Introduction

Surgical site infections (SSI) are still among the most frequent healthcare associated infections (HAI) and entail significant morbidity and mortality globally [1]. Considering antimicrobial resistance (AMR) a global threat, mutual international efforts must focus on reducing infection rates of key players such as SSI. In Europe, the most common causative pathogen is Staphylococcus aureus being part of the human skin microbiota [2]. S. aureus SSI is associated with prolonged duration of hospitalization, death rates and treatment costs [3]. In contrast to infections caused by antibiotic-resistant organisms like methicillin-resistant S. aureus (MRSA), S. aureus SSI rates are independent on regional epidemiological influences [1].

Recent efforts to understand and reduce SSI could diminish infection rates [2, 4,Role of the funding source

This study was an investigator-initiated trial with the University of Cologne as sponsor. The study was funded by a restricted research grant from Pfizer. The company provided advisory input into the trial design and was provided the primary raw data. Pfizer did not participate in site section, trial conduct, data analysis or manuscript writing.

Results

From 259 459 initially exported data sets of patients who had undergone surgery, 178 902 were included in the analysis; characteristics are depicted in Table 1. In total, 80 557 were excluded due to age < 18 years [7 855], year of surgery not 2016 (10 943), MCAR (29 845), eye or minimal invasive surgery (17 846) or duplicate entries (13 904) (Fig. 3). The procedure most frequently performed within the entire cohort was DER01 (Incision and excision of skin and subcutaneous tissue; n = 9483), followed by GYN08 (Caesarean section; n = 8146).

Table 1 Patient characteristics of the SALT cohort, S. aureus SSI cases, and controls
Full size table
Fig. 3
figure 3

Flow chart of case inclusion

Full size image

Data from 178 902 patients who had undergone surgery in five European countries (Fig. 1) in 2016 were analysed. Of those, 764 had S. aureus SSI constituting an overall incidence of 0.4% (Table 2). The overall relative proportions of superficial, deep, and organ/space SSI were 46.3%, 27.0%, and 26.7%, respectively (Additional file 3: Table S2). A total of 660 (86.0%) S. aureus SSIs were caused by methicillin susceptible S. aureus (MSSA) and 104 cases (14.0%) by MRSA.

Table 2 Overall incidence of S. aureus and of most common procedures and procedures with highest rates of SA SSI
Full size table

Cohort data and incidence of S. aureus SSI per country are depicted in Fig. 4. For the nested case–control part 20 (2.6%) of 764 cases were excluded due to missing documentation of the matched cases.

Fig. 4
figure 4

Cohort data and incidence rates of S. aureus SSI. A. Overall and country-specific S. aureus SSI incidence; number above columns indicate the numbers of participating centres per country. B. Country-specific number of patients contributed to the cohort. C. Incidence rates of S. aureus SSI by surgical discipline. Across the different surgical disciplines, S. aureus SSI incidence rate is 0.5 (IQR 0.3)

Full size image

Procedure-specific S. aureus SSI distribution

S. aureus SSI incidence did not differ across surgical specialties. Median discipline-specific incidence was 0.5 (IQR 0.265) as depicted in Fig. 4. Highest procedure-specific incidence (Table 2) was seen in visceral surgery (minimally invasive surgery of the bile duct, 4.6% [95% CI 1.2 -17.6%], and laparoscopic excision of the small intestine, 4.2% [95% CI 0.6–28.4%]) as well as cardiothoracic surgery (operation on the diaphragm, 3.6% [95% CI 0.5–24.5%]). With regard to distribution of S. aureus SSI, it most frequently occurred in caesarean Sects. (8.9%), operations on scull, brain, and meninges (5.8%), and open repair of fractured long tubular bones (4.32%). Surgery performed in cardiothoracic surgery (0.8%) and vascular surgery (0.6%) had a higher specialty-specific S. aureus SSI incidence than anticipated for the respective group along a prediction model (adjusted residual values 5.7 and 3.2, respectively). Further in depth-analyses revealed that patients undergoing surgery in one of those mentioned groups had significantly more comorbidities than patients from other surgical fields (data not shown) and were the oldest of the entire cohort (mean age in the cardiothoracic surgery group 64.3 years and in the vascular surgery group 63.3 years). A list of all procedure specific incidences is included in the Additional file 2.

Structural equality of cases and controls

Between cases and controls, no statistically significant differences in sex, age, or BMI were observed hence indicating successful matching; distribution of comorbidities was similar and significant differences were only seen in patients with peripheral vascular diseases, hemiplegia, and solid tumours (Table 1). Cases and controls were matched along the SALT (Staphylococcus aureus Surgical Site Infection Multinational Epidemiology in Europe [SALT] study) code [10] and 760 of 764 controls had undergone the same type of procedure as the respective case. For optimal matching, four patients were allocated to similar types of procedures – three within the respective specialty (two neurosurgical procedures) and one with a surgical procedure from another specialty (dermatological procedure with one visceral surgical procedure, Additional file 3: Table S3).

Sex did not differ across the entire cohort (51.7% female and 48.3% male), but there were more men than women among SSI cases (51.9% and 48.1%, respectively; p = 0.043; Table 1). Mean age was 56.7 years for controls and 58.1 years for cases. Distribution of age groups was comparable, and in both populations the majority of patients was older than 60 years. Within the group of SSI cases, most patients (65.4%) were overweight (BMI ≥ 25). However, there was no statistically significant difference between cases and controls (p = 0.77).

The most frequent comorbidity in the overall cohort was diabetes (11.4%), followed by solid malignancies (7.3%), chronic cardiovascular disease (CVD, 4.4%), peripheral vascular disease (3.4%), and chronic kidney disease (CKD, 3.2%) (Table 1). Within the population of S. aureus SSI cases, CVD (23.1%), solid malignancies (22.3%), diabetes (21.0%), and CKD (7.8%) were also the most frequent comorbidities. Comorbidities of the control group were distributed in a comparable manner (significantly differing prevalence only reported for hemiplegia, solid tumours, and peripheral vascular disease; Table 1).

Comparison of cases and controls

Univariate analyses showed that in comparison with matched uninfected controls, patients with S. aureus SSI were more likely to be re-admitted to the hospital (p < 0.005), more likely in need of revision surgery (p < 0.005) and hospitalized for a longer period (mean hospitalization 17 versus 12 days; p < 0.005) (Table 2). ICU stay after a surgical procedure was as frequent in cases as in controls (144 versus 130, p = 0.349). Out of 144 ICU stays in the case cohort, 42 (29.2%) were associated to SSI. An in depth-analysis of the different surgical categories showed, that case patients undergoing cardiothoracic, gynaecological, neuro- or vascular surgery were associated with significantly higher rates of readmission to hospital, revision surgery, and longer hospitalization periods (Table 3 and 4).

Table 3 Factors affecting likelihood of SA SSI (Logistic regression)
Full size table
Table 4 Outcome of SA SSI. Complications compared between cases and matched controls without infection
Full size table

In a multivariate analysis risk factors (Table 3) for S. aureus SSI were male sex, chronic cardiovascular disease (OR 1.9, confidence interval [CI] 1.3–2.6), peripheral vascular disease (OR 2.7, CI 2.0–3.7), lymphoma [OR 3.9, CI 1.6–9.4], solid tumour [OR 2.1, CI 1.6–2.8], diabetes (OR 1.7, CI 1.4–2.1) and chronic kidney disease (OR 1.7, CI 1.3–2.5). Smoking status was similar in both, cases, and controls (p = 0.774) (Table 5 and 6).

Table 5 Category-specific complications
Full size table
Table 6 Survival
Full size table

Discussion

We report results from the first multinational study of surgical site infection sufficiently powered to determine SSI rates irrespective of procedure type. Our approach allowed to detect and analyse 764 cases of culture-proven S. aureus surgical site infections among 178 902 included patients – an investigation of a scale comparable to the most recent ECDC SSI report (1 016 S. aureus SSI cases) – while providing data of much higher granularity than in single-country surveillance efforts [11].

The percentage of comorbidities in our cohort mirrors the distribution among the European population, in particular the diabetes rate of 11.4% compared to 9.5% in the general population [12]. Our study population was older than the European average (37.1% above 65 years in our cohort vs 18.4% to 22.8% in the respective countries [13]) reflecting that older patients are more likely to undergo surgery [14, 15].

The overall S. aureus SSI incidence was 0.43%, which is about 50% lower than assumed in our sample size calculation based on European averages. While lower than the European average, our findings are in line with prior publications by centres with similar expertise (ranging between 0.2% and 0.9%) [16], [17,18,19,20] Except for a higher incidence in the UK (1.9%), we saw a uniform S. aureus SSI distribution among all countries – most likely reflecting similar standards among leading surgical centres. The higher incidence in the UK might be a direct result of the lower-case volume at the sole participating UK centre and a resulting true higher SSI rate. Alternatively, it might be a statistical effect as a function of the lower number of contributed cases (9 168 vs mean of 42 424). As our trial was restricted to culture proven S. aureus SSI differences in incidence might also be the result in differences in clinical approaches to SSI (e.g. the frequency of obtaining cultures or use of antibiotic prophylaxis).

In line with our hypothesis that indicator procedures are not representative of their respective categories, S. aureus SSI incidence displayed a high degree of intra-disciplinary variability. Currently monitored indicator procedures were representative of their respective category in some disciplines (e.g., total prosthetic replacement of knee joint 0.4% [0.2–0.7%] for orthopaedic surgery 0.5% [0.4–0.6%]), but not others (e.g., revascularization of the heart 1.6% [1.1–2.2%] for cardiothoracic surgery 0.8 [0.6–1.0%]). While statistical interference testing for each procedure was beyond the scope of this work, non-overlap** confidence intervals strongly suggest real differences rather than random effects. These differences highlight the need to expand surveillance efforts beyond indicator procedures.

Beyond our expectation of relevant S. aureus SSI rates in all surgical domains our data show similar average S. aureus SSI rates across all subspecialties. This finding is explained by S. aureus pertaining to the skin microbiome common to all surgical sites in contrast to SSI caused by site-specific organisms (e.g., Enterobacterales in GI surgery). These characteristics further strengthen our assumption that S. aureus is a prototypical causative organism in SSI. This supports the notion that S. aureus can be used as a marker or sentinel pathogen in trials focusing on overall periprocedural care and host defence in a discipline-independent fashion, i.e., in contrast to current discipline-specific approaches [1]. This finding also highlights a need for further exploration of risk factors across all procedures and subspecialties.

The identified risk factors for S. aureus SSI were male sex, chronic cardiovascular disease, peripheral vascular disease, lymphoma, solid tumour, diabetes and chronic kidney disease reflecting a population with severe intern diseases. As recommended in general for these patients, high awareness for infections including adequate prophylactic treatment should guide medical decisions.

The choice of our retrospective study design can be regarded as either a strength or a weakness. We believe that, in the context of HAI by a known pathogen, a retrospective study prevents the Hawthorne effect of confounding by observation while not affecting detection rates [21, 22]. This study design cannot be easily applied to the study of HAI without a microbiologically proven pathogen. Furthermore, while the choice of S. aureus as a marker organism is indeed compelling, our study results cannot be extrapolated to culture-negative SSI or SSI caused by other organisms. However, the current work may be regarded as a proof of concept for a novel epidemiological approach to HAI.

We limited our investigation to adult patients; our findings are thus not applicable to children. No available risk stratification approach has been established in a procedure-independent fashion and research shows wide SSI rate variability within risk categories depending on the specific surgery type [7]. Consequently, we matched cases and controls by a propensity score rather than by more conventional approaches like the NHSN SSI risk index. The paucity of variables used in matching cases and controls resulted directly from limitations of available electronic health record data and hospital information technology capabilities. However, our subsequent analysis demonstrated successful matching.

As reported elsewhere [23] our study was initially designed to establish S. aureus SSI patterns representative of the overall European surgical population. However, during the centre selection process it became apparent that centres with the necessary technical expertise to generate exports and perform local matching would constitute a highly select sample and thus not be representative of the overall European S. aureus SSI epidemiology. We thus decided to focus on analysing high performance centres rather than trying to extrapolate results from a highly select sample to the overall surgical population. We believe this approach generates more valid and robust data. Post-hoc changes to study design risk can introduce bias and thus compromise validity of results. This risk should, however, not apply to our analysis as we merely abstained from over-interpreting our data by trying to extrapolate it to the overall European surgical population, while not deviating in any other aspect from the trial protocol or the statistical analysis plan.

Future endeavours in applying real-world data to epidemiological research in HAI will be aided by progressive digitalization of medicine, and the compatibility of database interfaces. Technical aspects regarding the harmonization of different procedure coding systems have been discussed elsewhere [9].

Conclusion

Our methodology and results diverge in important aspects from ongoing SSI surveillance and highlight the expanded possibilities provided by electronic health records and big data. Discrepancies between results from prospective trials and real-world evidence are a well-known issue. Contrary to prior hierarchical views of evidence levels, current approaches integrate real-world evidence with the results of more formal study types [24, 25]. We believe that such an integrative approach of supplementing prospective surveillance results with real-life data will provide a comprehensive grasp of the current state of HAI.