Risk factors for surgical site infection in head and neck cancer

Surgical site infection (SSI) frequently occurs in patients with head and neck cancer (HNC) after tumor resection and can lead to death in severe cases. Moreover, there is no definitive conclusion about the risk factors of SSI. Therefore, it is of great clinical significance to study the factors affecting the SSI. The HNC patients included in this study were all from the Department of Oral and Maxillofacial Surgery of the Second Xiangya Hospital of Central South University (CSU), and these patients received surgical treatment in the department from January 2018 to December 2019. The cross tabulation with chi-squared testing and multivariate regression analysis were applied to determine the risk factors of SSI. To identify the key risk factors of SSI, the caret package was used to construct three different machine learning models to investigate important features involving 26 SSI-related risk factors. Participants were 632 HNC patients who underwent surgery in our department from January 2018 to December 2019. During the postoperative period, 82 patients suffered from SSI, and surgical site infection rate (SSIR) was about 12.97%. Multivariate logistic regression analysis shows that diabetes mellitus, primary tumor site (floor of mouth), preoperative radiotherapy, flap failure, and neck dissection (bilateral) are risk factors for SSI of HNC. Machine learning indicated that diabetes mellitus, primary tumor site (floor of mouth), and flap failure were consistently ranked the top three in the 26 SSI-related risk factors. Diabetes mellitus, primary tumor site (floor of mouth), flap failure, preoperative radiotherapy, and neck dissection (bilateral) are risk factors for SSI of HNC.


Background
The frequency of surgical site infection (SSI) after head and neck cancer (HNC) surgery is high, especially in cleancontaminated surgery and major surgery [1]. Previous studies have shown that postoperative surgical site infection rate (SSIR) in HNC patients was about 8.1 ~ 45% [2][3][4][5][6]. It remains high even when the adequate prophylactic antibiotic are used [7]. SSI may have an adverse prognosis, cause significant morbidity, flap necrosis, poor cosmetic results, delayed wound healing, prolonged hospitalization, and, sometimes, even death [2,8]. Meanwhile, SSI increases the mortality during the perioperative period. In our previous study, we found that SSI is a risk factor for perioperative death of HNC and it is significantly correlated with poor overall survival (OS) of HNC patients [8].
Many risk factors of SSI have been reported in previous studies, such as diabetes, preoperative radiotherapy, tumor Chengwen Gan and Yannan Wang contributed equally to this work. location, advanced American Society of Anesthesiologists (ASA) grade, and contaminated wounds [7,[9][10][11][12][13]. However, due to the differences in study methodology, number of patients and sample size, there are still some disputes in different studies. For example, whether diabetes, tumor location, or previous radiotherapy can be regarded as potential risk factors for SSI causes controversy [11,14,15]. Coskun [16], Sepehr [17], and Hitomi [12] et.al found that diabetes, tumor location, and history of prior radiotherapy were not associated statistically with wound infection in head and neck surgery (HNS). However, Milap D. et.al [18] and Margita et.al [19] indicated that HNC patients with diabetes and tumor localization have significantly greater rates of postoperative infections. So far, there is no clear definition of risk factors for SSI of HNC surgery.
In addition, these SSI study cohorts appear to be highly heterogeneous if they included both head and neck surgical area and flap donor area infections. Therefore, in this study, we only focused on the SSI of head and neck incision after major surgery. Through this study, we found that diabetes mellitus, primary tumor site (floor in mouth), preoperative radiotherapy, flap failure, and neck dissection (bilateral) are risk factors for SSI of HNC. Through machine learning, for the first time, we found that diabetes mellitus, tumor site, and flap failure were consistently ranked the top three in the 26 SSI-related risk factors. Therefore, identifying these risk factors may help improve the perioperative care and management of HNC patients.

Ethics
This retrospective chart review study involving human participants was in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. The study was approved by the Ethics Committee of Second Xiangya Hospital of Central South University (No.202009717).

Study design and samples
A total of 632 consecutive patients underwent HNC major surgery with primary reconstruction at Department of Oral and Maxillofacial Surgery from January 2018 to December 2019. Surgical procedures include excision of the tumor, neck dissection, and flap reconstruction, if required. All head and neck wounds were categorized into class I to class IV according to the Centers for Disease Control (CDC) Surgical Wound Classification [13,20]. The criteria for enrollment in this study were as follows: (1) more than 1 month of follow-up postoperatively without a loss; (2) histologically confirmed diagnosis of head and neck malignant neoplasm which include but not limited to squamous cell carcinoma, sarcoma, adenoid cystic carcinoma, malignant melanoma, skin cancer, and malignant tumors derived from salivary glands, etc. [21]; and (3) only class I (clean) and class II (clean-contaminated) wounds were included. Exclusion criteria for this study were as follows: Class III (contaminated) and class IV (dirty-infected) wounds were excluded.

Data collection methods
The patients' data was collected and sorted from our hospital's electronic database, which contains prior medical records and postoperative follow-up information.

Demographics and clinical data
Data obtained from medical records included demographic distribution, pathology, clinical imaging, and treatment. A total of 26 putative risk factors were recorded in each patient and statistically analyzed to elucidate SSI-related factors. For demographics and comorbidities, age, sex, status of smoking and alcohol drinking, diabetes, TNM stage, and ASA grade were considered. Preoperative parameters included primary tumor site, history of radiotherapy and/or chemotherapy, preoperative WBC level, preoperative albumin level, and tumor recurrence. Parameters during surgery consisted of concurrent neck dissection, mandibular surgery, reconstructive procedures, airway management, operation time, estimated blood loss (EBL), and transfusion. The postoperative parameter of stage classification was ultimately determined based on the pathologic examination of the primary tumor and lymph nodes intraoperatively removed during the operation.
In this study, the definitions of SSI are as follows: (1) presence of purulent secretions from the incision or drain, (2) isolation of the infectious agent in the fluid found in the wound, and (3) diagnosis of infection by the physician (the diagnosis was mostly based on the presence of erythema, tenderness, swelling, fever, or elevated WBC level) [7,22]. Preoperative laboratory test results of blood cell analyses were defined according to parameters issued by the Ministry of Health of China in 2012 as follows: WBC (normal, 3.5-9.5 × 10 9 per l; low, < 3.5 × 10 9 per l; high, > 9.5 × 10 9 per l), albumin (normal, 40-55 g/l; low, < 40 g/l; high, > 55 g/l). The definition of smoking is participants who smoked more than 100 cigarettes were classified as smokers. The definition of drinking is the participant had drunk alcohol regularly (i.e., drank at least once a week on a regular basis) during the past 12 months, otherwise were considered nondrinkers [23].

Machine learning
Machine learning methods are particularly suited to predictions based on existing data and can improve the accuracy of prediction over the use of conventional regression models by capturing complex, nonlinear relationships in the data [24]. It is widely used in the prediction of clinical outcomes and the evaluation of the weights of prognostic factors [25,26]. To identify the key risk factors of SSI, we used the popular open-source R package "caret" to construct three different machine learning models to investigate features of importance involving 26 SSI-related risk factors. Neural network (nn), random forest (rf), and Gradient Boost Machine (gbm) were used to rank as the 26 risk factors by their status in two SSI states (positive and negative). The results were visualized by ggplot2 package. The R code and documentation for the analysis are available online (https:// github. com/ timod eist/ class ifier_ selec tion_ code). All classifiers were run on each dataset in a 1000-repeated nested fivefold cross-validation with hyperparameter tuning [27].

Statistical analysis
Statistical analyses were performed using SPSS 25.0 software (SPSS Inc., USA) and R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria). Pearson's chisquared test was used to determine whether there was a significant difference between each selected demographic or clinical factors and SSI. Multiple regression analysis was performed to reveal the relationship between SSI and demographic and clinical factors [28]. P value < 0.05 was considered statistically significant.

Demographic features of the patients.
A total of 632 patients with a median age of 57 (from 24 to 92) have been enrolled in this study, including 493 males and 139 females. Among them, SSI was present in 82 patients (12.97%). One hundred twenty-six patients had diabetes. Thirty-five patients underwent preoperative radiotherapy. Twenty-two patients received preoperative chemotherapy.

Univariate analysis of SSI risk factors in HNC patients
The cross tabulation with chi-squared testing was applied to analyze the difference in parameters between groups. The patient-related demographic characteristics are shown in Table 1. We found that the incidence of SSI was higher in patients with diabetes (***P < 0.001). These results also show that patients with a history of smoking are more likely to develop SSI (* P < 0.05).

Multivariate logistic regression analysis of risk factors of SSI in HNC patients
A multivariate logistic regression analysis was applied to analyze the relationship between SSI and the significant factors which has been identified in prior univariate analyses (shown in

Identify the key risk factors of SSI through Machine Learning
Here, for the first time, to evaluate the most important characteristics of the demographic and clinical factors which associated with SSI, we used the neural network (NN), random forest (RF), and Gradient Boost Machine (GBM) to build predictive models and rank 26 SSI-related risk factors, respectively ( Fig. 1A-C). We found that diabetes mellitus, tumor site, and flap failure were consistently ranked the top three of the 26 SSI-related risk factors and were selected as the most prognostically important locus.

Discussion
HNC always requires surgery-based comprehensive sequential therapy [29]. However, advances in medical care and the use of antibiotics have not significantly reduced the incidence of SSI in HNC patients after surgery. SSI can prolong hospital stays, increase medical costs, and, in severe cases, even lead to death [2,8]. Our previous studies have shown that SSI is associated with increased perioperative mortality and poorer OS in HNC patients [30]. Therefore, exploring the risk factors for SSI control of postoperative infection can not only reduce the burden on patients and doctors, but also reduce the occurrence of other postoperative complications and improve the quality of life and survival rate of patients. Several previous studies have identified diabetes as a risk factor for SSI [10,12,13,31]. Through this study, we further confirmed that diabetes is an independent risk factor for SSI. In multivariate logistic regression analysis, patients with diabetes had a more than threefold increased risk of developing SSI compared with those without diabetes (OR = 3.212). After further ranking the importance of the risk factors that may be related to SSI through three models constructed by machine learning, we found that diabetes ranks top in two of the models (Fig. 1A, C). These results suggest that diabetes is the one that deserves the most attention of all possible risk factors for SSI. The 2017 CDC guidelines recommend perioperative blood glucose control for the prevention of SSI in patients with or without diabetes, with a target blood glucose level of < 200 mg/dl [20]. Therefore, for HNC patients with diabetes and other SSI risk factors  (such as prior radiotherapy), blood glucose should be strictly controlled within the normal range during the perioperative period to prevent infection and any other complications. At present, it is generally believed that wound infection caused by diabetes is mainly related to microvascular disease and immunosuppression and interferes with wound healing [32,33]. It is worth noting that diabetes is closely related to flap failure [34,35]. In this study, we found that flap failure is a risk factor for SSI (Tables 3 and 4), suggesting both diabetes and flap failure may be risk factors of SSI. Whether prior radiotherapy is a risk factor for SSI is still controversial [12]. Girod DA et al. [31] and Penel N et al. [36] believe that there was no difference in SSI regardless of whether patients received radiotherapy before surgery, while studies by Robbins KT [37] and Hitomi [12] showed that preoperative radiotherapy increases the risk of SSI in HNC patients. In this study, we found a 2.9-fold increase in the odds of SSI in the preoperative radiotherapy group compared to the non-preoperative radiotherapy group (OR = 2.983). In this study, through machine learning, we found that prior radiotherapy ranks the fourth (Fig. 1A). These results suggest that preoperative radiotherapy is an independent risk factor for SSI. It is widely believed that preoperative radiotherapy can reduce collagen deposition and angiogenesis in the process of wound healing by inducing DNA mutations, microvascular damage, and soft tissue fibrosis, leading to wound healing problems [13,38]. There are also investigations indicating that preoperative radiotherapy may increase the risk of flap failure [39], which may further increase the risk of SSI [40]. Therefore, preoperative radiotherapy should be carefully considered in elderly and frail HNC patients with underlying diseases such as diabetes, especially in patients with squamous cell carcinoma (SCC), given the insensitivity of SCC to Fig. 1 The correlation between SSI and demographic or clinical parameters of HNC patients. Rank chart for the 26 SSIrelated risk factors by A NN, B RF, and C GBM radiotherapy and the increased risk of infection and flap crisis caused by radiotherapy [41,42].
In terms of surgery-related variables, we found that flap failure, primary tumor site, and neck dissection were independent risk factors for SSI (Tables 3 and 4). Especially when the flap failed, or the tumor was located at the floor of the mouth, or bilateral cervical lymph node dissection was selected, the risk of SSI increased by 4.562, 4.239, and 3.286 fold, respectively. The result is consistent with previous studies [10-13, 31, 43]. Notably, flap failure and tumor site ranked in the top three among the three machine learning models, which indicates that the probability of SSI is greatly increased in HNC patients with tumor located in the floor of the mouth and/or with flap failure. Some possible mechanisms underlying the association of flap failure, tumor site, and neck dissection with increased SSI are proposed, as follows: (1) When the tumor involves the floor of the mouth and metastasizes to the cervical lymph nodes, a penetrating defect in the mouth and cervical connection can be created during the removal of the tumor and the dissection of cervical lymph node. During postoperative recovery, saliva and food residues may enter the neck through these gaps and cause infection [44,45]. (2) In aggressive malignancies, complex three-dimensional defects are often formed during tumor resection and cervical lymph node dissection (especially bilateral neck lymph node dissection), which makes the shape of the flap and the tissue defect incompletely matched. Subsequently, the skin flap shrinks, resulting in the formation of postoperative dead space, which tends to cause fluid accumulation and provide a medium for bacterial growth [46,47]. (3) Flap failure occurs frequently in the process of flap reconstruction, and reoperation becomes an inevitable choice currently. Multiple operations, increasing operative time, and necrotic flap tissue all increase the risk of postoperative infection [48,49]. Therefore, for patients with tumor invasion of the floor of the mouth or bilateral cervical lymph node dissection, the wound should be carefully closed and the dead space should be tightly filled to prevent oral and neck traffic in the process of the flap reconstruction. Also, adequate antibiotics are used to prevent infection [50].
However, there are limitations in our study, some information about patient characteristics might be missed due to observer bias. And there are some unknown and unmeasurable confounders that need to be further measured or explored.

Conclusions
In conclusion, SSIR of HNC patients undergoing surgery in our hospital was about 12.97%. We have identified that diabetes mellitus, tumor site (floor of mouth), flap failure, preoperative radiotherapy, and neck dissection (bilateral) are risk factors for SSI of HNC. Among them, diabetes mellitus, tumor site (floor of mouth), and flap failure are the three risk factors most closely associated with SSI.