We followed MOOSE (Meta-analysis of Observational Studies in Epidemiology) guidelines to conduct and report this study (8). We searched PubMed, Embase, and the Cochrane Library for relevant studies with no language limitation from inception to February 1, 2021. The search strategy is illustrated in Table 1.

We included observational studies that examined the effects of tocilizumab on COVID-19. Two authors (Chi-Chung Chen and Hsien-Lung Tsai) independently scanned the search results, reviewed the full text of potential studies, and decided whether these studies met our selection criteria. The third author (Tao-Hsin Tung) was consulted if there was any disagreement.

The following data were extracted from the included studies: first author, year of publication, country, study subjects, and outcomes. Outcomes included mortality rates between tocilizumab and control groups. At the end of 2020, treatment of COVID-19 was various. Classes of drugs used are antiviral agents (e.g., lopinavir/ritonavir and remdesivir), inflammation inhibitors/antirheumatic drugs (e.g., tocilizumab, antimalarials, and methylprednisolone), low molecular-weight heparin, plasma, and hyperimmune immunoglobulins (3, 9). Definite therapy was still under development and research. Two authors (Chi-Chung Chen and Hsien-Lung Tsai) independently utilized the Newcastle-Ottawa Scale (NOS) to assess the quality of included non-randomized studies. The NOS applies three domains (selection of study groups, comparability, and outcome assessment) to evaluate the quality of these studies (10). A study could be awarded up to one star for each item in the selection and outcome domains and up to two stars for comparability. We considered a study of high quality if seven or more stars were awarded. Any disagreement was adjudicated by the third author (Tao-Hsin Tung).

Review Manager 5.4.1 (Copenhagen: The Nordic Cochrane Center, The Cochrane Collaboration, 2020) was used to conduct the meta-analysis. We presented the risk of outcome as risk ratio (RR), hazard ratio (HR), and 95% confidence interval (CI), and assessed heterogeneity using I² statistic, which evaluated the degree of variation across studies that was due to heterogeneity rather than chance alone. An I² value of ≥50% represented substantial heterogeneity. We conducted a random-effects model meta-analysis because we expected considerable clinical heterogeneity.

There were 784 studies found on Embase, 142 on PubMed. and 52 studies on the Cochrane Library. After removing duplicates, 853 articles were identified. A total of 15 studies meet the inclusion criteria, as shown in Figure 1. A total of 1,993 study subjects were involved in these 15 studies. Table 2 illustrates the characteristics of these included studies. Most of the included studies were conducted in Europe and the United States. All the included studies were focused on the effects of tocilizumab on the severity of COVID-19, and all the studies contained experiment groups (patients treated with tocilizumab) and control groups (patients not treated with tocilizumab).

Drugs used in control groups were various, such as hydroxychloroquine, azithromycin, remdesivir, and methylprednisolone (12, 14, 23). It is worth acknowledging that using adjunctive corticosteroid in patient with COVID-19 with acute respiratory distress syndrome could be a choice (12, 26).

The risk of bias assessment for included studies is also illustrated in Table 2. We used The Newcastle-Ottawa Scale (NOS) to assess the quality of these studies (10). Despite the short research time, there were two studies considered to not have a relatively long study period (<2 weeks) (8, 9). However, these two studies still had some valuable statistical information.

Figure 2 indicates the results of mortality in patients with COVID-19 receiving tocilizumab. Fifteen studies with a total of 854 COVID-19 patients receiving tocilizumab and 1139 patients as controlled group were collected. Our study found substantial heterogeneity (I² = 65%) among the 15 studies. Overall decreased mortality was found (pooled RR:0.62, 95% CI = 0.46–0.83). Subgroup heterogeneity was significant (I² = 65%).

Figure 3 illustrates the hazard ratio of using tocilizumab to treat patients with COVID-19. Overall decreased mortality was also found (pooled HR.61, 95% CI = 0.51–0.72). I² of subgroup heterogeneity was 0%. In summary, adult patients suffering from COVID-19 showed lower mortality compared to the controlled-group, which was treated with other therapy.

Among our studies, one of them showed a good effect of using tocilizumab on treatment of patients with COVID-19. Patients receiving tocilizumab showed significantly better survival rate than control groups (HR for death 0.035; 95% CI =0.004–0.347; p = 0.004) (22). After detailed analysis, the included subjects were all diagnosed as having COVID-19 but without mechanical ventilation. Maybe the relatively less severity of COVID-19 infection could explain the good effect of using tocilizumab.

Tocilizumab was associated with improved survival; however, it increased the proportion of patients with superinfections (54 vs. 26%; p < 0.001). Despite higher superinfection occurrence, there was no difference in 28-day case fatality rate (12).

Possibly because of the result of several biochemical, respiratory, and infectious complications, using tocilizumab in patients with COVID-19 was associated with longer hospital stay (HR 1.658, 95% CI = 1.088–2.524, p = 0.019) (23). By the way, a study in India showed longer days of hospitalization in the tocilizumab group (14 days) than in the control group (6 days) (p = 0.001) (25).

On the contrary, another study conducted in Sweden showed different results. Median length of stay in hospital was 20.5 (days) in the tocilizumab group and 30 (days) in the control group (p = 0.04) (16).

Treating COVID-19 with tocilizumab showed some variations in biochemistry indicators. After treatment, C-reactive protein (CRP) showed a clear reduction (P < 0.001). Meanwhile, D-dimer and lymphocyte both showed a significant increase (p < 0.001) (23). Despite short period, tocilizumab still showed positive influence on lowering CRP level (from 23.7 to 0.9 mg/dl) and ferritin level (from 1,522 to 940 ng/ml) and increasing lymphocyte count (from 800 to 1,700 cell/mmc) (17). Although these acute phase reactants are unreliable biomarkers for disease improvement, changes in biochemistry indicators could help us know about patients' current physical condition.

In addition to tocilizumab, a corticosteroid was used in eight of the fifteen studies (among both experiment and control groups). Since steroids had been reported to benefit patients with COVID-19 with acute respiratory distress syndrome (26), use of steroids had no influence on results of the studies (12, 14, 16, 19).

Publication bias was defined as publication or non-publication of studies depending on the direction and statistical significance of the results. Figures 4, 5 show that the funnel plots are not absolutely symmetry, which indicates some inevitable publication bias in this study.

In this meta-analysis of fifteen studies, as of February 2021, we observed that tocilizumab reduced the mortality of adult patients with COVID-19 pneumonia infection. The results indicated that early use of tocilizumab was associated with lower mortality in adults with COVID-19 pneumonia. Perhaps the relative less severity of COVID-19 infection could explain the good effect of using toxilizumab in treatment of patients with COVID-19. Lots of patients with COVID-19 suffered from severe complications, such as respiratory distress, intensive care unit (ICU) stay and even death. Current studies suggest that increased alveolar exudates caused by host immune response and inflammatory cytokine storm might contribute to mortality (27).

As observed in respiratory distress syndrome, interleukin-6 (IL-6) played a key role in inducing acute inflammation. The recombinant human monoclonal antibody tocilizumab, directed against the IL-6 receptor, mediated an inhibition strategy, preventing IL-6 binding to its receptor and, therefore, IL-6 biological activity (6).

The tocilizumab treatment showed inspiring clinical results (27). Tocilizumab is a potential recombinant monoclonal antibody against IL-6. Currently, tocilizumab is under investigation for management of patients with COVID-19 infection (28).

Previous studies have assessed the efficacy of tocilizumab in treatment of COVID-19 and showed that all-cause mortality of COVID-19 between tocilizumab and control groups was 16.3 and 24.1%, respectively. However, the difference did not have statistical significance (95% CI = 0.31–1.22). Besides, a similar risk of ICU admission was noted between the tocilizumab and control groups (35.1 vs. 15.8%, 95% CI = 0.33–6.78). From a clinical viewpoint, no clear evidence showed that tocilizumab could provide definite benefit to patients with COVID-19 infection (29).

From an evidence-based medicine viewpoint, ordinary meta-analyses on the efficacy of interventions obtain confident quality evidence from randomized controlled trials only (30–32). However, randomized trials are often not the best source of clinical evidence on harm, as the period of study is often too short to detect long-term or rare adverse outcomes (33, 34). Including observational studies in our analysis was a strength, as these studies could detect the effect of tocilizumab, which might be important in treatment of patients with COVID-19 pneumonia.

Another strength of our study was that we included most case-control studies regarding the possible therapeutic effects of tocilizumab on treatment of patients with COVID-19 pneumonia.

Several limitations should be considered when interpreting the results of this meta-analysis. First, the main limitation of the study was inclusion of only observational studies. There were not enough double-blind randomized controlled trials until February 2021. Second, research time was not long enough. Due to this urgent global crisis, nearly all studies to date have a relatively short research time. The longest research time is only more than a few months. Third, the main research subjects included in our studies were adults. Effect of the treatment on minors was still unknown. Lastly, the patients enrolled in the study were treated with different drugs and started treatment with tocilizumab in different stages of infection. Disease severity was inconsistent.

In this meta-analysis, heterogeneity may exist if sample estimates for the population risk were of different magnitudes. For existence of heterogeneity, it is crucial to assess heterogeneity in the meta-analysis. We aggregate studies that have different methodologies; however, heterogeneity in the results is still inevitable.