Deep Spinal Infection after Outpatient Epidural Injections for Pain: A Retrospective Sample Cohort Study Using a Claims Database in South Korea

The South Korean National Health Insurance Service sample database 2.0 was used in this retrospective cohort study. South Korea has used a single universal health insurance coverage system for all citizens since 1989, and the National Health Insurance Service has been the single insurer since 2000, comprising both National Health Insurance and Medical Aid.¹⁴  It has provided longitudinal sample data from approximately 1 million people from 2002 to 2015, which accounted for 2.2% of the total South Korean population eligible for health insurance in 2006.¹⁶  Individuals were stratified with proportional allocation according to age, sex, region, health insurance type (medical insurance premium or medical aid), and household income. Within each stratum, systematic stratified random sampling was conducted using the individual’s total annual medical expenses as a target variable for sampling to ensure representativeness. The sample database contains information about the patients’ age, sex, and type of insurance; a list of diagnoses based on the International Classification of Diseases, Tenth Revision; the medical costs claimed; prescribed medications; treatments covered by National Health Insurance, including interventions with their relevant diagnoses; hospital facility information; and socioeconomic and survival status. Data on socioeconomic status included the area of a patient’s residence (a rural area or small or large cities), level of income, and information on whether medical insurance premium or medical aid was used. The cohort population is refreshed annually by adding representative samples of newborns each year as preexisting patients become ineligible by death or emigration. Since 2006, information from medical aid beneficiaries has been incorporated into a single database¹⁷ ; therefore, we used the data from 2006 to 2015 in this study. Because an anonymized public database was used, the requirement for approval for the study was waived by the Seoul National University Hospital (Seoul, South Korea) Institutional Review Board (approval No. 1901-107-1005).

This study consisted of three parts: the first part describes the 10-yr national trend of outpatient-based single-shot epidural injections in pain practice in South Korea; the second part measured the incidence of deep spinal infections after single-shot epidural injections in outpatient pain practice, which was the primary outcome of the study; and the third part assessed the risk factors for infectious complications after these injections. The data analysis and statistical plan were written after the data were accessed.

We conducted a nation-specific descriptive analysis of the overall use of outpatient-based, single-shot epidural injections in pain practice from 2006 to 2015 in South Korea. The following procedures were identified as a single-shot epidural injection for the management of pain: “epidural injection (LA221),” “cervicothoracic epidural injection (LA321),” “lumbosacral epidural injection (LA322),” “spinal nerve block (LA252),” “selective nerve root block (LA354),” “dorsal root ganglion block (LA355),” and “epidurography (HA102).” The code of “epidural injection” has replaced “cervicothoracic” and “lumbosacral epidural injections” since 2008. Similarly, the code of “spinal nerve block” has been updated to “spinal nerve root block” or “ganglion block” since 2008. Various epidural procedures, such as caudal epidurography, transforaminal epidural injection, and epidural adhesiolysis, had been claimed using the same code (“epidurography”). Epidural injections for regional analgesia/anesthesia were distinguished from single-shot epidural injections for pain management; different injection codes were designated for epidural anesthesia. Additionally, inpatient and outpatient procedures were distinguished by different claim codes. Outpatient-based epidural injections could be performed repeatedly in a single patient; thus, demographics were presented only once per individual, although the characteristics of epidural injections were described per claim. Demographic data included sex, age, residential area (large city, small city, or rural area), and health insurance subscriber status (National Health Insurance or Medical Aid). The details of the epidural injection included the diagnosis relevant to the epidural injection; class of institution where the procedure was performed for managing pain (primary clinic, community hospital, secondary general hospital, or tertiary general hospital); medical department of the interventionalist who had claimed medical costs (anesthesiologist, spine surgeons [orthopedic surgeons and neurosurgeons], or others [rehabilitation doctors, internists, radiologists, and general surgeons]); type of the procedure (epidural injection, cervicothoracic epidural injection, lumbosacral epidural injection, spinal nerve block, selective nerve root block, dorsal root ganglion block, or epidurography); and use of corticosteroids for epidural injection (triamcinolone, dexamethasone, methylprednisolone, betamethasone, hydrocortisone, or none).

We assessed the primary outcome of the study, which was the incidence of deep spinal infection after only outpatient-based, single-shot epidural injection in pain practice. First, we defined deep spinal infection as a new-onset infection occurring within 90 days of the most recent outpatient-based single-shot epidural injection. For classifying a “deep” spinal infection, the infectious event needed to include hospitalization for at least 1 night and oral or intravenous antibiotic therapy for at least 4 weeks.^18–20  Therefore, the primary outcome included the following three major concepts: the deep spinal infection code, at least a 4-week course of antibiotic therapy that needed hospitalization for at least 1 night, and an epidural injection within the preceding 90 days. Injections were assumed to be independent of one another. Patients who had a new-onset deep spinal infection but died during the 4-week antibiotic treatment period were still included in the analysis. To distinguish between persisting and new deep spinal infections, we reviewed the medical history up to at least 1 yr before the registered date of deep spinal infection. Because we used the sample data from 2006 to 2015, deep spinal infections newly registered from 2007 to 2015 were counted in the second part. Spinal infection was identified using the International Classification of Diseases, Tenth Revision (Supplemental Digital Content 1, https://links.lww.com/ALN/C581); some infections were defined according to codes in a previous study on postoperative and drug abuse–related spinal infection,²¹  and other relevant codes were chosen by the authors. Extradural or subdural abscesses and postprocedural infections were included if they occurred in the spine. To eliminate superficial infections, codes regarding infection of muscles were excluded. Tuberculous spondylitis and Brucella spondylitis were excluded because their characteristics differ from those of pyogenic deep spinal infection.²² 

We analyzed the predisposing risk factors for deep spinal infection after an outpatient-based, single-shot epidural injection in pain practice. The date of the last epidural injection was set as the index date (fig. 1), and data on all explanatory variables for risk analysis were obtained based on the index date. In addition to the demographic data, such as age, sex, residential area, and income level, the Charlson Comorbidity Index²³  was estimated to investigate the predisposing comorbidity relevant to injection-related deep spinal infection. Comorbidities were reviewed up to 1 yr before the index date, which included myocardial infarction, congestive heart failure, peripheral vascular disease, hemiplegia or paraplegia, dementia, chronic pulmonary disease, rheumatologic disease, peptic ulcer disease, mild/moderate to severe liver disease, diabetes with/without chronic complications, renal disease, malignancy, and metastatic solid tumor.²⁴  Each comorbidity has a weighted score according to its potential influence on mortality, and the total score of the Charlson Comorbidity Index is the sum of these scores; a higher comorbidity score is associated with higher in-hospital mortality.²⁴  Additionally, we obtained data on the use of immunosuppressive agents, such as cyclosporine, tacrolimus, or methotrexate, or systemic steroids for 30 days or more within the 90 days before the index date. Moreover, the types of epidural procedures, the claims for epidural steroids, and the performance of multiple epidural injections within 90 days (three times or more) were compared between patients with deep spinal infection and those without infectious events.

The sample size was based on the available data from the South Korean National Health Insurance Service sample database 2.0, and no statistical power calculation was conducted before the study. The data were extracted using the SAS Enterprise Guide 7.1 (SAS Institute, USA). Statistical analyses were performed using R 3.3.3 (R Core Team, Austria),²⁵  which included the following packages: “haven,”²⁶  “tidyverse,”²⁷  “tableone,”²⁸  and “MASS.”²⁹  The data were reported as mean ± SD, median (interquartile range), and frequencies with percentages. The Shapiro–Wilk test was used to assess the normality of data. All P values presented are two-tailed, and P < 0 < 0.05 was considered statistically significant.

To determine the statistically significant predisposing risk factors, univariable and multivariable analyses were performed and presented with the estimated odds ratio and 95% CI of covariates. Patients with missing data were excluded from the analysis. For multivariable regression analysis, we constructed three models. Model 1 (full model) included all covariates except for variables with a low frequency of less than four. Model 2 was established with significant covariates (P < 0.05) from univariable analysis, and model 3 (final model) included covariates chosen from a backward stepwise selection. Statistically significant covariates (P < 0.05), which minimized the Akaike information criterion, remained in the final model. The estimates of each model were internally validated with relative bias and root mean square difference based on 200 bootstrapped samples. In the final model, multicollinearity between the covariates was tested by the variance inflation factor. The performance of the final model was assessed using the Hosmer–Lemeshow P value test.³⁰  In addition, sensitivity analyses were conducted for the final multivariable model, using broader and more restrictive time-based definitions (antibiotics for 2 weeks or more and for 6 weeks or more) of deep spinal infection.

In South Korea, the number of individuals who underwent single-shot epidural injection in pain practice steadily increased by twofold from 2006 to 2015, and most of them were outpatients (13.1 per 1,000 persons in 2006 to 28.8 per 1,000 persons in 2015; Supplemental Digital Content 2, https://links.lww.com/ALN/C582). During the observational period, the proportion of men increased from 36.7 to 40.5%, those aged 65 yr or more increased from 38.3 to 43.5%, those covered by National Health Insurance increased from 91.6 to 94.2%, and those living in large cities increased from 52.2 to 57.8%.

The number of epidural injections in the outpatient pain practice also doubled from 40.8 injections per 1,000 persons in 2006 to 84.4 in 2015. Although the actual number of outpatients who underwent epidural injections increased over time, the number of injections per patient was relatively constant during the period (a median of two times a year with the same interquartile ranges from one to three, except for 2006). Although most procedures were performed in clinics, claims by community hospitals increased from 9.5% in 2006 to 19.0% in 2015. Only half of the claims included corticosteroids for epidural injections during the study period (42.5 to 58.1%). Among the types of corticosteroids, triamcinolone had a 95.5 to 96.6% claim rate until 2012, which rapidly dropped to 48.9% in 2013 and 15.5% in 2015. The claim rate for dexamethasone was 4.4 to 5.4% in the 2000s, which increased sharply to 43.6% in 2012 and 74.6% in 2015.

Among 1,108,361 individuals in the sample cohort between 2007 and 2015, 111,098 underwent at least one outpatient-based single-shot epidural injection in pain practice. In addition, 96,083 individuals (86.5%) who did not undergo any inpatient-based single-shot epidural injections were included in the analyses (fig. 2; Supplemental Digital Content 2, https://links.lww.com/ALN/C582). Among them, 584 patients had codes for deep spinal infection; however, only 191 met the antibiotic treatment criteria of this study. Of these 191 patients, 143 underwent invasive procedure(s) within 90 days before the occurrence of infection; 91 patients were excluded because they had surgical operations (n = 68) or paraspinal injections (n = 23) other than epidural injections. Consequently, of 95,551 individuals who underwent 501,509 outpatient-based single-shot epidural injections in pain practice, 52 patients met our criteria of the deep spinal infection after the injection (fig. 2). The incidence rate of the deep spinal injection was 5.4 (95% CI, 4.1 to 7.2) subjects per 10,000 outpatients who underwent single-shot epidural injections and 1.0 (95% CI, 0.8 to 1.4) cases per 10,000 epidural injections (0.01%) in outpatient pain practice. Incidences were scattered from just 1 case in 2007 to 9 cases in 2013 (median, 7; interquartile range, 4 to 8; table 1). Among the 52 patients, 34 (65%) underwent a surgical operation to treat the infection, 4 (8%, including 1 surgical case) died in the succeeding month, and 9 (17%, including 3 surgical cases) died 6 months from the occurrence of the infection.

Univariable associations with deep spinal infection are presented in table 2, where age of 65 yr or more, living in a rural area, Charlson Comorbidity Index of 3 or higher, peripheral vascular disease, chronic pulmonary disease, mild liver disease, complicated diabetes, the use of immunosuppressive agents or systemic steroids, and frequent epidural injections (three times or more) within 90 days were significant variables (P < 0.05). There were no missing data in patients with deep spinal infection, whereas a small portion of data was missing in the control group (table 2).

In the multivariable analysis, three models, including model 1 (full model), model 2 (model with significant variables from univariate analysis), and model 3 (final model), were established (table 3). In the final model, age of 65 yr or more (odds ratio, 2.91; 95% CI, 1.62 to 5.5; P = 0.001), living in a rural area (odds ratio, 2.85; 95% CI, 1.57 to 5.0; P < 0.001), complicated diabetes (odds ratio, 3.18; 95% CI, 1.30 to 6.7; P = 0.005), repeated epidural injections (three times or more) within 90 days (odds ratio, 2.34; 95% CI, 1.22 to 4.2; P = 0.007), and recent use of immunosuppressants or systemic steroids (odds ratio, 2.90; 95% CI, 1.00 to 6.7; P = 0.025) remained statistically significant risk factors, with low multicollinearity among covariates (all variance inflation factors less than 2). In models 1 and 2, these factors were also statistically significant in association with the infectious complication after the procedure. The final multivariable model was well calibrated according to the Hosmer–Lemeshow goodness-of-fit test (P = 0.610). In the bootstrap adjustment, the relative biases of estimates on the final multivariable model were low (3.1% for advanced age, 0.7% for living in a rural area, −4.3% for complicated diabetes, and −3.3% for recent multiple epidural injections) or moderate (−7.1% for the recent use of immunosuppressants or systemic steroids).

Sensitivity analysis was performed to assess the multivariable model, which used the modified definition of deep spinal infection. When the lenient definition was used (the use of antibiotics for 2 weeks or more), 356 patients with deep spinal infection were identified. Among them, 64 were related to an outpatient-based epidural injection (1.3 [95% CI, 1.0 to 1.6] cases per 10,000 epidural injections). Age of 65 yr or more (odds ratio, 2.46; 95% CI, 1.47 to 4.3; P = 0.001), living in a rural area (odds ratio, 2.50; 95% CI, 1.44 to 4.2; P = 0.001), complicated diabetes (odds ratio, 3.45; 95% CI, 1.58 to 6.7; P = 0.001), repeated epidural injections (three times or more) within 90 days (odds ratio, 2.71; 95% CI, 1.55 to 4.6; P < 0.001), and recent use of immunosuppressants or systemic steroids (odds ratio, 2.88; 95% CI, 1.10 to 6.2; P = 0.015) remained in the model derived from patients with a 2-week period of antibiotic treatment. When we adopted a stricter definition of deep spinal infection (use of antibiotics for 6 weeks or more), 117 patients with deep spinal infection were identified. Among them, 37 were related to an outpatient-based epidural injection (0.7 [95% CI, 0.5 to 1.0] cases per 10,000 injections). Age of 65 yr or more (odds ratio, 2.98; 95% CI, 1.48 to 6.5; P = 0.003), living in a rural area (odds ratio, 2.57; 95% CI, 1.24 to 5.1; P = 0.008), complicated diabetes (odds ratio, 3.91; 95% CI, 1.45 to 8.8; P = 0.003), and repeated epidural injections (three times or more) within 90 days (odds ratio, 2.67; 95% CI, 1.26 to 5.3; P = 0.007) remained in the model focused on the primary outcome definition requiring 6 weeks of antibiotics. Recent use of immunosuppressive agents or systemic steroids was excluded in the 6-week model (odds ratio, 3.25; 95% CI, 0.96 to 8.3; P = 0.027).

Based on the South Korean National Health Insurance Service sample cohort, the number of outpatient-based, single-shot epidural injections per 1,000 persons in pain practice doubled from 40.8 in 2006 to 84.4 in 2015 in South Korea, which was twofold more than that in the United States (21.2 in 2000 and 42.3 in 2014 based on Medicare data).³¹  The incidence of deep spinal infection after the procedure was 5.4 (95% CI, 4.1 to 7.2) per 10,000 persons and 1.0 (95% CI, 0.8 to 1.4) cases per 10,000 injections (0.01%), representing 1 deep spinal infection per 10,000 outpatient single-shot epidural injections. Age of 65 yr or more (odds ratio, 2.91), living in a rural area (odds ratio, 2.85), complicated diabetes (odds ratio, 3.18), the use of immunosuppressants or systemic steroids (odds ratio, 2.90), and epidural injections three times or more within 90 days (odds ratio, 2.34) increased the risk of deep spinal infection after epidural injections in the outpatient pain practice by two- to threefold.

The increase in epidural injections performed in the outpatient pain practice may be explained by the increased supply and demand for specialized spinal pain management in South Korea. The recent aging trend in South Korea³²  has led to the rapidly increasing prevalence of degenerative spinal pain disorders,^33,34  potentially resulting in the increasing use of epidural injections in an outpatient setting. Pain management to improve the quality of life, which has become a global trend,³⁵  possibly steered the marked increase in the number of procedures and expenses of the National Health Insurance in South Korea.³⁶  Moreover, economic growth and a broad application of the National Health Insurance Service have allowed more South Koreans to have access to specialized pain interventions.³⁷  Although health disparity might exist in individuals between living areas (rural vs. cities), the increasing proportion of hospital-based epidural injections since early 2010 suggests that a growing number of specialty hospitals providing spinal interventions have been established in South Korea.³⁸  The rising number of epidural injections in outpatient pain practice could reduce disability and socioeconomic loss caused by spinal pain.³²  However, concerns about the unnecessary usage of epidural injection and increasing exposure to devastating complications, such as spinal infections, have been raised.⁴ 

Previously, Windsor et al.¹⁶  suggested that the severe infection rate associated with spinal injections ranged from 0.01 to 0.10%. In our study, the incidence of deep spinal infection after outpatient-based single-shot epidural injection for pain management (0.01%) was that of the lower value reported by Windsor et al. (0.01%).¹⁶  Defining deep spinal infections after epidural injections was the most challenging aspect of this study because a strict definition could omit true positives, whereas a lenient definition may include false positives. We assume that the definition of the infection in our study was closer to the strict one. It led us to perform sensitivity analyses using broader (antibiotics use for 2 weeks or more) and strict definitions (antibiotics use for 6 weeks or more), which did not substantially differ from our definition (antibiotics use for 4 weeks or more). Despite their rarity, deep spinal infections after epidural injections are catastrophic complications, which required surgical operation in most patients (n = 34, 65%) and led to death within 6 months in 27% (n = 13) of patients in this study. Moreover, the incidence of deep spinal infection caused by outpatient single-shot epidural injections in our study seems to be 10-fold higher than that reported in obstetric patients treated with epidural analgesia/anesthesia, which was 7 to 9 cases per million.³⁹  This could be attributed to the different characteristics of the subjects, such as advanced age with a high frequency of comorbidities in our study, compared to relatively young and healthy pregnant women in the previous study.³⁹  Concurrent administration of steroids may have also contributed to the difference between the studies. Therefore, healthcare providers must pay attention to high-risk patients who are scheduled for single-shot epidural injections in outpatient pain practice.

Regardless of the exogenic infectious source, predisposing comorbidities including diabetes mellitus, chronic renal disease, malignancy, and substance abuse are known risk factors for spinal infections, such as pyogenic osteomyelitis, epidural abscess, or discitis.^13,40  Our univariable analysis suggests that advanced age, Charlson Comorbidity Index of 3 or higher, complicated diabetes, chronic pulmonary disease, peripheral vascular disease, liver disease, and recent use of immunosuppressants are associated with epidural injection–related deep spinal infection. Although the exact reasons are unknown,⁴¹  the higher risk of deep spinal infection after epidural injection in patients with the aforementioned diseases could be attributed to immunosuppressive conditions in such diseases⁴²  or the effects of treatment, such as corticosteroid use in chronic pulmonary disease⁴³  or peripheral vascular disease.⁴⁴  In our multivariable analysis, patients with complicated diabetes had a more than threefold higher risk of deep spinal infection after the procedure in the outpatient pain practice than those without complicated diabetes. However, chronic pulmonary disease and peripheral vascular disease showed no statistical significance after adjusting for the recent administration of immunosuppressants. Furthermore, frequent epidural injections were associated with an increased risk of infectious complications, and those living in rural areas were more exposed to the risk of procedure-related deep spinal infection even after adjusting for age and several comorbidities in the multivariable regression. Although this could be explained by several assumptions, the main reason could be the health disparity between rural and urban populations in South Korea, including the relatively low accessibility to pain experts, low income levels, and insufficient social support in rural areas.⁴⁵ 

Another trend was noted in the use of corticosteroids in South Korea, wherein only half of the claims included corticosteroids (Supplemental Digital Content 3, https://links.lww.com/ALN/C583). Epidural injections may not always include corticosteroids for diagnostic purposes. The other assumption would arise from a new regulation by the Ministry of Food and Drug Safety (Cheongju-si, South Korea), which prohibits epidural injection of triamcinolone since the change in the indications of particulate steroids for epidural use.^46,47  Although the World Institute of Pain (Winston-Salem, NC) has recently insisted that dexamethasone is only safe for transforaminal injections, with little information on its long-term safety,⁴⁸  a dramatic change from triamcinolone to dexamethasone for epidural administration occurred between 2012 and 2013 in South Korea.²  Claims for epidural triamcinolone after the implementation of this national regulation may not be reimbursed by the Health Insurance Review & Assessment Service (Wonju-si, South Korea). They could be classified into unclaimed corticosteroids, reflecting a limitation of using this sample database. During lumbar surgical operations, epidural injections of corticosteroids have been suggested to increase postoperative infection rates.^49,50  Therefore, future studies are necessary to investigate whether the infection risk increases similarly during single-shot epidural injections with corticosteroids in pain practice.

This study has several limitations. First, our sample database contains the general shortcomings of a claims database, such as discrepancies in actual performances caused by limited access to patients’ medical records. Some valuable information, including patients’ symptoms and signs, is not included in the data. In addition, all causal relationships between invasive procedures and deep spinal infection were decided on the basis of the chronological registration to the health service system in this study; hence, there might be some ambiguous cases in which it was difficult to determine the causal relationship between invasive procedures and deep spinal infection. Additionally, the South Korean National Health Insurance Service database depends only on claimed medical benefits, which means that data on treatment paid out-of-pocket were missing. Moreover, despite the fine stratification, the sample database may still have a selection bias and may not be sufficient to assess the risk of rare complications. Second, patients who received adjunctive injections (e.g., acupuncture and trigger point injections) simultaneously with epidural injections and had deep spinal infection were included (n = 25 of 52, 48%); thus, the primary cause of the spinal infection may be confusing given the multiple interventions. Furthermore, the inclusion of the codes “selective nerve root block” and “dorsal root ganglion block” as epidural injections in this study may be an issue. These terms are interchangeably used with epidural injections in pain practice. However, they are technically not epidural injections, although epidural spreads of the injectates are frequently observed during the procedures. Another debate lies in defining “epidurography.” In South Korea, there is no claim code for transforaminal epidural injection, but “epidurography” is used for the procedure, which mainly accompanies epidural injection with local anesthetics and/or corticosteroids. Therefore, we included all these codes as epidural injections so as to not make an overly parsimonious process in identifying this rare complication. Third, in this study, we investigated deep spinal infection after outpatient-based, single-shot epidural injection, which is only one of the several critical adverse events after the procedure. Other complications resulting in catastrophic neurologic sequelae, such as spinal cord infarct or stroke, were not analyzed in this study; therefore, these complications require future investigation.⁵¹ 

In conclusion, the incidence of deep spinal infection after single-shot epidural injection in the outpatient pain practice in South Korea was very low (0.01% per injection). Advanced age, living in a rural area, complicated diabetes, use of immunosuppressants or systemic steroids, and multiple epidural injections within 90 days were associated with a two- to threefold risk of deep spinal infection after the procedure.

This study used the South Korean National Health Insurance Service sample cohort data (NHIS-2019-2-068 and NHIS-2019-2-203) provided by the South Korean National Health Insurance Service. The authors thank the Medical Research Collaborating Center of Seoul National University Hospital for statistical consultation.

Support was provided solely from institutional and/or departmental sources.

The authors declare no competing interests.