gms | German Medical Science

The morbidity and mortality rate associated with COVID-19 is not unrelated to co-infections with bacterial pathogens [1]. Respiratory and blood culture studies of hospitalized patients with severe acute respiratory coronavirus 2 (SARS-CoV-2) have shown that bacterial infections rather than the direct effects of the virus have resulted in a number of recorded fatalities [2].

Staphylococcus aureus (S. aureus) is persistently and asymptomatically present in the nares of 20% of the human population [3], [4]. In response to physiological changes during infection, bacterial factors involved in the shift of S. aureus from commensalism to pathogenesis is poorly understood [5]. The bacteria have been associated with secondary staphylococcal pneumonia following COVID-19 infection [6], [7]. However, the overlap of symptoms makes the identification of co-infected patients and the co-infecting pathogens laborious [8].

Although previous studies investigated the frequency of selected bacterial species in COVID-19 cases [9] and the overall extent of co-infection [10], the pooled rate of S. aureus among hospitalized COVID-19 patients is largely undocumented. This study aims to address this issue by conducting a meta-analysis to determine the burden of S. aureus co-infections in patients hospitalized with COVID-19. Knowledge about specific etiological agents may reduce the strain on the resources of healthcare systems worldwide and lead to more appropriate treatment and medication, as well as shorter hospitalization.

We examined databases for studies that reported data on S. aureus and MRSA co-infections in patients with COVID-19 infection. Studies with fewer than 10 participants and case studies were excluded. Searches were performed in Pub Med, Google Scholar, Web of Science and SCOPUS from 1^st of January 2020 up to 20^th of October 2020. The search terms included: ‘COVID-19 and MRSA’, ‘bacterial infection and MRSA’, ‘S. aureus and COVID-19’, ‘SARS-CoV-2 and MRSA or S. aureus’ ‘bacterial pathogens and COVID-19’. These were combined with search terms such as ‘hospital’, ‘healthcare’, ‘community-acquired’, ‘hospital-acquired’, ‘bacteremia’, ‘pneumonia’, ‘secondary infections’, ‘supra-infection’, ‘co-infection’. The bibliographies of identifed articles were also searched. The abstracts and full texts of identified studies were screened for eligibility by two reviewers (SSA and ABS). The quality of studies was evaluated using the Joanna Briggs Institute Checklist for Studies Reporting Prevalence Data [11]. Discrepancies in evaluation were settled by consultation with a mediator (SMG). PRISMA (preferred reporting items for systematic reviews and meta-analyses) protocols were used for this analysis [12].

The data gathered from the included studies comprised author’s name, country of study, type of study, setting, culture type, and number of patients with: influenza-like illness (ILI), COVID-19 positive results, co-infections, S. aureus co-infections or MRSA co-infections.

Random effects meta-analysis (REM) was performed because the studies included were sampled from a universe of different populations and high heterogeneity was anticipated. Comprehensive Meta-Analysis^® software, version 3.3070 (Bio-stat, Englewood, NJ, USA), was used to perform the analysis. Using the Cochran’s Q statistic, the observed dispersion (heterogeneity) among effect sizes was assessed. The percentage of total variability in the estimates of the effect size was calculated with the I² index. To check for publication bias, Egger weighted regression and Begg rank correlation methods with a meta-funnel plot were used. A p-value <0.05 was presumed to reflect a statistically significant publication bias. In order to structure the search results and document relevant studies, Zotero desktop^® (version 5.0.92) reference software was employed. Sensitivity analyses were carried out to gauge the impact of each study (by omission) on the pooled rates [13].

Meta-regression analysis was conducted to evaluate the variability between our outcomes and the covariates (study type, study quality, setting and country). Covariates were first tested individually in a univariate analysis and then simultaneously in a multiple meta-regression model through the computational options “methods of moments” followed by the “maximum likelihood” ratio. The R² analog was used to compute the total variance of all studies about the grand mean effects. Outlier diagnostics were performed using Cook’s distances, covariance ratios, heterogeneity test statistics and weights (Attachment Fig. S4).

Our search yielded 207 titles after removing duplicates, of which 148 were removed during the initial screening. At this point, the abstracts of the outstanding 59 studies were reviewed, and 41 studies were discarded because they did not meet the inclusion criteria, leaving 18 studies included in the study (Figure 1 [Fig. 1]). The study quality ranged from 6 to 9 (median 7.1) on the Joanna Briggs Institute scale. All included studies reported on S. aureus/COVID-19 co-infection amongst hospitalized patients. Table 1 [Tab. 1] provides data on 63,370 patients from the included studies. Five studies were from the United States (27.8%), 2 each from the United Kingdom, China, and Italy (33.3%), 1 each from France, Egypt, Saudi Arabia, Netherlands, Spain, Iran and Russia (38.9%). We retrieved data for the 63,370 patients hospitalized with influenza-like illness, about 14,369 (22.67%) of which tested positive for COVID-19 by rRT-PCR. Of this number, 8,249 (57.4%) of patient samples (respiratory and blood) were analyzed for co-infecting pathogens. Bacterial, fungal and viral agents were detected in 3,038 (36.8%) patients, and S. aureus in 1,192 (39.2%) patients. Only five studies reported MRSA co-infection in 66 patients.

The Forest plots (Figure 2 [Fig. 2], Figure 3 [Fig. 3], and Figure 4 [Fig. 4]) show the pooled rates of co-infections, S. aureus/COVID- 19 co-infection and frequency of MRSA among co-infected patients. Subgroup analysis was performed to evaluate whether the pooled effects differed for mono-center and multicenter studies. The red and blue summary symbols represent the overall and subgroup effect outcome of the analysis, respectively, at a confidence interval of 95%. The squares signify the point estimates of each study and the square's size denotes the weight given in the meta-analysis. From pooled analysis of 18 studies, 33.1% of patients reported co-infection (95%, CI 18.0 to 52.6%, Q=3473: df=17, I²=99·48%, p=0.00): mono-center, 30.4% (95% CI, 12.0 to 58.0%); multicenter, 35.7% (95% CI, 15.3 to 63.2%). Additionally, the pooled rate of S. aureus/COVID-19 co-infection among patients was 25.6% (95% CI: 15.6 to 39.0, Q=783.4, df=17, I²=97.702%, p=0.003): mono-center, 24.5% (95% CI, 12.2 to 43.2%, p=0.010); multicenter, 26.8% (95% CI, 12.9 to 47.4%, p=0.029). The overall pooled proportion of hospitalized COVID-19/S. aureus co-infected patients with MRSA was 53.9% (95% CI, 24.5 to 80.9, n=66, 5 studies, Q=29.32, df=4, I²=86.369%, p=0.000). The sensitivity analysis did not significantly affect the overall proportion of our results by excluding one study, nor did it affect the heterogeneity. Some indication of publication bias among the analysed studies is shown by the asymmetric distribution of the studies in the funnel plots (Attachement Fig. S1, Fig. S2, and Fig. S3 [Attach. 1]), highlighting the statistical heterogeneity observed. Egger’s and Begg’s tests (Attachement Tab. S1 [Attach. 1]) did not demonstrate statistical significance for bias in any of the analyses (p>0.05). None of our covariates were statistically significant using the “methods of moment’s” computation method. In the multivariate meta-regression model (maximum likelihood method), study type (p=0.029), study quality (p=0.000) and country of study (p=0.000) were significantly associated with heterogeneity of results (Attachement Fig. S4 [Attach. 1]). Of all the moderator variables, only study settings was not statistically significant (0.123). Overall, at a Q-value of 53.79, with df=13 and p=0.000, the covariates were associated with our observed effect. The variance of true effect sizes at any point on the regression line (T²) was 0.2048, p≤0.05. Only 83% (R²=0.83) of the variance in true effects can be explained by the covariates using our model (Figure 5 [Fig. 5]).

Normally, an underlying infection is expressed as symptoms. Traditional approaches (qualitative and quantitative) for detecting co-infections are not always effective due to overlapping symptoms. As a result, clinicians prefer empirical antibiotic therapy with an emphasis on etiological staphylococci and streptococcal agents [14], [15]. Although the danger posed by bacterial co-infections in COVID-19 patients is recognized, the extent of co-infection with S. aureus has hitherto not been systematically evaluated. This meta-analysis found that overall, in the included studies, more than one-fourth of COVID-19 hospitalized patients had a co-infection (bacterial, fungal or viral), underscoring the need for establishment of protocols for the detection of coinfection to improve clinical data and patient therapy. Similarly, in about one-fourth of recorded co-infections, S. aureus was the prevalent co-pathogen. This finding is consistent with a coinfection rate of 25% (S .aureus) previously reported by [16]. The finding that MRSA was associated with over half of patients hospitalized with COVID-19/S. aureus co-infection is consistent with the prevalence rate of 50% reported by [16] in the 2009 influenza pandemic. The rates observed may be attributable to widespread antibiotic use on skin and nasopharyngeal microbiota, which may degrade the respective ecosystem [17]. The reported MRSA rates may be linked to non-judicious administration of broad-spectrum antibiotics to a large proportion of patients. The sensitivity of S. aureus culture methods could have been limited by the excessive use of antibiotics, so that our findings may have underestimated the actual situation.

In our meta-analysis, significant heterogeneity exists, which led us to an enquiry into its origin. The maximum likelihood model explained that covariates such as study type, study quality and country of study were associated with heterogeneity. The unexplained heterogeneity (20.48%) may be due to differences between studies in terms of disease severity, patient co-morbidities, treatment differences, use of antibiotics prior to and during hospitalization, or other unidentifed covariates.

The strengths of the present study include our use of statistical models to assess the sources of heterogeneity, a systematic search strategy to classify potentially suitable studies from different sources, as well as scrutinizing the supplementary information of preprints and publications up to our search date. While research is ongoing, there are few studies documenting S. aureus and MRSA microbiological cultures among COVID-19 positive patients to date. This influenced the distribution of the studies covered. It is also likely that, considering the extraordinary conditions and immense burden on healthcare systems, patients with a suspected secondary infection would not have had extensive microbiological examinations. The data provided by the included studies did not distinguish between the sources of secondary infections and colonizers. This research focused solely on patients who were hospitalized and did not take into account patients who had not been hospitalized.

The pooled rates of S. aureus among COVID-19 patients documented in this study justify the concern of clinicians about the presence of S. aureus in co-infections. This data is not sufficient to support widespread- and often inappropriate empirical use of antibiotics in patients hospitalized with COVID-19, as reports of co-infection in admitted patients are scanty. Improved antibiotic stewardship can be accomplished through rapid diagnosis by longitudinal sampling of patients to allow targeted antimicrobial therapy.

The authors declare that they have no competing interests.

We thank Halima Salihu of the Department of Fisheries and Aquaculture, Usmanu Danfodiyo University Sokoto, Nigeria for her supportive remarks.

None was received.