This study diligently followed the stringent guidelines set forth by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), a universally recognized framework that promotes transparent and thorough reporting in systematic reviews and meta-analyses (7, 8). Furthermore, demonstrating our dedication to methodological precision, we proactively registered our research protocol with the International Prospective Register of Systematic Reviews (PROSPERO), receiving the unique registration number CRD42024497474. This proactive pre-registration aimed explicitly at fostering transparency, showcasing our commitment to upholding the highest standards in conducting systematic reviews and meta-analyses.

Population: Studies involving individuals aged 18 years and above diagnosed with COVID-19.

Non-RCT studies, including retrospective studies and crossover trials, were excluded from this review to ensure the robustness of the results and avoid potential biases such as carryover effects.

A thorough systematic search was executed across various databases, notably the Cochrane Central Register of Controlled Trials, PubMed, Embase and Scopus. The search encompassed articles from the inception of the databases up until December 28, 2023. Language restrictions were not applied during the search process. The complete search strategy can be found in Supplementary Table S1.

The systematic search strategy was conducted to identify relevant articles. Two independent reviewers (WX and YL) screened the titles and abstracts of the retrieved studies. Subsequently, the selected articles were thoroughly assessed in full-text format by the same reviewers. Discrepancies or uncertainties encountered during the review process were resolved through collaborative discussion, with a third reviewer available to serve as an arbitrator if needed.

Two independent reviewers (WX and YL) performed data extraction from the included trials. The extraction process focused on gathering information pertaining to the study population, number of participants, mean age, and intervention details. To ensure the accuracy and reliability of the extracted data, a cross-check was conducted by a third reviewer to identify any errors or discrepancies. In instances where discrepancies arose between the two initial reviewers, a consensus was reached through a discussion among all reviewers.

In the methodological assessment of the included trials, two independent reviewers (YL and WX) utilized the Cochrane Risk of Bias tool to evaluate the potential bias (9). Our analysis utilized a tool that assesses bias across seven distinct domains, assigning each trial a study-level score indicative of the risk of bias (low, high, or unclear) within each domain. Any disagreements between reviewers were addressed through detailed discussions. In instances where consensus remained elusive, a conclusive determination was made by a third author (YZ).

Two evaluators, (YT and WX), independently appraised the quality of evidence pertaining to both primary and secondary outcomes employing the Grading of Recommendation, Assessment, Development, and Evaluation (GRADE) framework. Subsequently, they classified the gathered evidence into four distinct tiers: high, moderate, low, or very low. This classification was grounded on an array of factors, including the design of the study, the risk of bias, variability in outcomes, the accuracy of the data, and the relevance of the trials reviewed.

Statistical evaluations were carried out utilizing RevMan software (version 5.3), furnished by The Cochrane Collaboration. For outcomes measured in a binary manner, we calculated the relative risk (RR) accompanied by 95% confidence intervals (CI) (10). A p-value below 0.05 was considered statistically significant, and the I2 test was employed to quantify the proportion of total variability attributable to heterogeneity, providing a measure of the degree of heterogeneity. This helps us ascertain whether the studies are sufficiently homogeneous to justify pooling their results. Random-effects models were applied in cases where I2 exceeded 50%, indicating substantial heterogeneity, whereas fixed-effects models were employed when I2 was below 50%, to enhance the reliability of the findings. By adopting these rigorous statistical methods and principles, we aimed to provide a comprehensive and robust analysis of the available data. The potential for small study effects was qualitatively evaluated through the visual inspection of the funnel plot. Additionally, quantitative analysis was conducted using the Egger test, Peters test, and Harbord test. This comprehensive approach allowed for a thorough evaluation of any potential bias caused by small study effects.

Subgroup analysis was conducted to assess the in-hospital mortality among critically ill patients. The analysis aimed to investigate potential differences in outcomes across various sub-groups.

We conducted a sensitivity analysis using the following methods: (1) Excluding studies with high or unknown risk of bias; (2) excluding trials with a weight less than 10%.

To reduce the risk of type I error, we performed a trial sequential analysis (TSA 0.9Beta), combining the estimated information size with a revised significance threshold. This approach aimed to maintain a 5% type I error risk and attain an 80% statistical power, using a two-sided trial sequential analysis.

The initial search strategy yielded 4,118 records. After removing duplicate entries, we were left with 3,298 unique records. These records were then carefully assessed through the examination of their titles, abstracts, and full texts. Through this thorough evaluation process, we identified 11 trials that satisfied the detailed criteria established for this systematic review. Please refer to Figure 1 for a visual representation of this process.

Table 1 provides an overview of the trial characteristics considered in this study. The identified trials were published between 2020 and 2023, featuring sample sizes that varied from 20 to 2,591 patients.

Risk-of-bias assessments are detailed in Supplementary Figures S1, S2. Among the evaluated trials, five were found to have a low risk of bias, three presented an unclear risk, and three exhibited a high risk. The quality of evidence for the primary outcome, as appraised using the GRADE methodology, ranged from moderate to high, as shown in Table 2.

This meta-analysis revealed that the overall in-hospital mortality rate was comparable between COVID-19 patients supplemented with vitamin C and those undergoing standard treatment (RR = 0.85; 95% CI: 0.62–1.17) (Figure 2). Furthermore, the trial sequential analysis of mortality indicated a significant shortfall in reaching the required information size (Figure 2). In light of the conducted sensitivity analysis, our findings remain robust, underscoring the reliability of our results (Supplementary Table S2). Moreover, according to the results of the symmetric funnel plot analysis, it is apparent that the funnel plot exhibited asymmetry, thereby underscoring the imperative for quantitative analysis (Supplementary Figure S3).

Subgroup analysis, specifically targeting mortality outcomes among critically ill patients, was conducted. However, this analysis did not reveal any statistically significant differences across any of the outcomes or sub-groups examined (Supplementary Figure S4).

Patients supplemented with vitamin C did not exhibit a prolonged ICU stay (OR = 0.86; 95% CI: −0.54 to 2.25; p = 0.23) or hospital stay (OR = −0.54; 95% CI: −3.10 to 2.01; p = 0.68) (Figure 3), with a comparable duration to those undergoing standard treatment.