Proportions of Staphylococcus aureus and Methicillin-Resistant Staphylococcus aureus in Patients with Surgical Site Infections in Mainland China: A Systematic Review and Meta-Analysis

Background Sufficient details have not been specified for the epidemiological characteristics of Staphylococcus aureus (S. aureus) and methicillin-resistant Staphylococcus aureus (MRSA) among surgical site infections (SSIs) in mainland China. This systematic review aimed to estimate proportions of S. aureus and MRSA in SSIs through available published studies. Methods PubMed, Embase and four Chinese electronic databases were searched to identify relevant primary studies published between 2007 and 2012. Meta-analysis was conducted on the basis of logit-transformed metric for proportions of S. aureus and MRSA, followed by pre-defined subgroup meta-analysis. Random-effects meta-regression was also conducted to explore the impact of possible factors on S. aureus proportions. Results 106 studies were included, of which 38 studies involved MRSA. S. aureus accounted for 19.1% (95%CI 17.2-21.0%; I2 = 84.1%) of all isolates in SSIs, which was roughly parallel to 18.5% in the United States (US) (P-value = 0.57) but significantly exceeded those calculated through the surveillance system in China (P-value<0.001). In subgroup analysis, S. aureus in patients with thoracic surgery (41.1%, 95%CI 26.3-57.7%; I2 = 74.4%) was more common than in those with gynecologic surgery (20.1%, 95%CI 15.6-25.6%; I2 = 33.0%) or abdominal surgery (13.8%, 95%CI 10.3-18.4%; I2 = 70.0%). Similar results were found in meta-regression. MRSA accounted for 41.3% (95%CI 36.5-46.3%; I2 = 64.6%) of S. aureus, significantly lower than that in the US (P-value = 0.001). MRSA was sensitive to vancomycin (522/522) and linezolid (93/94), while 79.9% (95%CI 67.4-88.4%; I2 = 0%) and 92.0% (95%CI 80.2-97.0%; I2 = 0%) of MRSA was resistant to clindamycin and erythromycin respectively. Conclusion The overall proportion of S. aureus among SSIs in China was similar to that in the US but seemed higher than those reported through the Chinese national surveillance system. Proportions of S. aureus SSIs may vary with different surgery types. Commonly seen in SSIs, MRSA tended to be highly sensitive to vancomycin and linezolid but mostly resistant to clindamycin and erythromycin.


Introduction
Competing Interests: This study was partly supported by MedImmune LLC, a wholly owned subsidiary of AstraZeneca, mainly for literature search and data collection. The funding was received directly by Peking University Health Science Center. None was paid individually to the researchers in this study for any employment or consultancy for the company. Antibiotics discussed in this review were not the patents of the company or its products in development or on market. There is no further relevant information about the conflicts of interest to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors.
concern from the perspective of clinical practice, are not accessible in the studies. In addition, this system only provided the overall proportion of S. aureus rather than proportions by year, region, hospital level, and surgery type, which are likely to have more significant impact on decision-making for clinical practice and public health.
Understanding the nationwide epidemiological situation of S. aureus and MRSA in SSIs is vital for policy makers and clinicians to develop appropriate preventive countermeasures. As the national data in China remain inadequate, this systematic review aimed to estimate the proportions of S. aureus and MRSA in SSIs, by summarizing and assessing the available observational studies in China published from 2007 to 2012, to provide further evidence.

Information sources and search strategy
We performed systematic search in six electronic databases, including PubMed, Embase (OVID), Chinese BioMedical Database (CBM), China National Knowledge Infrastructure (CNKI), VIP Chinese Science and Technique Journals Database, and Wanfang Database, to identify the relevant studies. Since the focus in the review was on the epidemiological characteristics of S. aureus and MRSA in SSIs during recent years, search was limited to the publication date from January 2007 to November 2012. A combination of Mesh words and free text words applied to PubMed, Embase and CBM, and free text words were used to search CNKI, VIP and Wanfang database. The following search terms were mainly used: "surgery", "wound infectionÃ", "postoperative wound infectionÃ", "surgical site infectionÃ", "S. aureus.", "Staphylococcus aureus", "methicillin", "MSSA" and "MRSA". Details of the search strategies for each database were summarized in S1 Table. Eligibility criteria Criteria of inclusion: • Patients: those with SSIs regardless of other characteristics; • Outcomes: S. aureus and MRSA isolates identified from SSIs; • Study types: observational studies including cross-sectional, monitoring, prospective, ambispective and retrospective study.

Criteria of exclusion:
• Duplicate studies; • Involvement of studied population from outside mainland China; • Therapeutic study including randomized controlled trial and observational research for comparative effectiveness; • Studies with data from the China Nosocomial Infection Surveillance System.

Study selection
According to the criteria of inclusion and exclusion, two reviewers independently screened each record by the title, keywords and abstract. The eligibility was determined further through the full texts if selection cannot be made only based on the screening. Any disagreement was resolved by the third reviewer.

Data extraction
An original extraction form was designed and then modified following a pilot test. The revised extraction form encompassed three parts: general information, clinical characteristics and numbers for calculating proportions of S. aureus and MRSA isolates. Two reviewers extracted information from each study independently. Any disagreement was also resolved by the third reviewer.

Assessment of risk of bias
As there were no acknowledged or standardized quality assessment tools for the included study designs, we used a checklist with 8 items adapted from a scale for case series [15], which was originally developed by the National Institute for Health and Care Excellence (NICE), a special health authority in the UK which is committed to providing national guidance and advice to improve health and social care. Low, high or unclear risk of bias for each item was determined according to the pre-specified criteria (S2 Table) and the graph of summary of risk of bias was developed with Revman 5.1. One point was scored if an item was judged low risk of bias. We defined study of higher quality with a total of at least 4 points.

Dealing with missing data
When information of the variables for analysis was missing from publications, the correspondent authors were contacted by email every one week. If the authors did not reply to the emails after our second contact attempt, the publications were excluded when the related variables were analyzed.

Statistical analysis
We conducted all the data analyses using R (Version 3. the China Nosocomial Infection Surveillance System. Statistical difference between the proportions in such comparisons was tested by Q statistic for heterogeneity [16]. P-value of less than 0.05 indicated statistical significance. Considering probable heterogeneity across all the observational studies, random-effects model with Der-Simonian Laird method was used a priori throughout the data analyses. Heterogeneity and subgroup analysis Q test and I 2 statistic were used to examine and quantify the heterogeneity of the logittransformed proportion across the studies. P-value of less than 0.05 or I 2 statistic of more than 50% were regarded as substantial heterogeneity [17]. Subgroup analysis was conducted to explore the possible sources of heterogeneity based on the pre-defined variables including study quality, sample size, region, level of hospital, provincial economic condition, types of surgeries. A map for the distribution of S. aureus was drawn through MapInfo Professional 11.0 according to the subgroup analysis by provinces. We determined small sample size if at most 20 bacteria isolates or S.aureus isolates were included in analysis for primary studies respectively reporting the proportion of S. aureus or MRSA. Based on whether the annual Gross Domestic Product (GDP) per capita of each province in 2011 was higher or lower than the national average (35,181RMB) in China, provinces were categorized into higher or lower provincial economic condition [18]. Informal comparisons were made between subgroups for the proportions of S. aureus and MRSA by directly comparing the magnitudes of proportions between different subgroups instead of significance tests which tend to be misleading for the comparison in subgroup analysis. Statistical significance was defined as non-overlap of the confidence intervals of the proportions between the subgroups [19].

Meta-regression for the proportion of S. aureus isolates
Meta-regression was used to explore the impact of pre-defined factors on the proportion of S. aureus isolates. We defined logit(P) as the dependent variable where P referred to proportion of S. aureus isolates. All the independent factors were initially selected based on the expertise in clinical microbiology and the availability of related information in the included articles, including study quality, sample size, region, level of hospital, provincial economic condition and type of surgery, all of which were defined as dummy variables. The factors without colinearity indicated by no correlation to each other (P-value!0.10) were finally included into the randomeffects meta-regression model with restricted maximum likelihood (REML) method. The statistical significance of any single coefficient was tested by Z-test and 0.05 was used as the threshold of P-value for statistically significant difference.

Publication bias
Egger's test served to assess the probability of publication bias for the overall S. aureus and MRSA proportion [20]. The test was based on the logit-transformed proportion and corresponding standard error. A P-value of less than 0.10 was regarded as statistical significance, indicating probable publication bias.

General information about included studies
We retrieved 2904 references from six databases, of which 106 studies were eligible for inclusion ( Fig. 1). All the studies, 105 published in Chinese and one in English, were hospital-based.

Methodological quality of studies
The methodological quality of studies was displayed in Fig. 2 with more details in S2 Table, S1 Fig The maximum score that studies achieved was 7 while the minimum was 0. We finally identified 38 studies with relatively high quality which reached at least 4 scores in our quality assessment scale.

Subgroup analysis
All the results of subgroup analysis were summarized in Table 3 and the forest plots were presented in S1 File.
The pooled proportion was 41 In addition, S. aureus proportion was higher in studies conducted in low economic condition, rural or non-tertiary hospitals or with small sample size (at most 20 isolates of bacteria), although significant differences between subgroups were not found. On the other hand, the proportions seemed similar in studies with high and low quality, those with retrospective and non-retrospective design, or those beginning before and since 2007 (Fig. A-I in S1 File).
Geographical differences in S. aureus proportions by different provinces or municipalities across China were shown in Fig. 5. Among 21 areas the maximum point estimate of S. aureus proportion among all the provinces with available data was 33.3% (95%CI 15.8-57.1%) in Ningxia province, followed by Tianjin municipality (30.2%, 95%CI 21.9-40.1%) and Jiangxi province (30.0%, 95%CI 16.9-47.4%) and the minimum was 11.5% (95%CI 8.1-16.1%) in Gansu province. However, there was only one study available, respectively, for the proportion estimate in Ningxia, Jiangxi and Gansu.
Regarding the pooled proportion of MRSA isolates (Fig. J-R in S1 File), the proportion was 55.0% (95%CI 21.4-84.5%, I 2 = 74.1%) for abdominal surgeries, 41.0% (95%CI 23.5-61.1%, I 2 = 0%) for gynecologic surgeries, 39.1% (95%CI 3.8-91.2%, I 2 = 94.0%) for thoracic surgeries and 26.6% (95%CI 15.3-42.2%, I 2 = 56.9%) for orthopedics surgeries. Furthermore, despite insignificant difference between subgroups, MRSA proportion tended to be higher in low economic condition, urban and tertiary hospitals as well as in studies with small sample size (at most 20 S. aureus isolates). Similar proportions can be found between studies with higher and lower quality, studies with retrospective and non-retrospective design, or studies with the start time of before 2007 and since 2007.  Meta-regression for the proportion of S. aureus isolates 97 studies without any missing data were included for meta-regression to identify related potential factors for heterogeneity with statistical significance. As we found a significant correlation coefficient between levels of hospital and region (coefficient = 0.570, P-value<0.001), provincial economic condition and region (coefficient=-0.198, P-value = 0.052), the region variable was therefore excluded out of the pre-defined independent factors for the metaregression ( Table 4).

Proportions of S. aureus
The overall proportion of S. aureus isolates (19.1%) was consistent with the reported proportion of 18.5% in the US in 2003 [21], but was significantly higher than both estimated in the China Nosocomial Infection Surveillance System (12.7% between 1999 and 2001, 13.5% between 1999 and 2005) [11,12]. This difference between our review and the Chinese   including tertiary and non-tertiary hospitals, urban and rural areas, which may be more representative of the real national status. Second, the sample size of bacteria isolates from the surveillance system (only 3,812 in total) was substantially less than that included in our review (13,608). Pooling data from 106 studies with a larger sample size may provide a more reliable estimate for national situation.
In addition, our finding can provide further useful information which was not available from the Chinese surveillance system, such as the stratified proportions by the surgery type, economics condition, hospital level and province. As shown in the subgroup analysis, S. aureus proportions varied between different surgery types-highest for thoracic surgeries (41.1%) whereas lowest for abdominal surgeries (13.8%). Meta-regression suggested the similar result * Others refer to: 1) multiple surgeries involved in the study which cannot be classified into a specific type of surgery or 2) a specific type of surgery, rather than orthopedic, abdominal, gynecologic, or thoracic surgeries, which was reported in a small number of studies. ** Non-retrospective design comprises prospective, cross-sectional, ambispective study and surveillance. that patients undergoing thoracic surgeries were much more vulnerable to SSIs by S. aureus, compared with patients with any other involved surgery type. This result was consistent with the guideline for prevention of SSIs, which concluded that S. aureus was the dominant pathogen causing SSIs following thoracic surgeries [8], indicating that S. aureus should be highly suspected in the case of SSIs after thoracic surgeries. On the contrary, with other influencing factors adjusted, patients with abdominal surgeries were less likely to suffer from SSIs by S. aureus. Priority may not be given to S. aureus in the SSIs after this surgery type because gram-negative bacilli, rather than S. aureus, tend to be predominant in the gastrointestinal tract usually involved in abdominal surgeries [22,23]. Impoverished regions, non-tertiary hospitals, and some provinces or municipalities such as Ningxia, Tianjin and Jiangxi, may also require more attention paid to S. aureus SSIs.

Proportions of MRSA
Our estimate focused on SSIs and thus may close the gap of the surveillance system in China which merely reported the MRSA proportion of 79.9% (3,177/3,975) in all kinds of hospital infections instead of in SSIs [12]. Comparison between the results indicated that the proportion of MRSA in SSIs may be lower than the average level among a diversity of hospital infections. Besides, our review concluded a significantly lower proportion than that reported in a recent multi-center study with a smaller sample size in the US [24]. However, the status quo necessitates further improvement since MRSA accounted for more than 40% of S. aureus in our review. Variation in the MRSA proportions was found between different surgery types: highest in abdominal surgeries (55.0%) and lowest in orthopedics surgeries (26.6%). While a recent study showed cases with MRSA SSIs accounted for 30.4% in those with S. aureus SSIs in the US [22], the high MRSA proportion (55.0%) following abdominal surgeries in our study provided an alarming picture that, despite S. aureus being subordinate pathogen in SSIs after abdominal surgeries, physicians still have to be highly cautious about MRSA in SSIs. On the contrary, orthopedic surgeries saw the lowest proportion of MRSA SSIs (26.6%) in spite of its high proportion of S. aureus. A study also concluded that the proportion of MRSA was the lowest in orthopedic surgeries among all the surgical procedures, although the proportion (31.9%) they calculated was higher than ours [10]. However, the mechanism seems still unclear and requires further studies to confirm.

Proportions of antibiotic-resistant MRSA
Based on our findings, vancomycin and linezolid appeared to be still effective for treating MRSA in SSIs in vitro. Vancomycin therapy is the primary option in the case of limited current therapeutic methods for patients with MRSA infections [25]. In China, the surveillance system suggested that none of MRSA were resistant to vancomycin (0/3,102) between 1999 and 2005 in a variety of nosocomial infections including SSIs [12], which was similar to our result: we identified none of the MRSA isolates resistant to vancomycin (0/522) in SSIs. However, it is necessary to raise the awareness of the resistance of vancomycin since there has been evidence suggesting the observed rise in minimum inhibitory concentrations (MICs) of vancomycin from less than 0.5μg/mL in 2005 to 1.0 μg/mL in 2010 [26]. Linezolid is the first available oxazolidinone antibiotic, which uniquely inhibits bacterial protein synthesis by preventing formation of 70S initiation complex [27]. Although surveillance data on linezolid was absent, one of 94 MRSA isolates in our findings was resistant to linezolid. But currently no robust clinical evidence can demonstrate whether linezolid or vancomycin is superior in the treatment of MRSA SSIs [28]. Continuous surveillance of drug resistance for both antibiotics in this treatment is necessary and crucial for the clinical practice.
On the other hand, clindamycin and erythromycin, inhibiting protein synthesis by their effect on ribosome function and commonly used in clinical practice for MRSA SSIs [27,29], may have a doubtful effectiveness. The proportion of MRSA resistant to clindamycin in our findings (79.9%) was also similarly suggested in the surveillance system for nosocomial infections-78.9% (1,137/1,445) [12]. In our review, more than 90.0% MRSA isolates were identified to be resistant to erythromycin, far higher than that in the UK bacteraemia surveillance where erythromycin resistance only occurred in 67% of MRSA [30]. As such, both treatments may not be the first choice when MRSA in SSIs is suspected.

Limitations
There are several limitations in this review. First, methodological quality of the included studies is the main concern for the combined estimates because less than half of the studies are of high quality according to our criteria. However, study quality seems not to be the main heterogeneity source as both subgroup analysis and meta-regression showed that the pooled result from studies of higher quality were consistent with that from those of lower quality. Second, we only included studies published after 2007 so as to understand the current proportion of S. aureus and MRSA in SSIs. Considering some studies had started before 2007, we conducted subgroup analysis for studies initiating before and after 2007 to ensure that it was reasonable to combine results from all included studies to provide more precise estimates and facilitate the metaregression. Third, none of the pre-defined variables can fully explain the variance in proportions of S. aureus (I 2 = 84.1%) and MRSA (I 2 = 64.6%) in subgroup analysis and meta-regression, which could result in uncertainty around the pooled proportions. The major obstacle of extensively exploring the potential source of variation is the limited information about the heterogeneity reported in the publication, such as the duration of surveillance, MICs and molecular epidemiology, which may be significantly associated with the heterogeneity but cannot be examined in our review. However, meta-regression did find that some factors with available information, such as the types of surgery and sample sizes, may partly contribute to the heterogeneity across studies. In addition, despite informal comparisons between subgroups by 95%CI rather than the significance test, the problem with multiple comparisons may be raised in the comparisons with no adjustment made with a stricter criterion for the significant difference. Further study may also be required to confirm some pooled results derived from limited number of included studies in our review.

Conclusion
In conclusion, the overall proportion of S. aureus causing SSIs in mainland China was similar to that in the US, and the proportion of MRSA was possibly lower. The real proportion of S. aureus may be higher than that reported from the Chinese surveillance system. Both proportions of S. aureus and MRSA tended to depend on types of surgeries. Therefore, clinicians should take into account the types of surgery when taking care of post-operative patients and managing S. aureus and MRSA SSIs. Vancomycin and linezolid appeared to be effective for MRSA in SSIs. Further well-designed studies on this topic, including surveillance and primary prospective studies, are required to provide further reliable evidence.
Supporting Information