). COVID-19 infection was observed in 251 participants receiving BCG and 244 participants not receiving BCG, HR 1.02 (anytime-valid 95%-CI 0.78-1.35). COVID-19 hospitalisations were observed in 13 participants receiving BCG and 7 not receiving BCG, resulting in an uninformative estimate (HR 1.88; anytime-valid 95%-CI 0.26-13.40). DISCUSSIONIt is highly unlikely that BCG has a clinically relevant effect on COVID-19 infections in healthcare workers. With only limited observations, no conclusion could be drawn for COVID-19 related hospitalisation. Due to the nature of ALL-IN meta-analysis, emerging data from new trials can be included without violating type-I error rates or interval coverage. We intend to keep this meta-analysis alive and up-to-date, as more trials report. For COVID-19 related hospitalisations, we do not expect enough future observations for a meaningful analysis. For BCG-mediated protection against COVID-19 infections, on the other hand, more observations could lead to a more precise estimate that concludes the meta-analysis for futility, meaning that the current interval excludes the HR of 0.8 predetermined as effect size of minimal clinical relevance. OTHERNo external funding. Preregistered at PROSPERO: CRD42021213069.
ABSTRACT
Substantial COVID-19 research investment has been allocated to randomized clinical trials (RCTs) on hydroxychloroquine/chloroquine, which currently face recruitment challenges or early discontinuation. We aimed to estimate the effects of hydroxychloroquine and chloroquine on survival in COVID-19 from all currently available RCT evidence, published and unpublished. We conducted a rapid meta-analysis of ongoing, completed, or discontinued RCTs on hydroxychloroquine or chloroquine treatment for any COVID-19 patients (protocol: https://osf.io/QESV4/). We systematically identified unpublished RCTs (ClinicalTrials.gov, WHO International Clinical Trials Registry Platform, Cochrane COVID-registry up to June 11, 2020), and published RCTs (PubMed, medRxiv and bioRxiv up to October 16, 2020). All-cause mortality was extracted (publications/preprints) or requested from investigators and combined in random-effects meta-analyses, calculating odds ratios (ORs) with 95% confidence intervals (CIs), separately for hydroxychloroquine and chloroquine. Prespecified subgroup analyses included patient setting, diagnostic confirmation, control type, and publication status. Sixty-three trials were potentially eligible. We included 14 unpublished trials (1308 patients) and 14 publications/preprints (9011 patients). Results for hydroxychloroquine are dominated by RECOVERY and WHO SOLIDARITY, two highly pragmatic trials, which employed relatively high doses and included 4716 and 1853 patients, respectively (67% of the total sample size). The combined OR on all-cause mortality for hydroxychloroquine was 1.11 (95% CI: 1.02, 1.20; I2=0%; 26 trials; 10,012 patients) and for chloroquine 1.77 (95%CI: 0.15, 21.13, I2=0%; 4 trials; 307 patients). We identified no subgroup effects. We found that treatment with hydroxychloroquine was associated with increased mortality in COVID-19 patients, and there was no benefit of chloroquine. Findings have unclear generalizability to outpatients, children, pregnant women, and people with comorbidities.