Prevalence of bacterial coinfection and patterns of antibiotics prescribing in patients with COVID-19: A systematic review and meta-analysis.

BACKGROUND: Evidence around prevalence of bacterial coinfection and pattern of antibiotic use in COVID-19 is controversial although high prevalence rates of bacterial coinfection have been reported in previous similar global viral respiratory pandemics. Early data on the prevalence of antibiotic prescribing in COVID-19 indicates conflicting low and high prevalence of antibiotic prescribing which challenges antimicrobial stewardship programmes and increases risk of antimicrobial resistance (AMR). AIM: To determine current prevalence of bacterial coinfection and antibiotic prescribing in COVID-19 patients. DATA SOURCE: OVID MEDLINE, OVID EMBASE, Cochrane and MedRxiv between January 2020 and June 2021. STUDY ELIGIBILITY: English language studies of laboratory-confirmed COVID-19 patients which reported (a) prevalence of bacterial coinfection and/or (b) prevalence of antibiotic prescribing with no restrictions to study designs or healthcare setting. PARTICIPANTS: Adults (aged ≥ 18 years) with RT-PCR confirmed diagnosis of COVID-19, regardless of study setting.
METHODS: Systematic review and meta-analysis. Proportion (prevalence) data was pooled using random effects meta-analysis approach; and stratified based on region and study design.
RESULTS: A total of 1058 studies were screened, of which 22, hospital-based studies were eligible, compromising 76,176 of COVID-19 patients. Pooled estimates for the prevalence of bacterial co-infection and antibiotic use were 5.62% (95% CI 2.26-10.31) and 61.77% (CI 50.95-70.90), respectively. Sub-group analysis by region demonstrated that bacterial co-infection was more prevalent in North American studies (7.89%, 95% CI 3.30-14.18).
CONCLUSION: Prevalence of bacterial coinfection in COVID-19 is low, yet prevalence of antibiotic prescribing is high, indicating the need for targeted COVID-19 antimicrobial stewardship initiatives to reduce the global threat of AMR.

1 Introduction

The first case of coronavirus disease 2019 (COVID-19) was reported in December 2019 [1, 2]. Since its emergence, the novel severe acute respiratory coronavirus 2 (SARS-CoV-2) has resulted in a global pandemic. As of January 14th 2022, a total of 318 million confirmed cases have been reported, with 5.5 million confirmed deaths [3]. The presence of bacterial co-infection in COVID-19 has been a widespread concern amongst healthcare professionals due to overlapping clinical features with bacterial pneumonia [4], and the increased risk of morbidity and mortality associated with bacterial co-infections [5]. Presence of bacterial co-infection had been observed during previous viral pandemics including the 1918 influenza pandemic and the 2009 influenza A (H1N1) pandemic [6, 7], with S. pneumoniae, β-hemolytic streptococci, H. influenzae, and S. aureus, being the most common causative pathogens of respiratory tract infections [8]. During winter months influenza-associated bacterial infections may account for up to 30% of community acquired pneumonia cases (CAP) [9]. Nevertheless, other respiratory viruses such as Middle East respiratory syndrome coronavirus (MERS-CoV) and SARS-CoV-1 have reported a very low prevalence of bacterial co-infection amongst infected patients [10, 11] potentially attributable to the comparatively small number of cases reported [12].

Concerns regarding bacterial co-infection in patients with COVID-19has led to widespread use of antibiotics empirically in both hospital and community settings [13-17]. The significant increase in antibiotic prescribing during the pandemic challenges antimicrobial stewardship programmes and risks emergence of multi-drug resistant bacteria [18-20], with their associated impact on morbidity, mortality and costs [21-24].

Prior meta-analyses suggest a bacterial coinfection prevalence of <4% - 8% in patients with COVID-19, nonetheless, these studies included a small number of patients [4, 25–27]. The prevalence of antibiotic prescribing in patients with COVID-19 was 74.6%, reported in a prior meta-analysis, which included literature mostly from Asia [28]. Consequently, this review aims at building on these publications through identifying the prevalence of bacterial co-infection, and the prevalence of antibiotic use in patients with COVID-19 across multiple countries and regions to guide future prescribing. This includes reducing the inappropriate use of antimicrobials during the COVID-19 pandemic where inappropriate use is a potential driver of antimicrobial resistance (AMR) [19, 20, 29].

2 Method

Search strategy

Electronic databases were systematically searched for published literature reporting bacterial coinfection and/or antibiotic use in patients with COVID-19. The databases searched included OVID MEDLINE, OVID EMBASE, Cochrane library and MedRxiv, with articles published between December 2019 and 29th June 2021. The search terms and keywords used included terms related to “COVID-19”, “Coinfections” and “Antibiotics” (See . The results of the search conducted were imported into Covidence online software for systematic reviews, in which duplicate publications were removed. Reporting was based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for systematic reviews. The study protocol was registered in the international register of systematic reviews, PROSPERO, under the following ID: CRD42021261734.

Study selection

Two reviewers (FA and ON) independently screened tittles and abstracts and read full texts to assess if they met the pre-set inclusion criteria, disputes were settled by third a reviewer (AK). All English language articles, irrespective of their primary outcomes, reporting bacterial coinfection rate and/or antibiotics use in, laboratory-confirmed (via Reverse transcription polymerase chain reaction (RT- PCR)), COVID-19 human adult patients (≥ 18 years) in all healthcare settings were included (Outpatients and Inpatients), neonates/children population were excluded due to potentially low prevalence of COVID-19 in the population during early waves of the pandemic with low morbidity and mortality. Studies in which patients with suspected COVID-19, based on clinical symptoms and not laboratory confirmed RT-PCR, were excluded. No restrictions to study design were applied. Case reports, case notes, editorials, letters, systematic review, meta-analysis and qualitative studies were excluded. Abstract only publications with no full text were also excluded.

Non-peer reviewed/ Pre-prints publications on MedRxiv were also included if the papers contained relevant information regarding the topic of interest. This approach of including non-peer reviewed in such meta-analysis during the COVID-19 pandemic had become common place, however, our rationale for this is to try to include larger number of patients in the meta-analysis, to address our outcome of interest.

Data extraction and quality assessment

Data was extracted into a standardised collection form that was created using Microsoft Excel 2016, by reviewers FA and ON. Data collected for information regarding the demographics of the studies included the following variables: first author; publication year; country of publication; study design (Retrospective, prospective, RCTs etc…); is the study multicentre; study setting (Community, hospital, mixed etc…); if the study was peer-reviewed; number of positive patients with COVID-19; proportion of male population; and the average age. Data was collected for the following variables: prevalence of bacterial coinfection (defined as a bacterial coinfection within 48 hours of positive COVID-19 diagnosis and hospital admission), studies looking into super-infection and/or secondary-infection (occurring at 48 hours of hospital admission), were not included; and prevalence of antibiotic use among patients with COVID-19, within first 48 hours of diagnosis. The following information, if reported, was also collected: bacterial species isolated; the prevalence of most common bacteria; most common site of infection of bacterial infection; clinical outcomes of co-infected patients; antibiotic class prescribed; timing of antibiotic initiation in relation to COVID-19 onset and clinical outcomes of patients prescribed antibiotics. The Newcastle-Ottawa Scale (NOS) was used to assess the quality of the observational studies included in the review [30].

Data synthesis, sensitivity analysis and publication bias

The two primary outcomes were the prevalence of bacterial coinfection in COVID-19 patients and the prevalence of antibiotics use in patients with COVID-19. Further sub-group analysis was conducted based on studies’ region/continent and design. Proportion (prevalence) outcome data across all studies were pooled using a random effect meta-analysis with Freeman and Tukey method [31]. Results were presented using forest plots, to demonstrate the studies’ effect size, and 95% confidence intervals (CI). Heterogeneity was assessed using I2 statistic. A value below 40% was considered to be low heterogeneity, 30–60% was considered to be moderate heterogeneity, 50–90% was substantial, and 75–100% is considerable heterogeneity [32]. Publication bias was assessed through Funnel plots followed by Egger’s asymmetry test [33]. All analyses were carried out using STATA/BE 17.0 for Windows (64-bit x 86–64) using the Metaprop command package.

3 Results

A total of 1183 studies were identified and 125 duplicates were removed. A total of 1058 studies were screened for title and abstract, 81 were screened by full-text screening and 22 studies were eligible for inclusion in the final analysis [24-45] (. Prevalence of bacterial coinfections was reported in 20 of the 22 studies included, whilst prevalence of antibiotics use was reported in 18 studies only (

Study characteristics

Retrospective cohort studies accounted for the majority of the studies involved (n = 18, 81%), whilst prospective cohort studies accounted for the remaining (n = 4, 18%). Of the 22 studies included, 3 (13%) studies were pre-prints [50, 51, 53], whilst the remaining (n = 19, 86%) were peer-reviewed. A total of 13 (59%) studies were conducted in multicentre settings, whilst the remainder (n = 9, 40%) were conducted in single centre settings. All of the studies included were conducted in hospital settings, whether it be in a normal, isolation or an intensive care ward. Twenty one (95%) out of the 22 studies have been classified as a “Good” rating during the quality assessment process .

Geographical distribution

The majority of the studies included in the review took place in the United States of America (USA) (n = 10, 45%), followed by the United Kingdom (UK) (n = 4, 18%), China (n = 3, 14%) and 1 study each in France, Germany, Indonesia, The Netherlands and Spain. Continent-wise, 10 (45%) studies were from North America, 8 (36%) from Europe and 4 (18%) were from Asia.

Patients characteristics

A total of 76,176 adult patients with RT-PCR confirmed COVID-19 were included from 22 studies, with studies by Russell et al. [44] (UK, 48,902 patients) and Puzniak et al. [34] (US, 17,003 patients) comprising 86.5% of the overall study population. The mean age of patients, was 61 years (IQR 59 67, range 36–74) and mean proportion of male subjects was 54% (IQR 50–63).

Of all the 20 studies (90%) reporting on bacterial coinfection, the most commonly reported bacterial organism was S. aureus (n = 8, 40%), followed by E.coli (n = 3, 15%). The most common source of bacterial coinfection was respiratory (n = 10, 50%), followed by blood (n = 2, 10%) and urine (n = 2, 10%).

The most commonly used class of antibiotics were the cephalosporins (8 out of 18 studies), with 7 out of 18 of the studies reporting that antimicrobial use was initiated on admission.

Meta-analysis of prevalence of bacterial coinfection in COVID-19 patients

A total of 20 studies of the 22 studies included in this review, comprising of 75,956 (99.7%) of the overall study population, investigated bacterial co-infection. Of which, only 3,645 (4.7%) patients were reported to have a confirmed diagnosis of bacterial co-infection. The random effects meta-analysis of all combined studies estimated that the prevalence of bacterial coinfection in patients with COVID-19was 5.62% (95% CI 2.26–10.31), with an I value of 99.69%, indicating considerable heterogeneity (, and an estimate of between-study variance Tau2 value of 0.15.

Meta-analysis of antibiotic use in COVID-19 patients

Antibiotic use was reported in 55,653 of the total 76,176 patients included in this review, with 18 studies (81%) reporting antibiotic use in patients with COVID-19. The random effects meta-analysis of all combined studies have estimated a prevalence of 61.16% (CI 50.95–70.90) of antibiotic prescribing in COVD-19, with an I value of 99.77%, indicating considerable heterogeneity (, and an estimate of between-study variance Tau2 value of 0.19.

Bacterial coinfection by region

The prevalence of bacterial coinfection was highest in North America (7.89%, 95% CI 3.30–14.18), followed by Asia (5.3%, 95% CI 4.03–6.73), with Europe having the lowest prevalence (3.57%, 95% CI 1.72–6) (). Heterogeneity was considerable in both North America and Europe, I = 98.89% and 96.75% respectively. Studies in Asia had low heterogeneity with an I value of 0%.

Bacterial coinfection by study design

Retrospective cohort studies had the highest prevalence of bacterial coinfection (5.92%, 95% CI 2.79–10.07), whilst prospective cohort studies had a prevalence of 3.97% (95% CI 0.38, 10.92) (). Heterogeneity was considerable in both retrospective and prospective, I = 98.88% and 98.62% respectively.

Antibiotic use by region

North America had the highest antibiotic use in patients with COVID-19per region (68.84%, 95% CI 62.27–75.05), followed by Europe (60.01%, 95% CI 25.50–89.67), with Asia having the lowest prevalence of antibiotic use (40.81%, 95% CI 7.75–79.65) (. Heterogeneity was considerable across all with studies in Europe being the most heterogeneous (I = 99.91%), followed by Asia (I = 99.18%), followed by North America (I = 97.28%).

Antibiotic use by study design

Prospective cohort studies had the highest estimate of antibiotic prescribing prevalence (77.83%, 95% CI 68.09–86.23), followed by retrospective cohort studies (56.02%, 95% CI 39.40–71.97) (. Heterogeneity was considerable in both Retrospective and Prospective cohort studies, with I value of 99.72% and 97.82%, respectively.

Bias assessment (publication bias)

As detected by the funnel plots generated (, there was no evidence of publication bias. This is further supported by the objective results (p-values) obtained through Egger’s asymmetry test for studies in both prevalence of bacterial coinfection and antibiotic use, p-values were 0.43 and 0.59, respectively.

4 Discussion

The aim of this systematic review and meta-analysis was to determine the prevalence of bacterial coinfection and antibiotic use in patients with COVID-19. The prevalence of bacterial coinfection amongst patients with COVID-19was 5.62% (95% CI 2.26–10.31), whilst, the use of antibiotic agents amongst patients with COVID-19 was 61.77% (CI 50.95–70.90). To the best of the authors’ knowledge, this is the first meta-analysis to investigate both outcomes at once as well as break the findings down by Region to provide future guidance.

With regards to bacterial coinfection in patients with COVID-19, the findings in this review are consistent with those of previously published studies and smaller systematic reviews addressing this issue (Range <4% - 8%) [4, 25–27]. Bacterial coinfection prevalence was low across all included studies, with the exception to Contou et al., Neto et al. and Puzniak et al. in which the reported prevalence rates were 28%, 19% and 16% respectively [34, 38, 40].

High prevalence rates reported by Contou et al. can be attributed to the study setting, which was the ICU. Symptomatic patients admitted to the ICU were tested for COVID-19 and for bacteriological pathogens afterwards; consequently, potentially reporting higher prevalence of bacterial coinfection. Nonetheless, Contou D et al. clearly differentiated in their study design between coinfections and nosocomial infections. Positive microbiological samples conducted within the first 48 hours of admission were labelled as coinfection, whilst positive microbiological samples after 48 hours were considered to be nosocomial ICU-acquired infections [38].

Urinary tract infections (UTIs) were the most prevalent source of bacterial coinfection (57%) as reported by Neto et al. [40]. The authors attributed the high UTI rate to the lack of a fixed defining clinical characteristics of bacterial coinfection and to high risk factors for UTIs amongst the study population, e.g. elderly hospitalised female patients and diabetic patients. It might be surprising that E coli was the second most commonly identified organism because E coli is an uncommon cause of community acquired pneumonia; this is likely to be driven by the studies including UTI among their coinfections.

High bacterial coinfection prevalence rates (16%) were reported by Puznik et al. [34], the second largest study included in this review, when compared to the low prevalence rates reported by Russell et al. [44] (0.65%), the largest study in the review. This may be due to a number of factors. These include the frequency of microbiological investigations, in which, investigation rates were higher in the study of Puznik et al. Interpretation of microbiological results in which gram-negative bacteria in sputum samples of non-ventilated patients were taken which may have over-estimated significance of bacterial coinfection [34].

The analyses conducted around bacterial coinfection in patients with COVID-19 suggests that bacterial coinfection prevalence rates are lower than seen in previous viral pandemics. During the 2009 swine flu pandemic, up to 55% of mortalities were as a result of bacterial pneumonia [46]. Previous pandemics have also reported that S. pneumoniae, β-hemolytic streptococci, H. influenzae, and S. aureus were the most commonly identified bacterial co-pathogens [8]. In this review, S. aureus has been the most identified bacterial co-pathogen.

This review also identified very high antibiotic use in patients with COVID-19, which is consistent with previous reviews including those of Langford et al. (2021) [28], which reported a prevalence of 74.6% (95% CI 68.3–80.0%). Differences between the results seen in this review and the review of Langford et al. may be attributed to the fact that the latter review also included case series with ≥10 patients. This can potentially be attributed to the time period of the pandemic in which the studies were conducted. There was scarceness of cohort studies in the review of Langford et al. (2021) [28], which is different to our study. This review also included a wider selection of nations in addition to a larger number of patients.

The increase in antibiotic use observed during this pandemic might have impacted and setback antimicrobial stewardship (AMS) efforts globally, especially in regions where AMS programmes are just starting as seen in Africa with previous knowledge and resource issues [56-58]. This is starting to change in Africa with a growing number of AMS activities to address identified concerns [59-61]. However, remarkably, in certain regions globally, specifically in Europe, there was a decline in antibiotic use overall in 2020, despite high antibiotic use in COVID-19 positive patients. This can potentially be attributed to a number of factors including social distancing measures and reduction in medical activities [62-64]. Nonetheless, inappropriate use of antibiotics during COVID-19 is a potential driver of the silent AMR pandemic [19, 65]. However, with current changes observed in global human behaviour, relating to personal hygiene, and increased interest in infection control since the emergence of this pandemic, we should see a rise in AMS activities globally [66].

Sub-group analysis based on the key regions demonstrated that the prevalence of reported bacterial coinfection was higher in North America followed by Asia and Europe at 7.89%, 5.30% and 3.57%, respectively. Antibiotic use was also higher in North America (68.84%), followed by Europe (60%) and Asia (40.81%). Our hypothesis suggests that the reason for higher prevalence of bacterial coinfection and antibiotic use in North America is due to the presence of larger number of studies and patients from the region in this review, in addition to possibly higher rates of microbiology investigation and over interpretation of microbiology results. Nevertheless, studies from Asia are reporting high use of antibiotics including the study of Hassan et al., which reported extremely high use of antibiotics (92%) in COVID-19 patients [67], however, this study was not included in our meta-analysis as it has not met our inclusion criteria. We are also aware of more recent studies in Asia reporting high rates since our analysis [17, 68].

In this review, investigating regional distribution of co-infection and antibiotic use was key. Its significance is directly correlated to the fact that antimicrobial use varies considerably across regions, albeit some convergence [69]. It is quite apparent that high antibiotic consumption is common in low- and middle-income countries (LMICs) in contrast to high-income countries (HICs) [69]. In addition, AMR rates vary considerably across countries and regions, with high AMR rates quite evident in regions such as South Asia and Sub-Sahara Africa, therefore, it was practical, in this review, to breakdown antibiotic usage rates by region [70, 71].

In terms of study design, sub-group analysis has demonstrated that retrospective studies had higher prevalence of bacterial coinfection than prospective ones at 5.92% vs 3.97% respectively. Whilst, on the other hand, antibiotic use was higher in prospective than retrospective studies, 77.83% vs 56.02%, respectively. The main hypothesis that might explain these variations in prevalence from the main meta-analyses is the study design itself. Prospective studies had well-defined processes to determine bacterial coinfection in patients with COVID-19, such as pre-defined clinical characteristics that prompt microbiological sampling [44]; hence likely lower bacterial coinfection rates but higher justifiable antibiotic use.

Despite having 10 out of 22 studies included in this review published in 2021, all the studies included have been conducted mainly in the first few months of the pandemic (February and April 2020) with the exception of one study conducted in June 2020 [55]. The results from this review demonstrates that there is insufficient evidence supporting considerable empiric antibiotic prescribing in patients with COVID-19due to a low prevalence of bacterial coinfection. Nonetheless, antibiotics use was high mirroring the findings in other reviews. As the pandemic evolves, and new COVID-19 specific therapeutics come into clinical practice, it will be important to assess their impact on antibiotic use. The early phase of the pandemic from which most of the published studies to date relate has been characterised by a lack of specific COVID-19 therapies and it may be as treatment options become available, and the understanding of the low prevalence of bacterial co-infection becomes more established, that there will be less reliance or defaulting to antibiotic prescribing. We will be following this up in future studies.

Strengths and limitations

We believe the key strengths of this review included a comprehensive search strategy spanning several databases, including both pre-prints and peer-reviewed studies, resulting in 22 studies being included, representing over 76,000 patients. In order to overcome any threats to the statistical validity of our pooled estimates (due to the nature of proportional/prevalence data including how its variance and hence study’s weight is calculated), we have used the double arcsine (freeman-Tukey) transformation in our meta-analysis as it is the recommended transformation method [72, 73]; this transformation overcomes both issues related to using the normal meta-analysis approach on untransformed prevalence data with the first issue being the problem of estimating a confidence interval that falls outside the 0–1 range, and the second issue of over-estimation of weights for studies with prevalence estimates that are at the extreme ends of zero to one [72]. We have used the metaprop command in STATA to conduct this double arcsine meta-analysis [74].

However, we are aware that this review was not without limitations. During the screening process, a significant number of studies have been excluded as they did not meet the inclusion criteria. The majority of the excluded studies included non-lab confirmed patients with COVID-19, therefore, bacterial coinfection and antibiotic use may be under- or over-reported. Disproportionate representation from North America and lack of eligible studies from regions other than Europe and Asia can also limit the generalizability of the results to other regions impacted by COVID-19, hence makes it difficult to make any conclusion about regional differences/variations; however, it is worth noting that the latter was not the objective of our study but rather we conducted a sub-group analysis by regions in order to explore the source of heterogeneity and as a sensitivity analysis to assess the sensitivity of the pooled estimates to the studies’ geographical location. Additionally, the majority of studies included were conducted within the first 6 month of pandemic. Consequently, data included might not be up to date, which again, can compromise the generalizability of the results. In addition, the majority of studies included in the meta-analyses were retrospective studies with their inherently associated bias and limitations.

Alongside this, determining the appropriateness and justifiable need of antibiotic therapy, which is likely to be higher in prospective studies in comparison to retrospective studies, was not possible, as studies have mainly reported the number of patients prescribed antibiotics. Information such as indications, initiation timing and duration of antibiotic could assist in determining future appropriateness. Diagnostic tests and measures used to determine bacteriological infections were also under-reported. This is crucial to determine whether the infection is a true infection or bacterial colonisation.

High heterogeneity was reported across all meta-analyses, which warrants caution and conservative interpretation of the results. Attempts have been made to explore the source of heterogeneity through sub-group analyses based on regions and study design, in addition, the exclusion of the two studies (Russell et al. [44] and Puzniak et al. [34]) with the highest population, all of which have yielded similar high heterogeneity. We believe this can be attributed to between -study variations such, how COVID-19 is diagnosed, definition of co-infection in each study, documenting of antibiotic use etc. Furthermore, it is worth mentioning that heterogeneity (measured by I2) is often high in proportional/prevalence meta-analysis studies representing either false heterogeneity (resulted from the nature of proportional data in that even with small sample size studies, small variance could be observed) or true heterogeneity (resulted from true differences in prevalence estimates due to variations in the time points and places where these prevalence estimates were measured in each individual studies [73]. In addition, clear overlapping of the confidence intervals can be observed, despite the sub-group analysis highlighting some difference in the point estimates, however, the overlapping between the studies demonstrates a not statistically significant result indicating consistent results (i.e., no clear difference exist) among the various sub-groups.

A clear asymmetry is observed in the funnel plot generated for antibiotic use ( which could be attributed to publication and/or heterogeneity of antibiotic use/prescription practices, nonetheless, this asymmetry was not statistically significant based on the P-value (0.59) of the Eggers test. However, results from funnel plot and Eggers test should be interpreted with cations in proportional/prevalence meta-analysis because these tests were originally developed for comparative/intervention meta-analysis data with the assumption that studies with positive results are more likely to be published compared to those with negative results but this assumption is not necessarily applicable and true for prevalence studies [73].

The inclusion of 3 (of the 22 studies included) non-peer reviewed studies might raise concerns regarding their quality [50, 51, 53]. However, two of these studies are now published, so it is unlikely to be of low quality [51, 53]. The remaining one, despite not being published, have still attained a “good” quality rating using the NOS, in addition, the study’s weight in the forest plot is small, and therefore unlikely to affect the overall results.

Future reviews and studies should aim at diversifying study regions, and to include or conduct studies that are more up to date. Studies should also include data on the appropriateness of antibiotic therapy, diagnostic tests and measures used to determine the infection. However, despite these limitations, we believe the findings give good guidance regarding the need to improve the rationality of antibiotic prescribing in patients with COVID-19 to reduce the occurrence of AMR within facilities.

Conclusion

This study demonstrates that the prevalence of bacterial coinfection amongst patients with COVID-19 was low, 5.62%, nevertheless, antibiotics use amongst COVID-19 patients was high (61.77%). However, the outcomes of this manuscript need to be interpreted with caution. Despite reporting low bacterial coinfection with the variability of the rate ranging between 2–10% amongst patients with COVID-19, when deciding to prescribe antibiotics to a patient, the difference between 2 and 10% prevalence would not be considered significant to most clinicians, and if antibiotic administration is delayed in patient with bacterial coinfection, it could result in poor prognosis. The findings of this study encourages a more rational approach to antibiotics prescribing in COVID-19 patients, an approach based on laboratory-confirmed diagnosis of coinfection, rather than clinical, advocating for more antimicrobial stewardship (AMS).

PRISMA 2020 checklist.

(DOCX)

Click here for additional data file.

PRISMA 2020 for abstracts checklist.

(DOCX)

Click here for additional data file.

(DOCX)

Click here for additional data file.

2 May 2022

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Eili Y. Klein, PhD

Academic Editor

PLOS ONE

Journal Requirements:

1. When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

3. Please amend your list of authors on the manuscript to ensure that each author is linked to an affiliation. Authors’ affiliations should reflect the institution where the work was done (if authors moved subsequently, you can also list the new affiliation stating “current affiliation:….” as necessary).

4. We note that this manuscript is a systematic review or meta-analysis; our author guidelines therefore require that you use PRISMA guidance to help improve reporting quality of this type of study. Please upload copies of the completed PRISMA checklist as Supporting Information with a file name “PRISMA checklist”.

Additional Editor Comments:

This is an interesting and timely paper, however the reviewers have raised some concerns, so I am returning this for revision. The major concerns are around the descriptions of bias and the heterogeneity in the underlying studies. One reviewer raised significant concerns about the methodology, so please justify the methods, or add additional information about the heterogeneity and consider some sub-analyses to ensure the results are consistent. Additionally, in line with the sub-analyses, greater justification of regional breakdowns is needed, it is not clear why they would be broken down that way rather than by study design. The first reviewer raises several points that are worth addressing regarding this heterogeneity in terms of how each study was performed in looking for bacterial pathogens. Adding descriptions of the methods within each study, and potential grouping by similar studies, would go a long way to address these concerns. Both reviewers raised some concerns about the biases, so please add a bit more of an explanation about the issues there. Additionally, I would suggest stratifying the results by risk of bias in an additional sub-analysis.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: No

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Prevalence of Bacterial Coinfection and Patterns of Antibiotics Prescribing in Patients with COVID-19: A Systematic Review and Meta-Analysis.

Manuscript Number: PONE-D-22-06815

Summary

The topic of this manuscript is very important. The course of the COVID-19 pandemic remains uncertain and seems increasingly likely it will continue to impact medical systems into the future. What is more certain is the pandemic of antimicrobial resistance (ref). All use of antibiotics contributes to this pandemic, particularly antibiotic use in patients without a bacterial infection. Bacterial coinfection is a real, but very difficult to diagnose complication of viral pneumonia. For this reason, clinicians struggle to determine which patients are likely to benefit from antibiotics. Establishing baseline risk factors for bacterial coinfection in COVID-19 is critical in aiding clinicians to utilize antibiotics in this population. A systematic review and meta-analysis is an excellent tool to determine that risk. The authors perform a systematic review and meta-analysis of the available literature on coinfection in COVID-19. They find that coinfection occurs in about 5% of patients within 48 hours of COVID-19 diagnosis. They also find that antibiotic prescription is over 60% in that same population. They conclude that antibiotics are over prescribed for coinfection in COVID-19.

Major Comments:

While the concept of this study is very important for clinicians and antibiotic stewards, this not unique and is somewhat behind the literature. Multiple meta-analyses and systematic reviews, including one in PLOS ONE, have already been published with findings comparable to this manuscript. In response to these earlier studies, national guidelines statements on antibiotic use in COVID-19 have been published. Currently, clinicians and stewardship programs are advocating more restricted antibiotic use in COVID-19. While these studies have been done, this manuscript is the largest meta-analysis that this reviewer has come across which would serve to cement the literature on incidence of coinfection in COVID-19, validate ongoing antimicrobial stewardship efforts, and affirm clinician’s choices to use less antibiotics in COVID-19.

The authors stratified coinfection and antibiotic use by location. I have not previously seen much literature to suggest the pathophysiology of COVID-19 and coinfection are different between regions. Some regions have different baseline population ages and prevalence of comorbidities which can affect disease severity, and indirectly, risk for coinfection or antibiotic use. What is likely different between regions is robustness of diagnostic modalities for coinfection, availability of various antibiotics, and prevalence of antimicrobial stewardship programs. While some differences exist between the regions, the general message of relatively uncommon coinfection and antibiotic over utilization remains similar between groups. The major weakness of this regional analysis is significant bias towards the United States and Europe. The data are sparce from Asian locations, and there are no studies from Africa or South America. It is difficult to make any conclusions about regional differences given large lack of data from other geographic regions.

Three terms were used in the search. All of these terms are reasonable. This reviewer wonders if they may have picked up more literature if “bacterial,” or “super-infection” were used.

Studies performed in children were not included. It is likely few studies were preformed in children given COVID-19 is less morbid in children. None-the-less, inclusion might have been useful as children are often overlooked in the medical literature and left behind by evidence based medicine.

Non-peer reviewed publications were included in this analysis. This reviewer finds inclusion of non-peer reviewed manuscripts problematic. The rapid push to describe COVID-19 and publish experiences during the pandemic has resulted in many pre-print and non-peer reviewed articles being disseminated. These premature disseminations have resulted in retractions and risk undermining public trust of the medical literature. The authors appropriately list this as a potential limitation of the study later in the manuscript.

The authors chose to only include coinfection occurring within 48 hours of COVID-19 diagnosis. This is helpful as hospital acquired pneumonia occurs after prolonged hospital exposure and is pathophysiologically distinct from community acquired pneumonia.

Diagnosing coinfection is difficult and often done heterogeneously. Diagnosis can be based on clinical factors such as oxygenation, radiography, fever, physical exam findings. Diagnosis can also be aided with laboratory and microbiologic testing, such as inflammatory markers, cultures, and PCR. This is much more difficult in the setting of a pre-existing infection, which can mimic bacterial infection. It would be nice to see a description of how each study defined coinfection. This might possibly explain the high degree of heterogeneity between coinfection rates.

Figure 2 is an easy to read figure clearly demonstrating the finding of a roughly 5% coinfection rate, ranging from 2-10%. This is the figure that most readers will remember as the principle finding of the meta-analysis. This coinfection rate is consistent with rates described by other groups. The variability in rate is understandable given the heterogeneity of the study populations and coinfection diagnosis practices. While the mathematical variability within the 95% confidence interval (2 and 10%) is real, this degree of variability is clinically smaller. When deciding to prescribe antibiotics to a patient, the difference between 2 and 10% prevalence would not be considered significant to most clinicians.

The prevailing organisms causing coinfection are known common causes of pneumonia. It is surprising that E coli was the second most commonly identified organism though. Perhaps this is driven by studies including urinary tract infection (UTI) among their coinfections. If UTIs are driving this finding, it is worth mentioning by the authors, as E coli is an uncommon cause of community acquired pneumonia.

Studying antibiotic use adjacent to coinfection rates is a natural comparison that underlies the importance of determining the true risk of coinfection in COVID-19. It also emphasizes the critical need to effectively identify patients with coinfection and appropriately administer antibiotics. This reviewer is not surprised they found considerable heterogeneity of antibiotic prescribing practices for a number of reasons. Antimicrobial stewardship is performed heterogeneously between hospitals. It is likely that antimicrobial prescription practices were lower in studies performed at hospitals with robust antimicrobial stewardship programs.

funnel plot for antibiotic prescribing practices does not have ideal symmetry. This asymmetry could be attributed to publication bias or heterogeneity of antibiotic prescription practices. A comment to this effect may be warranted.

The authors perform a brief discussion of some of the high impact or outlier studies included in the analysis. This reviewer appreciates this discussion as it lends incite in to the design of included studies. It also lends incite into the high degree of heterogeneity of coinfection literature. One major factor in potential over diagnosis of coinfection is diagnostic culture of non-sterile body sites. The authors comment directly on this challenge in regards to colonization of the urinary track in older diabetic patients as well as colonization of the airways of mechanically ventilated patients.

The authors suggest that there is insufficient evidence for high coinfection rates to justify considerable empiric antibiotic prescribing. This reviewer believes that this manuscript is unable to completely support this statement. While coinfection rates are low, there is a proportion of patients who do have coinfection. It is very likely those patients would have a much worse prognosis if antibiotic administration was delayed. This reviewer takes this manuscript as evidence we need to better identify patients who are unlikely to have a coinfection and avoid using antibiotics in them. Given there is a (small) rate of coinfection, antibiotic use will continue to be necessary in COVID-19, the key is identifying those few patients and sparing antibiotics in the rest.

This study provides the most complete and thorough review and meta-analysis of coinfection and antibiotic prescription practices in COVID-19 to date. Though, the message of the review is not dissimilar to previous systemic reviews and meta-analyses on the topic. That said, the message is abundantly clear, coinfection is a relatively uncommon complication of COVID-19, despite this, antibiotic prescription is very high. This manuscript serves to bolster the ongoing efforts to reduce unnecessary antibiotic prescribing in COVID-19. This study will no doubt serve to support further studies into the timely identification of patients with coinfection and striking an appropriate balance between antibiotic prescription and reducing morbidity and mortality in bacterial coinfections.

Minor Comments:

Line 100: Categorizing information regarding hydroxychloroquine use as “misinformation” strikes this reviewer as problematic and potentially political. There is a growing body of empirical evidence on hydroxychloroquine use in COVID-19. The evidence indicates hydroxychloroquine is not beneficial and likely harmful. Regardless, this study is not the appropriate position to determine what information should be considered “misinformation.”

Line 184: This reviewer is not a trained statistician and unable to comment on the appropriateness of the statistical methods.

Figure 1: Figure is clean and makes understanding study inclusion/exclusion simple.

Table 1: Table is easy to read and compare study similarities and differences.

Figure 3: Sub stratification by region is an interesting aspect of this analysis, this figure demonstrates some of the subtleties between the selected regions

Figure 4: stratifying the findings by study design is good, it shows small benefit of the superior prospective studies, but doesn’t change the message of the manuscript by much.

Figure 5: This figure will but just behind figure 2, in memorability for readers. It contrasts nicely with figure 2, drawing stark contrast between coinfections and antibiotic use.

Line 229: nearly 90% of the patients included in analysis were from two studies. It is likely these studies bear outsized weight on the final outcome of the analysis. The authors recognize this outsized contribution appropriately.

Line 402-406: This reviewer agrees with the hypothesis that coinfection rates are likely linked to diagnostic practice patterns at various hospitals.

Line 420-427: This reviewer agrees with the hypothesis that study design likely influenced coinfection rates owing to more rigorous coinfection definitions in prospective studies.

Reviewer #2: This manuscript was a systematic review and metanalysis (SRMA) of the prevalence of bacterial co-infections and use of antibiotics in hospitalized patients with COVID-19. Similar to prior SRMAs, the authors describe relatively low prevalence of co-infection and high antibiotic use.

The methods used in the meta-analysis need to be further examined. The authors used a random-effects model, however the within-study and between-study variance (Tau squared statistic) is not reported (both of which influence study weight). As is currently reported for all analyses in this manuscript, the weights of each included study appears to be nearly equal (all in the 4-6% range). However the authors acknowledge (lines 229-231) that two studies made up 86.5% of the total population. Therefore it is not clear how these two studies appear to be weighted nearly equally as the other included studies. Is this the correct weight distribution for these random effects models? If this is statistically accurate, mention should be made of this.

Another methodological concern is the extent to which the authors assess risk of biases in this SRMA. Publication bias was assessed using Funnel plots and Egger’s asymmetry test. The authors conclude there is no publication bias, however Figure 8B appears to be quite asymmetric with numerous studies laying outside of the funnel lines. This is confusing. It is also concerning that it does not appear that other biases were considered or assessed. For example, many SRMAs will have an entire section of the manuscript for a risk of bias assessment. If the included studies are not examined for different biases, how is it known that these biases are not reflected in the SRMA? The Cochrane Collaboration has a Risk of Bias tool, which could serve as an example. A risk of bias assessment is also part of the PRISMA checklist. Additionally many SRMAs would make their PRISMA checklist available as supplemental information, which was not done here.

All of the meta-analyses included had exceedingly high levels of heterogeneity (I2). The heterogeneity is listed in the results section, but this needs to be discussed in the discussion of the paper. It is very difficult to truly draw conclusions with such high degrees of heterogeneity. Perhaps this could speak to some of the methodological concerns addressed above, or the true heterogeneity of the included studies. Additionally, one way to explore heterogeneity may be with subgroup analyses. For example, eliminating the two studies with the highest populations (and therefore possibly the highest weights), or studies with large within-study variance.

Numerous results of sub-group analyses (retrospective vs prospective, region-specific) are discussed in the results and extensively in the discussion. However, it doesn’t appear as there are truly any clear differences, as all of the confidence intervals for the sub-groups overlap. This is not brought up in the discussion, and is also ignored when attempting to draw conclusions that differences exist.

Additional comments:

- Please spell out “COVID-19” the first time it is used (line 88)

- Numerous times throughout the manuscript the phrase “COVID-19 patients” is used. This should be replaced with “patients with COVID-19”.

- Parts of the introduction (lines 91-98) do not seem relevant to the current study, and it is detracting from the manuscript.

- Clarify in the inclusion criteria for antibiotic use, at what time this is (line 173). Is this antibiotics within the first 48 hours?

- In the methods, please clarify if trial/study characteristics and outcomes were all pre-specified

- In all figures, please better label information in the figure itself. Spell out or clarify what “ES” is (effect size?). is this just proportion? It may be more clear if labeled that way. It is also often helpful to have the population of each study included

- Figures 4 and 6, the I2 statistic and p-value are missing for the prospective cohort sub-analysis.

- Address line 329

- Consider re-wording lines 337-338 as “The aim of this systematic review and meta-analysis was to determine the prevalence of bacterial coinfection and antibiotic use in patients with COVID-19.”

- Line 344 should read “the findings in this review are consistent…”

- Line 346 should read “Bacterial coinfection prevalence was low…”

- Lines 366-368: Was this study looking at co-infection (i.e., first 48 h) or hospital-acquired/nosocomial bacterial coinfection?

- It is not clear why “new variants, updated treatment regimens, and variations in measures for SARS-CoV-2 testing” would impact bacterial co-infection (lines 456-458)

- Consider removing “so it is unlikely to be of low quality” (line 472)

- The argument in lines 472-474 does not make sense, as discussed above, all studies were weighted nearly equally.

- Consider rephrasing the paragraph in lines 476-481. A systematic review is limited by the available data published, which answers your questions developed a priori.

- The last sentence in the conclusion (lines 489-491) seems out of place and not as relevant to the findings of this study.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

Submitted filename: PLOS Coinfection Meta-analysis review.docx

Click here for additional data file.

18 Jun 2022

Reviewer #1

The major weakness of this regional analysis is significant bias towards the United States and Europe. The data are sparse from Asian locations, and there are no studies from Africa or South America. It is difficult to make any conclusions about regional differences given large lack of data from other geographic regions.

[Reply] Thank you for this comment. We do acknowledge this, however, during the data collection phase, relevant data from other regions were not available at the time being, thus further studies in lacking regions need to be conducted in to understand regional differences. Furthermore, we want to clarify that making conclusions about regional differences was not the main objective of our study but rather we conducted a sub-group analysis by regions in order to explore the source of heterogeneity and as a sensitivity analysis to assess the sensitivity of the pooled estimates to the studies’ geographical location.

We have now added a statement to the “Strengths and Limitations” section to reflect this. We hope that this is acceptable.

Three terms were used in the search. All of these terms are reasonable. This reviewer wonders if they may have picked up more literature if “bacterial,” or “super-infection” were used

[Reply] Thank you for your comment. In regards, to the use of the term “Bacterial”, we planned so start broad in our key search strategy to include as much literature as possible, therefore the term “Coinfection” was used, and later narrowed it down to bacterial infection, kindly refer to Appendix 1

In regards to not including the term “Superinfection”, we would like to clarify that we did not include superinfection because it implies that the second infection is superimposed on an earlier one that it being treated with an antibacterial agent; our systematic review meta-analysis, however, was primarily focused on co-infections occurring within 48 hours of confirmed SAR-CoV-2 infection. We hope this is OK with you.

Studies performed in children were not included. It is likely few studies were performed in children given COVID-19 is less morbid in children. None-the-less, inclusion might have been useful as children are often overlooked in the medical literature and left behind by evidence based medicine.

[Reply] Thank you for your comment. We acknowledge this, but because children and adult have different pathophysiology and could well have different antibiotic use patterns, we believe it would not have been relevant to combine these two populations together. In addition, there was a potentially low prevalence of COVID 19 in children during earlier waves of the pandemic with low morbidity and mortality. We have now added this into the paper. We accept this is beginning to change alongside equal concerns regarding the over use of antimicrobials in this population (some of the co-authors have recently published on this in Bangladesh with a study on the situation in India recently submitted). Consequently a similar study using children population is now being planned. We hope this is acceptable

Non-peer reviewed publications were included in this analysis. This reviewer finds inclusion of non-peer reviewed manuscripts problematic. The rapid push to describe COVID-19 and publish experiences during the pandemic has resulted in many pre-print and non-peer reviewed articles being disseminated. These premature disseminations have resulted in retractions and risk undermining public trust of the medical literature. The authors appropriately list this as a potential limitation of the study later in the manuscript.

[Reply] Thank you for your comment. Although at the time of the analysis only 3 of the included studies were unpublished,, now two of these have become published, the exception being Crotty MP et al 2020. Nevertheless, to address the reviewer’s comments, we have conducted a sensitivity analysis excluding the non-peer reviewed studies and the results remained consistent. We hope this is acceptable.

Furthermore, we would like to highlight that this approach of including non-peer reviewed publications in such meta-analysis during the covid-19 pandemic has been a common practice with further sub-group analysis based on peer-review status which is what we have done in our current study as well. An example of a previous systematic review/meta-analysis which used this approach is the study by Kurdi A et al, 2020 “A systematic review and meta-analysis of the use of renin angiotensin system drugs and COVID-19 clinical outcomes: What is the evidence so far?” We have now added this rationale into the Methodology, and hope this is now acceptable.

The authors chose to only include coinfection occurring within 48 hours of COVID-19 diagnosis. This is helpful as hospital acquired pneumonia occurs after prolonged hospital exposure and is pathophysiologically distinct from community acquired pneumonia

[Reply] We thank the reviewer for this valuable comment.

Figure 2 is an easy to read figure clearly demonstrating the finding of a roughly 5% coinfection rate, ranging from 2-10%. This is the figure that most readers will remember as the principle finding of the meta-analysis. This coinfection rate is consistent with rates described by other groups. The variability in rate is understandable given the heterogeneity of the study populations and coinfection diagnosis practices. While the mathematical variability within the 95% confidence interval (2 and 10%) is real, this degree of variability is clinically smaller. When deciding to prescribe antibiotics to a patient, the difference between 2 and 10% prevalence would not be considered significant to most clinicians.

[Reply] We thank the reviewer for this valuable comment, this has now been acknowledged by the authors in the conclusion section.

The prevailing organisms causing coinfection are known common causes of pneumonia. It is surprising that E coli was the second most commonly identified organism though. Perhaps this is driven by studies including urinary tract infection (UTI) among their coinfections. If UTIs are driving this finding, it is worth mentioning by the authors, as E coli is an uncommon cause of community acquired pneumonia.

[Reply] Thank you for your comment. We have now added a statement in the discussion to address the reviewer’s comment.

Funnel plot for antibiotic prescribing practices does not have ideal symmetry. This asymmetry could be attributed to publication bias or heterogeneity of antibiotic prescription practices. A comment to this effect may be warranted.

[Reply] Thank you for your comment. We have now added a paragraph in the limitations section to address the reviewer’s concerns. We hope that this is acceptable now.

The authors perform a brief discussion of some of the high impact or outlier studies included in the analysis. This reviewer appreciates this discussion as it lends incite in to the design of included studies. It also lends insight into the high degree of heterogeneity of coinfection literature. One major factor in potential over diagnosis of coinfection is diagnostic culture of non-sterile body sites. The authors comment directly on this challenge in regards to colonization of the urinary track in older diabetic patients as well as colonization of the airways of mechanically ventilated patients.

[Reply] We thank the reviewer for this valuable comment.

The authors suggest that there is insufficient evidence for high coinfection rates to justify considerable empiric antibiotic prescribing. This reviewer believes that this manuscript is unable to completely support this statement. While coinfection rates are low, there is a proportion of patients who do have coinfection. It is very likely those patients would have a much worse prognosis if antibiotic administration was delayed.

[Reply] Thank you for your comment, this has now been elaborated in the conclusion section. We hope that this is acceptable now.

This reviewer takes this manuscript as evidence we need to better identify patients who are unlikely to have a coinfection and avoid using antibiotics in them. Given there is a (small) rate of coinfection, antibiotic use will continue to be necessary in COVID-19, the key is identifying those few patients and sparing antibiotics in the rest.

[Reply] Thank you for highlighting this important point, the authors agree with the reviewer’s comment.

This study provides the most complete and thorough review and meta-analysis of coinfection and antibiotic prescription practices in COVID-19 to date. Though, the message of the review is not dissimilar to previous systemic reviews and meta-analyses on the topic. That said, the message is abundantly clear, coinfection is a relatively uncommon complication of COVID-19, despite this, antibiotic prescription is very high. This manuscript serves to bolster the ongoing efforts to reduce unnecessary antibiotic prescribing in COVID-19. This study will no doubt serve to support further studies into the timely identification of patients with coinfection and striking an appropriate balance between antibiotic prescription and reducing morbidity and mortality in bacterial coinfections.

[Reply] The authors thank the reviewer for his positive overall feedback on this manuscript.

Minor comments

[Reply] The authors thank the reviewer for his comments, all relevant comments which needs addressing have been addressed accordingly

Reviewer #2

The methods used in the meta-analysis need to be further examined. The authors used a random-effects model, however the within-study and between-study variance (Tau squared statistic) is not reported (both of which influence study weight

[Reply] Thank you for your comment, the Tau Squared Value for both meta-analysis have now been mentioned in the results section, please review lines (240-241 & 252)

As is currently reported for all analyses in this manuscript, the weights of each included study appears to be nearly equal (all in the 4-6% range). However the authors acknowledge (lines 229-231) that two studies made up 86.5% of the total population. Therefore it is not clear how these two studies appear to be weighted nearly equally as the other included studies. Is this the correct weight distribution for these random effects models? If this is statistically accurate, mention should be made of this.

[Reply] Thank you for your comment. We would like to clarify that the variance of proportional/prevalence data; hence, the given weight for each study, unlike the variance of association measures such as odds ratio, is determined not only by the sample size but also the estimated prevalence; this means that pooling prevalence data via the normal meta-analysis approach would violate the statistical validity of the pooled estimates because the variance for studies with a prevalence estimate value at the extreme ends of zero to one would be toward zero which in turn results in artificial inflation of the estimated weights for these studies in the final pooled prevalence estimate. Therefore, transformation of the prevalence data is required. Accordingly, we have used the double arcsine (freeman-Tukey) transformation in our meta-analysis as it is the recommended transformation method. Therefore, the weights of studies with large sample size might be similar to the weight of much smaller studies because even with small studies, little variance still be observed resulting is much larger weight for these smaller studies; in fact this has been observed in other published prevalence meta-analysis studies whereby the given weights for large and smaller studies were comparable as in the study by Munn, Z et. al (2015).

We have now included a statement in the discussion to reflect this. We hope that this is acceptable.

Another methodological concern is the extent to which the authors assess risk of biases in this SRMA. Publication bias was assessed using Funnel plots and Egger’s asymmetry test. The authors conclude there is no publication bias, however Figure 8B appears to be quite asymmetric with numerous studies laying outside of the funnel lines. This is confusing. It is also concerning that it does not appear that other biases were considered or assessed. For example, many SRMAs will have an entire section of the manuscript for a risk of bias assessment. If the included studies are not examined for different biases, how is it known that these biases are not reflected in the SRMA?

[Reply] Thank you for your comment, we acknowledge that the funnel plot looks asymmetrical, however this was not statically significant from the eggers test. Publication bias is bias assessment, and risk of bias is similar to quality assessment (potential confusion), therefore, table 2, has been added, table 2 illustrates the risk of bias assessment using Newcastle-Ontario Scale (NOS) that has been undertaken in this study. Furthermore, regarding Funnel plots and Egger’s asymmetry test (as we explained in our reply to reviewer one above), we would like to mention that results from funnel plot and Eggers test should be interpreted with cations in proportional/prevalence meta-analysis because these tests were originally developed for comparative/intervention meta-analysis data with the assumption that studies with positive results are more likely to be published compared to those with negative results but this assumption is not necessarily applicable and true for prevalence studies

We have now added Table 2 and a statement in the discussion to reflect these changes. We hope that this is now acceptable.

All of the meta-analyses included had exceedingly high levels of heterogeneity (I2). The heterogeneity is listed in the results section, but this needs to be discussed in the discussion of the paper. It is very difficult to truly draw conclusions with such high degrees of heterogeneity. Perhaps this could speak to some of the methodological concerns addressed above, or the true heterogeneity of the included studies. Additionally, one way to explore heterogeneity may be with subgroup analyses. For example, eliminating the two studies with the highest populations (and therefore possibly the highest weights), or studies with large within-study variance

[Reply] Thank you for your comment. In terms of the high levels of heterogeneity, we would like to mention that it is worth mentioning that heterogeneity (measured by I2) is often high in proportional/prevalence meta-analysis studies representing either false heterogeneity (resulted from the nature of proportional data in that even with small sample size studies, small variance could be observed) or true heterogeneity (resulted from true differences in prevalence estimates due to variations in the time points and places where these prevalence estimates were measured in each individual studies as well as variations in the definitions of COVID-19 diagnosis, bacterial con-infections etc. Furthermore, as recommended by the reviewer, we have conducted further sub-group analysis by excluding the two studies with the highest population and the pooled estimates were comparable

We have now revised the manuscript to reflect the above changes. We hope that this is acceptable now.

Numerous results of sub-group analyses (retrospective vs prospective, region-specific) are discussed in the results and extensively in the discussion. However, it doesn’t appear as there are truly any clear differences, as all of the confidence intervals for the sub-groups overlap. This is not brought up in the discussion, and is also ignored when attempting to draw conclusions that differences exist.

[Reply] Thank you for your comment. We have now added a paragraph in the discussion section to reflect this. We hope that this is acceptable now.

Minor Comments Sections

[Reply] The authors thank the reviewer for his comments, all relevant comments which needs addressing have been addressed accordingly, with the exception of the following:

• In all figures, please better label information in the figure itself. Spell out or clarify what “ES” is (effect size?). Is this just proportion? It may be clearer if labelled that way. It is also often helpful to have the population of each study included

[Reply] Thank you for this comment. All graphs are generated automatically by STATA, only certain components of the graph can be modified by the user, unfortunately ES (Effect Size) is automatically generated as an acronym by STATA, however, to address this concern we have spelled out ES in line (244) to make it clearer for future readers.

• Figures 4 and 6, the I2 statistic and p-value are missing for the prospective cohort sub-analysis.

[Reply] Thank you for this comment. This has now been addressed and the plots have been modified

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

19 Jul 2022

Prevalence of Bacterial Coinfection and Patterns of Antibiotics Prescribing in Patients with COVID-19: A Systematic review and Meta-Analysis

PONE-D-22-06815R1

Dear Dr. Alshaikh,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Eili Y. Klein, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

I believe that the responses were adequate and no additional revisions are needed.

Reviewers' comments:

22 Jul 2022

PONE-D-22-06815R1

Prevalence of Bacterial Coinfection and Patterns of Antibiotics Prescribing in Patients with COVID-19: A Systematic review and Meta-Analysis

Dear Dr. Alshaikh:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Eili Y. Klein

Academic Editor

PLOS ONE

Table 1

Summary of study and patients’ characteristics.

	Author, Year	Country	Study Design	Multicentre? (Y/N)	Study Setting	Peer-reviewed? (Y/N)	Quality Rating	COVID-19 Patients, n	Male, n (%)	Age (SD/IQR)	Bacterial Coinfection, n (%)	Most common Bacteria, N (%)	Antibiotic Use, N (%)	Antibiotic Class (most common)
1.	Puzniak L, 2021[34]	USA	Retrospective Cohort	Y	Hospital	Y	Good	17003	9026 (53)	61.7 (18)	2889 (16.99)	Enterobacterales, 1594 (55.17)	11562 (68)	Cephalosporins
2.	Wang L, 2020 [35]	UK	Retrospective Cohort	Y	Hospital	Y	Good	1396	903 (65)	67.4 (16.2)	37(2.65)	E. Coli, 6 (16.22)	36 (2.58)
3.	Michael S, 2020 [36]	USA	Retrospective Cohort	Y	Hospital	Y	Good	73	35 (48)				27 (36.99)	NR
4.	S. Hughes, 2020 [37]	UK	Retrospective Cohort	Y	Hospital	Y	Good	836	519 (62)	69 (55–81)	27 (3.23)	S. aureus, 4 (14.81
5.	Contou D, 2020 [38]	France	Retrospective Cohort	N	Hospital	Y	Good	92	73 (79)	61 (55–70)	26 (28.26)	S. aureus, 10 (38.46)	39 (42.39)	Cephalosporins
6.	Cheng, L, 2020 [39]	China	Retrospective Cohort	Y	Hospital	Y	Good	147	85 (58)	36 (24–54)	4 (2.72)		52 (35.37)	Cephalosporins
7.	Neto A G M,2020 [40]	USA	Retrospective Cohort	N	Hospital	Y	Good	242	123 (51)	66 (14.75)	46 (19.01)	E. Coli, 12 (26.09)	162 (66.94)	Cephalosporins
8.	Lardaro T, 2020 [41]	USA	Retrospective Cohort	Y	Hospital	Y	Good	542	269 (50)	62.8 (16.5)	20 (3.69)	S. aureus, 7 (35)
9.	Chen S, 2020 [42]	China	Retrospective Cohort	N	Hospital	Y	Good	408	196 (48)	48 (34–60)	25 (6.13)	mycoplasma pneumonia, 13 (52)	60 (14.71)	NR
10.	Baskar V, 2021 [43]	UK	Retrospective Cohort	Y	Hospital	Y	Good	254	164 (65)	59 (49–69)	14 (5.51)	S. aureus, 4 (28.57)
11.	Russell C D, 2021 [44]	UK	Prospective Cohort	Y	Hospital	Y	Good	48902	28116 (58)	74 (59–84)	318 (0.65)	E. Coli, 64 (20.13)	39528 (80.83)	Penicillin/B-lactams
12.	Lehmann C J,2021 [45]	USA	Retrospective Cohort	N	Hospital	Y	Poor	321	155 (48)	60 (17)	7 (2.18)	S. aureus, 2 (28.57)	222 (69.16)	NR
13.	Vaughn V,2021 [46]	USA	Retrospective Cohort	Y	Hospital	Y	Good	1705	885 (52)	64.7 (53–77)	59 (3.46)		965 (56.6)	Cephalosporins
14.	Miao Q, 2021 [47]	China	Retrospective Cohort	N	Hospital	Y	Good	323			17 (5.26)	Klebsiella pneumonia, 11 (64.71)
15.	Karami Z, 2020 [48]	Netherlands	Retrospective Cohort	Y	Hospital	Y	Good	925	591 (64)	70 (59–77)	7 (0.76)	S. aureus, 4 (57.14)	556 (60.11)	Cephalosporins
16.	Garcia-Vidal C, 2021 [49]	Spain	Retrospective Cohort	N	Hospital	Y	Good	989	552 (56)	62 (48–74)	31 (3.13)	Streptococcus pneumonia, 12 (38.71)	917 (92.72)	Macrolide
17.	Crotty M P, 2020 [50]	USA	Prospective Cohort	Y	Hospital	N	Good	289		58.6 (14.4)	25 (8.65)	S. aureus, 5 (20)	271 (93.77)	NR
18.	Wei W, 2020 [51]	USA	Prospective Cohort	N	Hospital	N	Good	147	87 (59)	52			87 (59.18)	Cephalosporins
19.	Karaba S, 2020 [52]	USA	Prospective Cohort	Y	Hospital	Y	Good	1016	543 (53)	62 (48–74)	53 (5.22)		717 (70.57)	NR
20.	Martin A, 2020 [53]	USA	Retrospective Cohort	Y	Hospital	N	Good	208	105 (51)	69 (60–80)	24 (11.54)	S. aureus, 5 (20.83)	172 (82.69)	Cephalosporins
21.	Rothe K, 2021 [54]	Germany	Retrospective Cohort	N	Hospital	Y	Good	140	90 (64)	63.5 (17–99)	3 (2.14)		116 (82.86)	Penicillin/B-lactams
22.	Asmarawati T P, 2020 [55]	Indonesia	Retrospective Cohort	N	Hospital	Y	Good	218	120 (55)	52.45 (14.44)	13 (5.96)	NR	164 (75.23)	Quinolones

Table 2

Risk of bias assessment using Newcastle-Ottawa scale (NOS).

	Author, Year	Score per Domain			Quality Rating
		Selection	Comparability	Outcome
1.	Puzniak L, 2021	4	2	2	Good
2.	Wang L, 2020	4	2	2	Good
3.	Michael S, 2020	4	2	2	Good
4.	S. Hughes, 2020	4	2	2	Good
5.	Contou D, 2020	4	2	2	Good
6.	Cheng, L, 2020	4	2	2	Good
7.	Neto A G M,2020	4	2	2	Good
8.	Lardaro T, 2020	4	2	2	Good
9.	Chen S, 2020	4	2	2	Good
10.	Baskar V, 2021	4	2	2	Good
11.	Russell C D, 2021	4	1	1	Poor
12.	Lehmann C J,2021	4	2	2	Good
13.	Vaughn V,2021	4	2	2	Good
14.	Miao Q, 2021	4	2	2	Good
15.	Karami Z, 2020	4	2	2	Good
16.	Garcia-Vidal C, 2021	4	2	2	Good
17.	Crotty M P, 2020	4	2	2	Good
18.	Wei W, 2020	4	2	2	Good
19.	Karaba S, 2020	4	2	2	Good
20.	Martin A, 2020	4	2	2	Good
21.	Rothe K, 2021	4	2	2	Good
22.	Asmarawati T P, 2020	4	2	2	Good