Vitamin D, acute respiratory infections, and Covid-19: The curse of small-size randomised trials. A critical review with meta-analysis of randomised trials

Background Randomised trials conducted from 2006 to 2021 indicated that vitamin D supplementation (VDS) was able to prevent severe COVID-19 and acute respiratory infections (ARI). However, larger randomised trials published in 2022 did not confirm the health benefits of VDS in COVID-19 patients. Objective To examine through a systematic review with meta-analysis the characteristics of randomised trials on VDS to COVID-19 patients and admission to intensive care unit (ICU), and of randomised trials on VDS for the prevention of ARI. Method A systematic search retrieved randomised trials on VDS to COVID-19 patients and admission to ICU. Data on VDS and ARI were extracted from the meta-analysis of Jolliffe et al. 2021. Groups were formed including trials with total numbers of patients below or above the median size of all trials. The associations between VDS vs no VDS, and admission to ICU were evaluated using random-effects models from which summary odds ratios (SOR) and 95% confidence intervals (CI) were obtained. Meta-analyses were done for all trials and for each group of trials, which allowed testing a possible effect modification of trial size. Publication bias was assessed using the Louis-Furuya-Kanaruori (LFK) index (no bias if index between -1 and +1) and the trim and fill method. Results Nine trials on VDS for preventing admission to ICU were identified, including 50 to 548 patients. The summary odds ratio (SOR) was 0.61 (95% CI: 0.39-0.95) for all trials, 0.34 (0.13-0.93) for trials including 50 to <106 patients and 0.88 (0.62-1.24) for trials including 106 to 548 patients (interaction p = 0.04). The LFK index was -3.79, and after trim and fill, the SOR was 0.80 (0.40-1.61). The SOR for the 37 trials on VDS for ARI prevention included 25 to 16,000 patients. The SOR was 0.92 (0.86-0.99) for all trials, 0.69 (0.57-0.83) for trials including 25 to <248 patients and 0.98 (0.94-1.03) for trials including 248 to 16,000 patients (interaction p = 0.0001). The LFK index was -3.11, and after trim and fill, the SOR was 0.96 (0.88-1.05). Conclusion Strong publication bias affected small randomised trials on VDS for the prevention of severe COVID-19 and of ARI. Systematic reviews should beware of small-size randomised trials that generally exaggerate health benefits.