Prevalence of and risk factors for surgical site infections after pancreaticoduodenectomy: a systematic review and meta-analysis

Background: Surgical site infections (SSIs) are one of the most common complications after pancreaticoduodenectomy (PD); however, the global prevalence and risk factors for SSIs after PD remain unknown. Objectives: To investigate the prevalence of and risk factors for SSIs after PD. Methods: The PubMed, Embase, Cochrane Library, Web of Science, and Science Direct databases were systematically searched from inception to 1 December 2022. Observational studies reporting adjusted odds ratios (ORs) and 95% confidence intervals (CIs) of risk factors for SSIs in patients undergoing PD were included. Two independent reviewers in teams performed data extraction, risk of bias assessment, and level of evidence analysis. The pooled results were estimated using a random-effects model. The I 2 statistic and Q χ 2 statistic were used to assess heterogeneity. Funnel plots, Egger’s regression test, and the trim-and-fill method were used to determine publication bias. The primary outcomes were identifying risk factors for SSIs after PD. The secondary outcomes were the pooled prevalence rates of SSIs. Results: A total of 98 704 patients from 45 studies were included, and 80% of the studies were considered high quality. The estimated pooled prevalence of SSIs was 23% (0.19–0.27, I 2=97%). The prevalence of SSIs was found to be higher in Japan and lower in USA. Preoperative biliary stenting, higher body mass index (BMI), longer operation time, postoperative pancreatic fistula, soft pancreatic texture, perioperative blood transfusion, and cardiac disease were identified as significant risk factors for the development of SSIs after PD. Additionally, broad-spectrum antibiotics were a significant protective factor against SSIs. Subgroup analysis and sensitivity analysis showed that the results were robust. Conclusion and relevance: The prevalence of SSIs remains high and varies widely among regions. It is necessary to take effective preventive measures and carry out more prospective studies to further verify these results.


Introduction
Pancreaticoduodenectomy (PD) is a complicated medical surgical procedure used to treat diseases that are found in the periampullary region.It is one of the most challenging surgical procedures due to the complexity of the region being operated on and the high prevalence of postoperative complications [1,2] .Nearly one-third of patients have experienced complications after PD, among which surgical site infections (SSIs) are one of the most common complications, with a prevalence of up to 23.5% [1,3] .
According to the US Centers for Disease Control and Prevention (CDC) criteria, SSIs include superficial incisional SSIs, deep incisional SSIs, and organ/space SSIs [4] .A correlation has been observed between SSIs and various adverse outcomes, such as heightened postoperative pain and psychosocial distress, impaired wound healing, prolonged hospitalization, frequent readmissions, and elevated healthcare expenses [5][6][7] .

HIGHLIGHTS
• This is the first systematic review and meta-analysis of multiple risk factors for surgical site infections (SSIs) after pancreaticoduodenectomy (PD).• Nearly 23% of patients after PD had SSIs.And the prevalence of SSIs remains high and varies widely among regions.• Preoperative biliary stenting, higher body mass index, longer operation time, postoperative pancreatic fistula, soft pancreatic texture, perioperative blood transfusion, and cardiac disease were significant risk factors for SSIs.• Further studies are needed to elucidate the differences among regions and to determine the potential risk of SSIs.
Therefore, it is necessary to implement measures to safeguard patients at a high risk of developing SSIs and to prevent potentially severe outcomes.Multiple factors, such as preoperative biliary stenting [8,9] , antibiotic prophylaxis [10] , and wound protector use [11] , have been reported to be related to SSIs.However, this evidence is scattered and difficult to identify.In addition, most of the studies were focused on one specific factor [8][9][10][11] .The main objective of this study was to systematically explore risk factors for SSIs after PD.We also explored and estimated the prevalence of SSIs based on the literature included in the analysis of risk factors.

Search strategy
Search strategies were developed by a senior health sciences investigator.A pilot search was performed via PubMed to determine the search strategy.Then, we systematically searched the PubMed, Embase, Cochrane Library, Web of Science, and Science Direct databases up to 1 December 2022 to identify relevant articles.The search strategy was adjusted based on the syntax and subject headings of each database.The retrieval strategy adopted a combination of subject heading terms and keywords.The details of the search strategy for each database are presented in the Supplemental Digital Content 3, Table 1 (http:// links.lww.com/MS9/A300).

Inclusion/exclusion criteria
Studies were included if they met the following criteria: (1) patients who underwent PD; (2) evaluated the associated factors of SSIs; (3) reported adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for SSIs; (4) observational study designs (cohort study, case-control study, and cross-sectional study); (5)  published in English language; and ( 6) full text available.Except for not meeting the inclusion criteria, the exclusion criteria were as follows: (1) failure to determine the outcome of interest and (2) conference abstracts, study protocols, case reports, reviews, letters, and animal studies.

Study selection
All research literature retrieved was imported into EndNote X9.1.The study selection process included initial screening, secondary screening, and a final check.After removing duplicates, two reviewers screened titles and abstracts to exclude unrelated studies and conference abstracts according to predefined selection criteria.The eligible studies were included through full-text screening according to the criteria.If a conference abstract met our inclusion criteria, we proceeded to search for the corresponding full-text articles.In cases where full-text articles were not available, we contacted the principal investigators to verify whether the studies were published and to acquire data.Detailed discussion took place if there were any disagreements between the two reviewers.A third investigator would join and make the decision if the agreement could not be reached.

Risk of bias assessment
Quality assessment of the included studies was independently assessed by two reviewers using the Newcastle-Ottawa scale (NOS) [14] , a revised Cochrane risk-of-bias tool for observational studies that include the following domains: the selection of the study groups (0-4 points), the comparability of the groups (0-2 points), and the determination of either the exposure for cohort study or the outcome of interest for case-control study (0-3 points).The maximum score is 9 points.A score greater than 6 points was considered to indicate high quality.A third researcher was consulted to resolve the disagreement in case of any discrepancy.

Data extraction
Two reviewers collected data independently from the eligible studies and recorded them in predesigned study information tables in Microsoft Excel 2019.The following data were extracted and recorded: first author, publication year, region, study period, type of surgery, the definition of SSIs, data source, sample size, study design, type of SSIs, the prevalence of SSIs, risk factors, and adjusted ORs with 95% CIs.Any disagreement was settled by discussion.
SSIs were categorized into different groups: total SSIs, superficial incisional SSIs, deep incisional SSIs, organ/space incisional SSIs, and superficial/deep incisional SSIs according to the US CDC criteria and study description.Total SSIs indicated that SSI types were not distinguished in the included study.Superficial incisional SSIs, deep incisional SSIs, and organ/space incisional SSIs are consistent with the US CDC criteria.

Statistical analysis
To obtain the pooled prevalence of SSI and ORs with 95% CIs, a meta-analysis using the random-effects model of DerSimonian and Laird [15] was performed, given the unavoidable heterogeneity among the included studies.In addition, the results of the common-effects model (namely the fixed-effects model) are also presented in the figures and tables.A normality test was performed on the prevalence, which was transformed by four estimation methods, including logarithmic, logit, arcsine, and Freeman-Tukey double arcsine transformations.The normality of the untransformed prevalence was also assessed before the meta-analysis.Then, we selected the results that were close to the normal distribution.ORs and corresponding 95% CIs were also log transformed before analysis.
Heterogeneity was tested using the Cochrane's Q χ 2 statistics (significance level at P < 0.10) and the I 2 statistic (significance level at I 2 > 50%) [16] .In case of heterogeneity, subgroup analysis was carried out based on the characteristics of studies, including region, study design, and type of SSIs.Sensitivity analysis was performed by excluding one study at a time to test the robustness of the results.In order to analyze the prevalence of and risk factors for SSIs in specific countries, two countries with a substantial number of studies can be selected.
Moreover, potential publication bias was assessed by examining the symmetry of the funnel plot and Egger's regression test (significance level at P < 0.10) [17] .The trim-and-fill method was used to determine the impact of publication bias and determine whether the results were robust [18] .
Descriptive analysis was performed by using Microsoft Excel to provide summative figures.Meta-analysis was performed using the software R 4.2.2.The threshold for statistically significant differences was a two-tailed P value less than 0.05, except for the tests of heterogeneity and publication bias (P < 0.10).

Search results
In the initial search, 875 studies were identified.Of these, after excluding duplicated studies and other unrelated articles, 45 studies were eligible and ultimately included in the meta-analysis  . The detils of the screening process are shown in Supplemental Digital Content 3, Figure 1 (http://links.lww.com/MS9/A300).

Risk factors for SSI
A total of 73 risk factors for SSI were identified in the analysis.However, 60 factors were excluded without sufficient data for quantitative analysis.We included 13 risk factors that were reported in more than three studies for meta-analysis.

Pancreatic texture
Five studies [42,53,54,58] were finally included in the analysis.The results showed that a soft pancreatic texture could increase the risk of developing SSIs by 2.65 times compared with a hard texture (OR = 2.65, 1.43, 4.91, I 2 = 74%) (Fig. 2H).Similar results were observed in the subgroup analysis (Supplemental Digital Content 3, Figs 6-8, http://links.lww.com/MS9/A300).Preoperative albumin A total of six studies [28,32,33,38,58,61] evaluated the relationship between preoperative albumin and SSIs.Of these, one study [58]  was excluded due to the inconsistency of the direction of the reported effect (high levels of preoperative albumin compared to low levels).According to the meta-analysis of the remaining studies, people with lower levels of preoperative albumin were 2.28 times more likely to develop SSIs than patients with higher levels (OR = 2.28, 0.59-8.71,I 2 = 81%), but the difference was not significant (Fig. 2I).Subgroup analyses revealed similar results (Supplemental Digital Content 3, Figs 6-8, http://links.lww.com/MS9/A300).

American Society of Anesthesiologists (ASA) classification
The association between ASA score and SSIs was estimated in three studies [28,44,47] .The meta-analysis results showed that patients with an ASA score ≥ 3 were more likely to develop SSIs than those with an ASA score <3 (OR = 1.07, 0.68-1.67,I 2 = 28%), but the difference was not statistically significant (Fig. 2M).

Sensitivity analyses
The significance of the result changed when one of the three studies [42,53,54] was removed at a time for pancreatic texture using a random-effects model.For the sex factor, removing Tee et al. [58] from the analysis in the random-effects model and Morikane [48] in the common-effects model changed the significance of the conclusion.The exclusion of the study of Gavazzi et al. [34] or Yamamoto et al. [59] in the analysis of cardiac disease using a random-effects model and the removal of Gavazzi et al. [34] in the common-effects model resulted in a change in the significance of the conclusion.Other sensitivity analysis results showed that the pooled results were robust after omitting any of the studies once a time.The pooled results of any other risk factors did not change the significance, regardless of whether a random-effects model or common-effects model was used (Supplemental Digital Content 3, Fig. 9, http://links.lww.com/MS9/A300).

Assessment of publication bias
There was a publication bias in the analysis of preoperative biliary stenting, BMI, operation time, broad-spectrum antibiotics, and pancreatic texture based on the asymmetric funnel plots.The results of Egger's regression test also indicated publication bias (Table 2).The funnel plot asymmetry corrected using the trim-and-fill method suggested the existence of some publication bias and unrobust results (Table 2; Supplemental Digital Content 3, Fig. 10, http://links.lww.com/MS9/A300).

Discussion
In this study, we systematically estimated the prevalence of SSIs and evaluated the potential risk factors in patients undergoing PD.The analysis of 98 704 patients from 45 included studies suggested that SSIs were a serious postoperative complication after PD.The pooled prevalence of total SSIs was 23%.Preoperative biliary stenting, higher BMI, longer operation time, postoperative pancreatic fistula, soft pancreatic texture, perioperative blood transfusion, and cardiac disease were identified as significant risk factors for the development of SSIs after PD.However, antibiotic prophylaxis (second-generation or thirdgeneration cephalosporin), sex, preoperative albumin, diabetes mellitus, age, and ASA score were not significant risk factors.Moreover, we identified the use of broad-spectrum antibiotics as a significant protective factor against SSIs.Sensitivity analysis showed the robustness of the base-case results.
To the best of our knowledge, few reviews and meta-analyses have focused on the prevalence of SSIs after PD.According to the included studies, the prevalence rate of SSIs after PD varied from 3.2% to 51.0%.The pooled total SSI rate was 23% (0.19-0.27,I 2 = 97%), which was consistent with findings from previous research, up to 23.5% [1] .For the SSI type, the meta-analysis results showed that the lowest prevalence was estimated in deep incisional SSIs (2%), and the highest prevalence was in organ/ space SSIs (17%), which is similar to the published studies [22,23,26] .In terms of the region-specific SSI rate, Japan was reported to have the highest prevalence of SSIs compared with the United States, China, and other countries.
Several studies [8][9][10][11][64][65][66] have explored the potential risk factors associated with SSIs after PD; however, they were limited to a narrow scope with a particular factor. We sysematically reviewed the risk factors in the analysis.Preoperative biliary stent placement is a surgical procedure used for preoperative biliary drainage in patients with obstructive jaundice who are about to undergo neoadjuvant chemotherapy [1] .Preoperative biliary stenting has been demonstrated to increase the development of postoperative SSIs in previous studies [8,9] , which was consistent with our findings.A possible explanation is the transport of enteric bacteria from bile, which can cause bacteremia and then lead to infections [1] .A longer duration of drainage could be a potential risk factor for SSIs, although the optimal duration of drainage has not been determined [19] .The superiority of metal stents with fewer stent-related complications compared to plastic stents has been demonstrated [67] .The application of the common bile duct should be strictly limited to clear indications [68] .Thus, preoperative biliary stent placement using metal stents should be considered a priority.
A higher BMI may increase the risk of developing SSIs after PD, which is similar to other abdominal surgery studies [69] .Although BMI cannot directly and accurately reflect the body's fat composition, it can reflect the thickness of subcutaneous fat and the area of visceral fat to a certain extent.BMI may be a reliable parameter that has an impact on infection after laparotomy [70,71] .
It is noteworthy that compared with first-generation cephalosporins, there was no statistically significant protective effect of second-generation or third-generation cephalosporins against SSIs, while broad-spectrum antibiotics show a protective effect.One study [10] also indicated that the use of targeted broadspectrum antibiotics was linked to a significantly lower prevalence of SSIs after PD compared to standard antibiotic therapy.This is important for patients undergoing preoperative biliary drainage [41] .
As expected, a longer operating time was an independent risk factor for SSIs.Usually, a longer operation time leads to a longer exposure time to the environment and an increase in blood loss.In addition, a longer operation time itself may also reflect the complexity of the surgical procedure.A positive linear relationship between operation time and SSIs was demonstrated in a previous study [72] .In addition, there was a strong relationship between perioperative blood transfusion and the development of SSIs after PD.Blood transfusions may lead to immunosuppression and transmission of pathogens [73,74] .Therefore, it is recommended that clinicians prioritize their surgical skills, aim to decrease operation time, and minimize both intraoperative blood loss and perioperative blood transfusions.
Postoperative pancreatic fistula was found to be an independent risk factor for SSIs in our analysis.Pancreatic fistula is one of the most common complications after PD [75] .Pancreatic fistula can cause pancreatic secretions (digestive enzymes and other substances) to leak into the surrounding tissues or abdominal cavity, leading to SSIs.For the texture of the pancreas, soft pancreas is a risk factor for SSIs compared to hard pancreas.A soft pancreas is leaked more easily during and after PD than a hard pancreas, which can increase the risk of infection [76,77] .
Although not statistically significant, our analysis found that preoperative albumin was a potential risk factor for SSIs.Albumin could be used as an indicator of nutritional status.Low levels of albumin often suggest malnutrition, which can increase the risk of infection [78] .Contrary to the findings of previous studies on the risk factors for SSIs, our analysis did not find a significant association between diabetes mellitus and increased susceptibility to SSIs [69,79] .Similarly, male sex was not a significant risk factor in our study.Cardiac disease can be a risk factor for SSIs.Kent et al. [80] also found that coronary artery disease is a predictive factor for SSIs in a prospective study.
The ASA score is a classification system of the ASA based on the patient's physical condition and the risk of surgery.The results showed that an ASA score ≥ 3 was not significantly associated with an increased risk of developing SSIs.This finding is inconsistent with previous research, which has suggested that a higher ASA score is linked to a higher risk of SSIs [69] .Similarly, older age, although a potential risk factor for SSIs, was not significant in our study or in previous studies [64,65] .
In addition to the established risk factors for SSIs after PD, our analysis suggested some other potential risk factors that have not been extensively explored (Supplemental Digital Content 3, Fig.21, http://links.lww.com/MS9/A300).To explore these potential risk factors, we used two different approaches.One method was to treat different types of SSIs as separate studies within the same study, thus categorizing at least three independent groups.Pancreatic duct not over 3 mm compared to over 6 mm, unknown incision compared to midline incision, and preoperative cholangitis were risk factors for SSIs.Robot-assisted surgery compared with open surgery and wound protection were identified as protective factors against SSIs.The second method was to conduct subgroup analysis using studies that did not clearly specify their category.For example, we analyzed the results of neoadjuvant therapy under both neoadjuvant chemotherapy and neoadjuvant radiotherapy categories.The pooled effects showed that neoadjuvant chemotherapy, contaminated wounds compared to clean wounds, and dirty wounds compared to clean wounds were risk factors for SSIs, while neoadjuvant radiotherapy was not a significant risk factor.In addition to wound protector use [11] and robot-assisted surgery [81] , which have been shown to be protective factors for SSIs after PD, further research is required to investigate other potential risk factors.Additionally, Vasavada and Patel [66] also found that laparoscopic PD was associated with fewer SSIs than open PD.
There are some limitations.First, only English-published studies were included, which may limit the findings.Second, not all included studies were designed to report prevalence rates, which can lead to inaccuracies in the merged data.Third, it is important to note that some studies rely on retrospective analysis, which may be affected by nonrandomized strategies and unmeasured confounding factors.Fourth, although sensitivity and subgroup analysis were performed, heterogeneity among the included studies cannot be eliminated, especially for the unclear or inconsistent criteria of SSI diagnosis, surgical approach, the cause of the surgery, and the baseline characteristics variation of each study.

Conclusion
The prevalence rate of SSIs remains high, and potential risk factors related to SSIs should be given more attention.Regional variation exists, and a higher prevalence is observed in Japan.More efforts should be devoted to monitoring and preventing the development of SSIs in different regions, and more prospective studies are needed to assess potential influencing factors in the future.

Table 1
Baseline characteristics of the included studies
a 1. b

Table 2
Results of publication bias assessment of SSIs