Risk Factors for Postoperative Deep Infection after Instrumented Spinal Fusion Surgeries for Degenerative Spinal Disease: A Nationwide Cohort Study of 194,036 Patients

Previous studies to identify risk factors for postoperative deep infection following instrumented spinal fusion surgery for degenerative spinal disease are based on insufficient information and have limited use in clinical practice. This study aims to fill this gap by assessing the risk factors and their adjusted relative risks through a comprehensive analysis, including all core information. In this nationwide, population-based, cohort study, data were obtained from the Korean National Health Insurance claims database between 1 January 2014, and 31 December 2018. This study included a cohort of 194,036 patients older than 19 years, who underwent instrumented spinal fusion surgery for degenerative spinal disease. We divided this population into cases (patients with postoperative deep infection) and controls (patients without postoperative deep infection); risk factors for postoperative deep infection were determined by multivariable analysis. The definition of postoperative deep infection varied, and sensitivity analyses were performed according to each definition. The estimates of all the statistical models were internally validated using bootstrap samples. The study included 767 patients (0.39%) with postoperative deep spinal infections and 193,269 controls. The final multivariable model identified the following variables as significant risk factors for postoperative deep infection: age between 60–69 years (OR = 1.6 [1.1–2.3]); age between 70–79 years (OR = 1.7 [1.2–2.5]); age > 80 years (OR = 2.1 [1.3–3.2]); male sex (OR = 1.7 [1.5–2.0]); rural residence (OR = 1.3 [1.1–1.5]); anterior cervical approach (OR = 0.2 [0.1–0.3]); posterior cervical approach (OR = 0.5 [0.2–1.0]); multiple approaches (OR = 1.4 [1.2–1.6]); cerebrovascular disease (OR = 1.5 [1.2–1.8]); peripheral vascular disease (OR = 1.3 [1.1–1.5]); chronic pulmonary disease (OR = 1.2 [1.0–1.4]); rheumatologic disease (OR = 1.6 [1.3–2.1]); liver disease (OR = 1.4 [1.1–1.7]); diabetes (OR = 1.5 [1.3–1.7]); hemiplegia or paraplegia (OR = 2.2 [1.5–3.3]); allogenous transfusion (OR = 1.6 [1.3–1.8]); and use of systemic steroids over 2 weeks (OR = 1.5 [1.1–2.0]). Our results, which are based on homogenous patient groups, provide clinicians with an acceptable tool for comprehensive risk assessment of postoperative deep infection in patients who will undergo instrumented spinal fusion surgery for degenerative spinal disease.


Introduction
Degenerative spinal disease is one of the most common chronic disorders affecting members of an aging population [1], lowering their quality of life with chronic pain and physical disability. Most patients with degenerative spinal disease can be treated conservatively with pain medication. However, the severity of the disease is closely associated with age; in an aging population, a significant number of people undergo surgery for advanced disease, and various types of spinal instruments are inevitably used 2 of 14 to support their weak spinal structures before or after surgery. Postoperative infection in these patients is one of the most devastating postoperative complications.
Studies do not consider age a consistent risk factor for infection after spinal surgery. However, patients' comorbidities and the level of surgical invasiveness are believed to be associated with a higher risk of postoperative spinal infection [2,3]. Because members of an aging population are more predisposed to developing medical comorbidities than those of a younger population, the incidence of postoperative spine infection is also expected to be increased in them.
Postoperative deep infection occurring in patients who have undergone instrumented spinal fusion surgeries can cause mortality or disastrous complications resulting from long-term antibiotic use and immobilization. These infections can also necessitate further surgeries including removal of a previously inserted instrument to control the infection or additional instrumentation for stabilization [4]. Therefore, when spine surgeons plan instrumented spinal fusion surgery for patients with degenerative spinal disease, identifying those at high risk of postoperative deep infection is a prerequisite to avoiding postoperative complications.
Studies have attempted to identify the risk factors for postoperative spinal infections. However, there were several limitations. Most of the studies were based on a limited number of patients, which was not representative of the general population [3,5]. There have also been several studies with large sample sizes [6][7][8][9][10][11][12][13][14][15]; however, these studies had methodological limitations. First, the target diseases for which spinal surgery was performed were not clearly outlined or analyzed individually. Second, the patient population was heterogeneous, including patients who underwent all types of spinal surgery, regardless of spinal instrumentation. The risk factors for postoperative infection in patients with different diagnoses and the surgical methods varied. Third, postoperative spinal infection was not properly defined or classified. The clinical outcomes of superficial and deep wound infections, including those of the vertebral body or intervertebral disc, are different and should be analyzed separately. Fourth, several types of core clinical information that influence the occurrence of postoperative infection, such as comorbidities and surgical approach, were not comprehensively evaluated. Finally, events that occurred between the index spinal surgery and the diagnosis of postoperative spinal infection, such as epidural injection, were not considered.
The purpose of our study is to investigate the risk factors for postoperative deep infection among patients who underwent instrumented spinal fusion surgery for degenerative spinal disease and to present their relative risks through a comprehensive analysis, including all possible core information. To overcome these limitations, our study used a domestic national claims database that covers the entire population.

Database
In this nationwide, population-based, cohort study, data were obtained from the Health Insurance Review and Assessment Service (HIRA) (which contains data from the Korean National Health Insurance (NHI)) and the National Medical Aid (NMA) databases. The NHI covers approximately 97% of the population, and the NMA program covers the remaining 3%. The HIRA database reviews all data from the NHI and NMA. Therefore, the HIRA database contains all inpatient and outpatient data from hospitals and community clinics in Korea, making a nationwide cohort study feasible. The diagnostic codes were assigned according to a modified version of the tenth revision of the International Classification of Disease (ICD-10) and the seventh revision of the Korean Classification of Disease (KCD-7). Drug use under diagnosis was identified using the Anatomical Therapeutic Chemical and HIRA general name codes. This study was approved by the institutional review board of our hospital (IRB No. 2020-03-009-001).

Definitions
We defined postoperative deep infection as any infection occurring within one year of the index spinal fusion surgery. A minimum follow-up period of one year was mandatory for inclusion in the study. To determine a postoperative deep spinal infection, an infectious event was expected to have required intravenous antibiotic therapy over 4 weeks for the defined

Definitions
We defined postoperative deep infection as any infection occurring within one year of the index spinal fusion surgery. A minimum follow-up period of one year was mandatory for inclusion in the study. To determine a postoperative deep spinal infection, an infectious event was expected to have required intravenous antibiotic therapy over 4 weeks for the defined ICD-10 codes (Supplementary Table S1) [16]. Postoperative deep spinal infection was identified using the following ICD-10 codes [8,17] Table S1) [16]. Postoperative deep spinal infection was identified using the following ICD-10 codes [8,17]: intraspinal abscess (G06.1); osteomyelitis of vertebra (M46.

Covariates
Preexisting comorbidities relevant to postoperative deep infection were identified according to the ICD-10 codes (Supplementary Table S2), within one year prior to the index surgery, based on the Charlson comorbidity index (CCI). The CCI score is the sum of the weighted scores of each comorbidity and has shown good agreement with ICD-10 codes [18,19]. Data regarding transfusion (allogenous or autologous) for the index surgery (Supplementary Table S3) and use of steroids (Supplementary Table S4) and immunosuppressants within two weeks before and after the index surgery were retrieved [20]. The type of hospital and region of residence were defined in accordance with previous studies [20,21].

Statistical Analysis
Data are reported as mean ± standard deviation for numerical variables and as numbers and frequencies (%) for categorical variables. Logistic regression models were used to construct prediction models for postoperative deep infections. All significant independent variables (p < 0.05) from the univariate analysis were included in the first multivariable model (Model 1) and subsequently chosen by backward stepwise selection (Model 2, final model). Multicollinearity between covariates was tested using the variance inflation factor. The performance of the prediction model was assessed by the area under the receiver operating characteristic curve, for discriminative ability, and by the Hosmer-Lemeshow goodness-of-fit statistics, for calibration.
Postoperative deep infection was additionally defined using the following methods (Table 1): shorter (2 weeks or more) and longer (6 weeks or more) duration of therapeutic intravenous antibiotics for postoperative deep spine infection, and limiting postoperative deep infection as an infection that occurred within 60 days of the index surgery. Sensitivity analysis was performed according to the three definitions. The estimates of all statistical models were internally validated with a relative bias based on 1000 bootstrapped samples.

Covariates
Preexisting comorbidities relevant to postoperative deep infection were identified according to the ICD-10 codes (Supplementary Table S2), within one year prior to the index surgery, based on the Charlson comorbidity index (CCI). The CCI score is the sum of the weighted scores of each comorbidity and has shown good agreement with ICD-10 codes [18,19]. Data regarding transfusion (allogenous or autologous) for the index surgery (Supplementary Table S3) and use of steroids (Supplementary Table S4) and immunosuppressants within two weeks before and after the index surgery were retrieved [20]. The type of hospital and region of residence were defined in accordance with previous studies [20,21].

Statistical Analysis
Data are reported as mean ± standard deviation for numerical variables and as numbers and frequencies (%) for categorical variables. Logistic regression models were used to construct prediction models for postoperative deep infections. All significant independent variables (p < 0.05) from the univariate analysis were included in the first multivariable model (Model 1) and subsequently chosen by backward stepwise selection (Model 2, final model). Multicollinearity between covariates was tested using the variance inflation factor. The performance of the prediction model was assessed by the area under the receiver operating characteristic curve, for discriminative ability, and by the Hosmer-Lemeshow goodness-of-fit statistics, for calibration.
Postoperative deep infection was additionally defined using the following methods (Table 1): shorter (2 weeks or more) and longer (6 weeks or more) duration of therapeutic intravenous antibiotics for postoperative deep spine infection, and limiting postoperative deep infection as an infection that occurred within 60 days of the index surgery. Sensitivity analysis was performed according to the three definitions. The estimates of all statistical models were internally validated with a relative bias based on 1000 bootstrapped samples. Data extraction and statistical analysis were performed using SAS Enterprise Guide 6.1 (SAS Institute, Cary, NC, USA).

Results
Among 216,033 patients who underwent instrumented spinal fusion surgery for degenerative spinal disease between 2014 and 2018, 3060 were initially identified as having the ICD-10 codes for postoperative deep spinal infection ( Figure 1). Among them, 262 (8.6% of 3060) who had undergone additional local invasive spinal procedures within 90 days prior to the diagnosis of deep infection, were excluded ( Figure 1). Of the remaining 2798 patients, 2031 (72.6%) received intravenous antibiotics for less than 4 weeks and were excluded. Finally, our study included 194,036 patients divided into two cohorts: 767 and 193,269 patients with and without postoperative deep spinal infection, respectively. The median interval between the index surgery and the occurrence of infection was 54 days (interquartile range, 29-123 days). Table 1 presents the incidence of postoperative spinal infections according to the four definitions. The incidence rate of postoperative infection decreased as the defined duration of therapeutic intravenous antibiotics increased: it was 0.5%, 0.39%, 0.3% with antibiotics over 2, 4, and 6 weeks, respectively ( Table 1). Regardless of the definition, the incidence of postoperative spinal infection decreased during the study period ( Figure 3).

Results
Among 216,033 patients who underwent instrumented spinal fusion surgery for degenerative spinal disease between 2014 and 2018, 3060 were initially identified as having the ICD-10 codes for postoperative deep spinal infection (Figure 1). Among them, 262 (8.6% of 3060) who had undergone additional local invasive spinal procedures within 90 days prior to the diagnosis of deep infection, were excluded (Figure 1). Of the remaining 2798 patients, 2031 (72.6%) received intravenous antibiotics for less than 4 weeks and were excluded. Finally, our study included 194,036 patients divided into two cohorts: 767 and 193,269 patients with and without postoperative deep spinal infection, respectively. The median interval between the index surgery and the occurrence of infection was 54 days (interquartile range, 29-123 days). Table 1 presents the incidence of postoperative spinal infections according to the four definitions. The incidence rate of postoperative infection decreased as the defined duration of therapeutic intravenous antibiotics increased: it was 0.5%, 0.39%, 0.3% with antibiotics over 2, 4, and 6 weeks, respectively ( Table 1). Regardless of the definition, the incidence of postoperative spinal infection decreased during the study period ( Figure 3).

Comparison of the Two Patient Cohorts with and without Postoperative Deep Infection: Univariable Analysis
Baseline characteristics of the patients in the two cohorts are presented in Table 2. Male sex, older age groups, rural residence, higher hospital volume, CCI score, and various comorbidities were associated with an increased risk of postoperative deep infection. The treatment profiles of the two cohorts are presented in Table 3. Surgical approach, cage use, allogeneic transfusion, and prolonged use of systemic steroids (over 2 weeks) were associated with an increased risk of postoperative deep infection.

Sensitivity Analysis
Sensitivity analysis was performed to assess our prediction model according to the dif-  Table 4) remained consistently significant in this prediction model, and the posterior thoracic approach was chosen as a significant predictor (Supplementary Table S5). When defined as an infection requiring over 6-week intravenous antibiotics, 584 patients were identified as having postoperative deep infection. Age >80 years and posterior thoracic approach, which were significant predictors for the final model, were not significant predictors for this prediction model (Supplementary Table S6). Additionally, end-stage renal disease was identified as a significant predictor.   All significant independent variables (p < 0.05) from the univariate analysis were included in Model 1. Relative bias was estimated as the difference between the mean bootstrapped regression coefficient estimates and the mean parameter estimates of Model 1, divided by the mean parameter estimates of Model 1. Sensitivity analysis was also performed when postoperative deep infection was defined as an infection that occurred within 60 days of the index surgery; 417 patients were identified, and the median interval between the index surgery and postoperative infection was 30 days (interquartile range, 21-42 days). Rural residence, cervical posterior approach, chronic pulmonary disease, cerebrovascular disease, hemiplegia or paraplegia, and use of systemic steroids over 2 weeks, which were significant predictors for the final model (Model 2 in Table 4) were not significant predictors for this definition (Supplementary Table S7).
For these three additional models, the bootstrap-adjusted odds ratios and confidence intervals are displayed in Figure 5. All significant independent variables (p < 0.05) from the univariate analysis were initially included and subsequently chosen by backward, stepwise selection in Model 2. Relative bias was estimated as the difference between the mean bootstrapped regression coefficient estimates and the mean parameter estimates of Model 2, divided by the mean parameter estimates of Model 2.
After bootstrap validation, the relative bias of the estimates in Model 1 was high, between −557.1% and 202.7% (Table 4). After backward selection, the relative bias of the estimate in Model 2 was low, between −7.6% and 10.1% (Table 5). Bootstrap-adjusted odds ratios and confidence intervals for Model 2 are shown in Figure 4.  Table 5).  Table 5).  Table 4) remained consistently significant in this prediction model, and the posterior thoracic approach was chosen as a significant predictor (Supplementary Table S5). When defined as an infection requiring over 6-week intravenous antibiotics, 584 patients were identified as having postoperative deep infection. Age >80 years and posterior thoracic approach, which were significant predictors for the final model, were not significant predictors for this prediction model (Supplementary Table S6). Additionally, endstage renal disease was identified as a significant predictor. Sensitivity analysis was also performed when postoperative deep infection was defined as an infection that occurred within 60 days of the index surgery; 417 patients were identified, and the median interval between the index surgery and postoperative infection was 30 days (interquartile range, 21-42 days). Rural residence, cervical posterior approach, chronic pulmonary disease, cerebrovascular disease, hemiplegia or paraplegia, and use of systemic steroids over 2 weeks, which were significant predictors for the final model (Model 2 in Table 4) were not significant predictors for this definition (Supplementary Table S7).
For these three additional models, the bootstrap-adjusted odds ratios and confidence intervals are displayed in Figure 5.

Discussion
The recent incidence of postoperative spinal infection ranges from 0.2% to 16.7% [3,5], and that of deep infection is relatively low (<1.8%) [5,8,22,23]. In our study, 194,036 patients who underwent instrumented spinal fusion surgeries for degenerative spinal disease between 2014 and 2018 were analyzed, and the mean incidence rate of postoperative deep spinal infection in them was 0.39% (767 of 196,329, Table 1). Multivariable analysis identified the following variables as significant risk factors for postoperative deep spinal infection after instrumented spinal fusion surgery for degenerative spinal disease: older age, male sex, rural residence, multiple surgical approach, comorbidities including cerebrovascular disease, peripheral vascular disease, chronic pulmonary disease, rheumatologic disease, liver disease, diabetes, hemiplegia or paraplegia, allogeneic transfusion, and systemic steroid use over 2 weeks (Figure 4). In contrast, the anterior and posterior cervical approaches were associated with a lower risk of infection.
Previous studies have shown conflicting results in evaluating the factors related to the risk of infection such as age [3], and some have shown methodological limitations [6][7][8][9][10][11][12][13][14][15]. Our study has several advantages compared with previous studies with similar research purposes. First, our study is the largest study of this type, and is based on a homogenous group of patients with clear target diseases who underwent spinal fusion surgeries using instrumentation due to degenerative spinal disease. Second, the clinical endpoint (postoperative deep infection) was clearly defined, based on the patients with a minimum of one-year mandatory follow-up after the index surgery ( Figure 2). Third, rigorous statistical calculation was performed to overcome the essential limitations of claims databases. Patients were stratified into three categories according to the duration of the postoperative intravenous antibiotics (over 2, 4, and 6 weeks) and into two categories according to the interval between the index surgery and the postoperative infection (within 60 days and all intervals). Sensitivity analysis was performed according to each defined category (Table 5,  Supplementary Tables S5-S7). Bootstrap adjustment was performed for all prediction models (Table 5, Supplementary Tables S5-S7). Fourth, core clinical information, including precise surgical approach and precise comorbid conditions, could be included using the domestic national claims database covering the entire population. We could also assess all clinical events that occurred between the index surgery and postoperative infection, and exclude patients who underwent additional invasive spinal procedures (Figure 1). This study has several limitations. First, the claims database is not designed for clinical research; thus, important clinical information, such as laboratory and radiologic data and precise surgical profiles including different sterilization procedures, surgical protocols, and techniques among hospitals or surgeons, was not included in the study, which may have led to bias. However, the consistency of our results was confirmed by a sensitivity analysis using various definitions of postoperative deep infection [24]. Possible discrepancies between the diagnostic codes in the database and the real disease in patients could be potential sources of bias. However, high-revenue invasive procedures and drug use are well documented in the administrative databases [25][26][27] which our study utilized. Several important degenerative diseases, including adult spinal deformity, were not included in our study owing to the limited data capacity for analysis. In our study, patients who underwent multiple surgical approaches showed an increased rate of postoperative deep infection, and those with deformities were also expected to have an increased rate of infection. Future studies including such patients are required.
In conclusion, postoperative deep spinal infection after instrumented spinal fusion surgery for degenerative spinal disease is one of the most devastating complications of spinal surgery. A precise and comprehensive risk assessment for postoperative deep spinal infection is a prerequisite for the clinical success of instrumented spinal fusion surgery for degenerative spinal disease. Previous studies have consistently reported that comorbidities and surgical invasiveness are associated with postoperative spine infection. However, evidence for precise variables such as individual comorbidities or surgical methods was limited [8,[28][29][30]. In this regard, our results, which are based on a large number of homogenous patient groups, provide clinicians with an acceptable tool for comprehensive risk assessment of postoperative deep infection in patients who are set to undergo instrumented spinal fusion surgery for degenerative spinal disease.