We included interventional studies with/without randomization and blinding. Each study must’ve reported evidence of ethical clearance by the local research ethics committee. Only articles in English with a full manuscript available were included. The type of intervention was MSC administration, without any restrictions on cell seeding/harvesting method and MSC dose. There were no restrictions on the patient's social status, ethnicity, race, or nationality. The expected comparisons were administration of a standard therapy regimen according to local health protocols, placebo, or MSC vehicle.

We excluded studies that included patients with: 1) age under 18 years (pediatric patients); 2) multiorgan failure at the start of study; or 3) inherited/acquired immunity disorder. Each study reported at least one of these variables: mortality rates, adverse events, need for ventilators, treatment duration, time for clinical improvement, and changes in inflammatory markers. Case reports or case series studies were also excluded from this systematic review. Mortality was defined as deaths that occurred during hospitalization. Adverse events were evaluated within six hours of MSC administration, which included urticaria, palpitations, and pulmonary edema. Cardiac arrest within 24 hours after MSC administration was also considered as an adverse event.

A systematic literature search of (Pubmed/MEDLINE, EBSCO/CINAHL, ProQuest, and Scopus from 1 January 2020 to 20 February 2021 was performed using the search strategy outlined in Table 1. ¹⁵ Additional records were retrieved from Google Scholar. Duplicate articles were removed after the initial search. Two authors independently screened the title and abstract of the articles. Articles that passed the screening were assessed in full text based on the eligibility criteria. Disagreements were resolved by discussion with the senior author. This study was conducted following the Preferred Reporting Item for Systematic Reviews and Meta-Analysis (PRISMA) statement. ¹⁶

We assessed the suitability of each study that appeared in the search engines by title and abstract. All appropriate articles were input in our database and duplicated using Zotero 5.0 (RRID:SCR_013784) citation manager application. All authors conducted an independent assessment of the eligibility of all studies. At this stage, the article eligibility assessment was conducted based on the full text of the assessed article. Any discrepancies that occurred were resolved by discussion.

All authors extracted the data independently. We developed a data extraction sheet that was referred to by the Cochrane Consumers and Communication Review Group. ^{15 , 17} Some of the data we extracted included study design, age, number of participants, comorbidities, type of intervention, and all of the previously mentioned outcomes. Numerical representations were preferred over graphs to avoid misinterpretation/estimation. However, graphic presentation in each article was not a reason to exclude an article.

This meta-analysis compared MSC safety and efficacy in COVID-19 patients against primary (mortality) and secondary (adverse events, treatment duration, need for mechanical ventilation) clinical outcomes. For treatment duration, the input data are numerical, therefore we used the mean difference from each study to be processed in this meta-analysis. We used a 2 × 2 format to assess mortality, side effects, and need for mechanical ventilation outcomes, therefore OR (Odd Ratio) were obtained from each study. We compiled all the outcomes from each study by considering their weight, using the RevMan 5.0 (RRID:SCR_003581) application. Studies with high heterogeneity were analyzed using the DerSimonian and Laird random-effects model and Mantel-Haenszel fixed effect model was used if heterogeneity was low.

Quality assessment was conducted by two independent reviewers (A and B). If there were differences in bias assessment results, a senior reviewer (C) was involved in making the final decision. Bias risk assessment in this study used the Risk Of Bias In Non-randomized Studies - of Interventions (ROBINS-I) tool and the revised Cochrane risk-of-bias (RoB) tool for randomized trials. ¹⁸ The RoB instrument was used to assess the quality of interventional randomized studies with/without blinding. ¹⁸ The ROBINS-I instrument was used to assess bias risk in interventional studies without randomization, according to Cochrane recommendations. The ROBINS-I tool used low, moderate, and serious as terminologies for risk stratification of the assessed studies. ¹⁸

There were 137 articles that appeared in the initial search of the entire electronic database. Checking for duplicated articles was conducted using Zotero 5.0 (RRID:SCR_013784) citation manager application, where 133 articles were found free from duplication. We screened articles based on title and abstract suitability, and only seven articles matched. Two articles were excluded because they did not use controls (comparators) and did not report outcomes that matched the outcomes set in this meta-analysis. The PRISMA flow diagram detailing the literature search can be seen in Figure 1.

We included five interventional studies with a total of 193 participants. Two out of the five studies did not apply randomization. One of the three randomized studies did not apply blinding to either participants or medical professionals. MSCs were taken from human umbilical cord, except for Leng et al., who did not specify the MSC origin used. ¹⁹ Shi et al., administered MSC therapy at a dose of 2 × 10 ⁶ cells/kg, ²⁰ while Lanzoni et al., used a fixed dose of 100 ± 20 × 10 ⁶ cells. ²¹ The other three studies did not mention the MSC dose administered explicitly.

All studies excluded pediatric patients. The mean age of participants across the studies was over 45 years. Meng et al., and Leng et al., excluded patients with comorbidities. ^{14 , 19} The other three studies included comorbid patients, with a distribution/proportion that did not differ significantly between groups. The study characteristics and outcomes reported for each study can be seen in Tables 2 and 3.

Study quality assessment based on risk of bias using the RoB and ROBINS-I tools showed that the included studies were of moderate to good quality ( Figure 2 and Table 4). A study by Leng et al., had moderate quality because there was no difference in the analysis of group characteristics. ¹⁹ This might lead to confounding bias. Furthermore, a study by Shu et al., did not apply blinding, thus triggering performance bias, detection bias, and attrition bias. ²¹

Four out of five studies reported mortality within 28 days after MSC therapy. Lanzoni et al., reported mortality incidence in one in 11 patients treated with MSC, and seven in 12 patients who were given standard therapy. ²¹ This led to a statistically significant difference in patient mortality rates between groups, with an OR of 0.07 [CI 95%: 0.01, 0.75]. Leng et al., and Shu et al., did not report any mortality in the intervention group. ^{18 , 21} Meanwhile, Shi et al., did not report any mortality in two groups. ¹⁹

Compilation of the study findings was conducted using a forest plot. The pooled OR for mortality risk for patients with MSC therapy was 0.13 [95% CI: 0.02, 0.68]. This value was obtained by the Mantel-Haenszel fixed effect model method because the heterogeneity was 0% and p value for heterogeneity was 0.76 ( Figure 3).

Adverse event is a parameter that could represent the safety of a product. The adverse events referred to were urticaria, palpitations, or pulmonary edema (within six hours) and/or cardiac arrest (within 24 hours) after MSC administration or placebo or MSC vehicle (without MSC). All studies reported this parameter in their articles. None of these studies reported a significant difference in adverse event incidence between groups. Meng et al., and Shu et al., found no adverse events in either group. ^{14 , 21}

With the Mantel-Haenszel fixed effect model method, it was found that the pooled OR for the MSC group that experienced adverse events was 0.91 [CI 95%: 0.45, 1.86], with a heterogeneity value of 51% ( Figure 4).

The need for a mechanical ventilator could represent the respiratory distress severity of a patient. In this systematic review, three out of five studies reported the need for a mechanical ventilator post-MSC therapy during the treatment period. None of the three studies reported a significant difference in the number of mechanical ventilators needed between groups. Only Shi et al., reported a higher percentage of mechanical ventilator use in the MSC group. ¹⁹

In this study, pooled OR for the MSC group that required a mechanical ventilator during treatment was 0.42 [95% CI: 0.12, 1.47]. This figure was obtained by Mantel-Haenszel fixed effect model, with a heterogeneity of 0% ( Figure 5).

There were only two studies reporting hospitalization duration. Meng et al. reported that the mean hospitalization duration in patients on MSC therapy was 20.5 ± 2.0207 days, and 23.25 ± 2.0361 days for patients without MSC administration. ¹⁴ Shu et al., also agreed that the mean hospitalization duration in patients on MSC therapy was shorter than patients who received standard therapy, with a mean difference of 4.00 [−5.44, −2.56] days. ²¹ The results of meta-analysis using the Mantel-Haenszel fixed effect model method found that the pooled mean difference was −3.54 [CI 95%: −4.68, −2.40] with 7% heterogeneity ( Figure 6).

Leng et al., found a significantly greater increase in the IL-10 ratio and decrease in TNF-α in the intervention group compared to the control group within 10 days post-MSC administration. ¹⁸ Meng et al., reported a significant reduction in IL-6 levels in patients receiving MSC therapy. ¹⁴ Changes in IL-6 levels were not found to be significant in patients on standard therapy. Lanzoni et al., found a significant decrease in various inflammatory markers at six days post-MSC administration. ²⁰ Inflammation markers that were found to be significantly decreased were TNF-α, IL-2, IL-6, and IL-7. There was no significant decrease in these inflammatory markers in the control group. Shu et al., reported significantly lower levels of CRP and IL-6 in the MSC group compared to the control group at days three and seven post-MSC administration. ²¹ On the same day, oxygenation index and lymphocyte count were found to be significantly higher in the MSC group. ²¹

This systematic review and meta-analysis included five interventional studies. We assessed the heterogeneity of each outcome with the parameter I2 value in percent (%). Heterogeneity was noted to be quite low in all outcomes, except for the side effect outcome, which had moderate heterogeneity (51%). Moderate heterogeneity of side effects may result from different definitions or types of side effect reported in each study. Using a funnel plot, we noted that there was no significant risk of publication bias in all outcomes ( Figure 7). The studies are evenly distributed in the right and left areas of the triangle forming a symmetrical funnel plot.

All data underlying the results are available as part of the article and no additional source data are required.

Figshare: PRISMA checklist for ‘Mesenchymal stem cell therapy efficacy in COVID-19 patients: A systematic review and meta-analysis’, https://doi.org/10.6084/m9.figshare.16602203.v1. ¹⁶

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).