Baseline parameters and the prediction of treatment failure in patients with intravenous drug use–associated spinal epidural abscesses

Justin Baum Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Justin Baum in
jns
Google Scholar
PubMed
Close
 MD
,
Stephanus V. Viljoen Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Stephanus V. Viljoen in
jns
Google Scholar
PubMed
Close
 MD
,
Connor S. Gifford Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Connor S. Gifford in
jns
Google Scholar
PubMed
Close
 MS
,
Amy J. Minnema Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Amy J. Minnema in
jns
Google Scholar
PubMed
Close
 MS
,
Ammar Shaikhouni Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Ammar Shaikhouni in
jns
Google Scholar
PubMed
Close
 MD, PhD
,
Andrew J. Grossbach Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Andrew J. Grossbach in
jns
Google Scholar
PubMed
Close
 MD
,
Shahid Nimjee Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by Shahid Nimjee in
jns
Google Scholar
PubMed
Close
 MD, PhD
, and
H. Francis Farhadi Department of Neurological Surgery, The Ohio State University Wexner Medical Center, Columbus, Ohio

Search for other papers by H. Francis Farhadi in
jns
Google Scholar
PubMed
Close
 MD, PhD
Free access

OBJECTIVE

Despite the increasing incidence of spinal epidural abscess (SEA), the baseline parameters potentially predictive of treatment failure remain poorly characterized. In this study, the authors identify the relevant baseline parameters that predict multimodal treatment failure in patients with either intravenous drug use (IVDU)–associated SEA or non-IVDU–associated SEA.

METHODS

The authors reviewed the electronic medical records of a large institutional series of consecutive patients with diagnosed SEA between January 2011 and December 2017 to characterize epidemiological trends as well as the complement of baseline measures that are predictive of failure after multimodal treatment in patients with and without concomitant IVDU. The independent impact of clinical and imaging factors in detecting treatment failure was assessed by performing stepwise binary logistic regression analysis.

RESULTS

A total of 324 consecutive patients with diagnosed SEA were identified. Overall, 226 patients (69.8%) had SEA related to other causes and 98 (30.2%) had a history of recent IVDU. While non-IVDU SEA admission rates remained constant, year-over-year admissions of patients with IVDU SEA nearly tripled. At baseline, patients with IVDU SEA were distinct in many respects including younger age, greater unemployment and disability, less frequent diabetes mellitus (DM), and more frequent methicillin-resistant Staphylococcus aureus infection. However, differences in length of stay, loss to follow-up, and treatment failure did not reach statistical significance between the groups. The authors constructed independent multivariate logistic regression models for treatment failure based on identified parameters in the two cohorts. For the non-IVDU cohort, the authors identified four variables as independent factors: DM, hepatitis B/C, osteomyelitis, and compression deformity severity. In contrast, for patients with IVDU, the authors identified three variables: albumin, endocarditis, and endplate destruction. Receiver operating characteristic and area under the curve (AUC) analyses were undertaken for the multivariate models predicting the likelihood of treatment failure in the two cohorts (AUC = 0.88 and 0.89, respectively), demonstrating that the derived models could adequately predict the risk of multimodal treatment failure. Treatment failure risk factor point scales were derived for the identified variables separately for both cohorts.

CONCLUSIONS

Patients with IVDU SEA represent a unique population with a distinct set of baseline parameters that predict treatment failure. Identification of relevant prognosticating factors will allow for the design of tailored treatment and follow-up regimens.

ABBREVIATIONS

AUC = area under the curve ; CRP = C-reactive protein ; DM = diabetes mellitus ; ESR = erythrocyte sedimentation rate ; IVDU = intravenous drug use ; MRSA = methicillin-resistant Staphylococcus aureus ; ROC = receiver operating characteristic ; SEA = spinal epidural abscess ; WBC = white blood cell .

OBJECTIVE

Despite the increasing incidence of spinal epidural abscess (SEA), the baseline parameters potentially predictive of treatment failure remain poorly characterized. In this study, the authors identify the relevant baseline parameters that predict multimodal treatment failure in patients with either intravenous drug use (IVDU)–associated SEA or non-IVDU–associated SEA.

METHODS

The authors reviewed the electronic medical records of a large institutional series of consecutive patients with diagnosed SEA between January 2011 and December 2017 to characterize epidemiological trends as well as the complement of baseline measures that are predictive of failure after multimodal treatment in patients with and without concomitant IVDU. The independent impact of clinical and imaging factors in detecting treatment failure was assessed by performing stepwise binary logistic regression analysis.

RESULTS

A total of 324 consecutive patients with diagnosed SEA were identified. Overall, 226 patients (69.8%) had SEA related to other causes and 98 (30.2%) had a history of recent IVDU. While non-IVDU SEA admission rates remained constant, year-over-year admissions of patients with IVDU SEA nearly tripled. At baseline, patients with IVDU SEA were distinct in many respects including younger age, greater unemployment and disability, less frequent diabetes mellitus (DM), and more frequent methicillin-resistant Staphylococcus aureus infection. However, differences in length of stay, loss to follow-up, and treatment failure did not reach statistical significance between the groups. The authors constructed independent multivariate logistic regression models for treatment failure based on identified parameters in the two cohorts. For the non-IVDU cohort, the authors identified four variables as independent factors: DM, hepatitis B/C, osteomyelitis, and compression deformity severity. In contrast, for patients with IVDU, the authors identified three variables: albumin, endocarditis, and endplate destruction. Receiver operating characteristic and area under the curve (AUC) analyses were undertaken for the multivariate models predicting the likelihood of treatment failure in the two cohorts (AUC = 0.88 and 0.89, respectively), demonstrating that the derived models could adequately predict the risk of multimodal treatment failure. Treatment failure risk factor point scales were derived for the identified variables separately for both cohorts.

CONCLUSIONS

Patients with IVDU SEA represent a unique population with a distinct set of baseline parameters that predict treatment failure. Identification of relevant prognosticating factors will allow for the design of tailored treatment and follow-up regimens.

In Brief

In this study, the authors identify the relevant baseline parameters that predict multimodal treatment failure in patients with either intravenous drug use (IVDU)–associated or non-IVDU–associated spinal epidural abscess (SEA). Patients with IVDU SEA represent a unique population with a distinct set of baseline parameters (including albumin levels as well as concomitant endocarditis and endplate destruction) that predict treatment failure. Knowledge of relevant prognosticating factors will allow for the design of tailored treatment and follow-up regimens.

Spinal epidural abscess (SEA) is a collection of pyogenic material in the epidural space between the dura mater and the periosteum. Traditional risk factors for SEA include diabetes mellitus (DM), chronic renal failure, chronic steroid use, alcoholism, and recent spinal procedures.1 Historically, the incidence of SEA has been estimated at 0.2–1.2 per 10,000 hospital admissions annually.2 However, since 2005, in part because of the expanding national opioid epidemic and an associated increase in intravenous drug use (IVDU), the incidence of SEA has increased more than threefold nationally.3,4

Despite the availability of combinatorial therapeutic strategies including intravenous antibiotics and surgical intervention,5–9 treatment failure is variously estimated to occur in 26.3%–35.1% of cases, often with grave consequences.10 Patients with IVDU-associated SEA represent a particularly challenging group for clinicians given their concomitant social stressors, frequent hospitalizations, and discharges against medical advice.4 Despite the increasing incidence of IVDU SEA, there is a paucity of knowledge regarding the distinguishing baseline demographic, clinical, laboratory, and radiological measures that specifically characterize these distinct patients as compared to these measures in patients with non-IVDU SEA.11 Further complicating the development of rational treatment regimens is the failure of studies to have defined a uniform range of SEA treatment failures10 despite identified risk factors including an age over 65 years, DM, methicillin-resistant Staphylococcus aureus (MRSA) infection, and neurological impairment on admission.12

In this study, we evaluated a large tertiary care center series of patients with SEA to characterize epidemiological trends as well as the complement of baseline measures that are predictive of failure after multimodal treatment in patients with and without concomitant IVDU. Identification of relevant prognosticating factors will allow for the design of optimized treatment regimens for patients with diagnosed SEA of various etiologies.

Methods

Study Population

Ohio State University Institutional Review Board approval was obtained, and all guidelines were followed throughout the course of the study. We used our institutional Information Warehouse database to identify consecutive patients admitted to The Ohio State University Wexner Medical Center with a primary diagnosis of SEA from January 1, 2011, through December 31, 2017. International Classification of Diseases, Ninth (code 324.1) and Tenth (code G06.1) Revision codes for intraspinal abscess were used to identify patients. Electronic medical records were reviewed in all cases to confirm radiographic evidence of SEA for study inclusion. All patients with less than 1 full year of clinical follow-up from the time of initial admission were defined as lost to follow-up and were excluded from the outcome analysis.

Study Parameters

We recorded baseline demographics (age, sex, race, and employment status), habits (IVDU, opioid use, stimulant use, and smoking status), and comorbidities (DM, HIV status, hepatitis, endocarditis, organ transplant history, and history of chemotherapy over the prior 3 months). Examination findings on presentation (fever defined as admission temperature > 98.6°F, motor deficits, and bladder dysfunction) and hospital length of stay (in days) were also recorded. The recorded laboratory data included initial white blood cell (WBC) counts, erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), and albumin (g/dl) levels. Culture data obtained from blood samples, image-guided biopsies, and surgical samples were recorded, as available. Treatment modalities included antibiotics and supplemental surgical decompression with or without instrumented arthrodesis.

Several CT and MRI study parameters were measured (Fig. 1) including location of the abscess within the spine (cervical, cervicothoracic, thoracic, thoracolumbar, lumbosacral, and multifocal), location of the abscess within the canal (anterior, posterior, circumferential), presence of concurrent discitis and/or osteomyelitis at the level of the abscess, and presence of vertebral bony endplate destruction at the level of the abscess. We measured overall length and maximal width of the abscess (mm) as a surrogate measure of the severity of local infection. Further, as a measure of local bone density, we evaluated Hounsfield unit values at the midvertebral body level closest to, as well as two vertebral body levels away from, the area of maximal abscess thickness. Finally, we assessed for the presence and severity of associated compression deformities as measured by the vertebral body Cobb angle at the level of maximal abscess thickness (defined as 0° in the absence of a compression deformity).

FIG. 1.
FIG. 1.

Representative spinal images obtained in patients with SEA, illustrating various baseline imaging parameters. A: SEA length and maximal width measurements (arrows) in a patient with lumbar discitis and osteomyelitis but relative bony endplate preservation. B: SEA with severe lumbar endplate destruction (arrows). C: Sample average Hounsfield unit calculation at the level of a lumbar SEA. D–F: Compression deformities of mild, moderate, and severe levels as assessed by vertebral body Cobb angle calculations (12°, 19°, and 30°, respectively). Figure is available in color online only.

Outcome Variables

We opted to include three specific initial treatment failure patterns in our analysis to more generally reflect real-world clinical circumstances. Clinical failure of initial treatment requiring readmission was defined as 1) documented laboratory and/or radiographic progression of an SEA requiring further intervention, 2) increased kyphotic deformity at the level of the abscess with worsened pain requiring further intervention, or 3) hardware failure defined as the subsidence of grafts or loosening of implants requiring revision surgery (in patients who underwent surgery as part of their initial treatment course).

Statistical Analysis

Descriptive statistical methods were used to identify demographic and study parameters. Continuous data are presented as means with standard deviations. Categorical data are presented as the number of subjects in the category, along with the percentages. Chi-square tests were used to compare the impact of clinical factors on outcomes in univariate analyses. Univariate logistic regression analysis was conducted to identify potential factors that predict the likelihood of treatment failure.

The independent impact of various clinical and imaging factors in detecting treatment failure was then assessed by performing stepwise binary logistic regression analysis. Multivariate receiver operating characteristic (ROC) analysis was conducted to evaluate the predictive ability of the models separately for the non-IVDU and IVDU SEA cohorts. This method is currently considered to be the gold standard to assess the accuracy of predictive distribution models.13 Therefore, we constructed independent multivariate models evaluating the combined effect of various distinct admission parameters by ROC analysis for the non-IVDU and IVDU SEA cohorts.

The development of the points-based prediction model to determine the risk of treatment failure was undertaken as described in detail by Sullivan and colleagues.14 The identified risk factors from the binary logistic regression were categorized separately for each cohort, and reference values were determined for each factor. Next, a referent risk factor profile was determined by setting a base category for each continuous risk factor that was assigned 0 points in the scoring system (e.g., for the IVDU SEA cohort, albumin > 3.4 g/dl). After calculating how far each risk factor category was from the base category in regression units, we set a constant value (i.e., the number of regression units corresponding to 1 point) to reflect the increased risk associated with a decrease of, for instance, 0.5 g/dl albumin. Points associated with each of the risk factor categories were then calculated. Finally, we derived a separate list for each cohort with the risk corresponding to each possible point total. The Hosmer-Lemeshow goodness-of-fit test and area under the multivariate ROC curves were used to confirm the performance of the derived prediction models.

The sensitivity, specificity, positive predictive value, negative predictive value, Youden Index, and cutoff points were calculated for the most predictive set of baseline factors (as identified by the ROC analysis). The Youden Index is a commonly used measure of overall predictive effectiveness. The maximum of the Youden Index occurs at the optimal cutoff point for each model, thus maximizing the number of correctly classified individuals.15

A p value < 0.05 was considered statistically significant. All data were analyzed using IBM SPSS Statistics version 26 (IBM Corp).

Results

Patient Characteristics

We identified a total of 324 consecutive patients with diagnosed SEA at our institution over the 7-year study period. Overall, 226 patients (69.8%) had SEA related to other causes, whereas 98 patients (30.2%) had a history of current IVDU. Annual admission numbers from 2011 to 2017 are shown in Fig. 2. While non-IVDU SEA admission rates remained constant, year-over-year admissions of patients with IVDU SEA nearly tripled in 2016 and 2017. Over half (54.6%) of this patient cohort presented in these last 2 years, which correspond to a period of rapidly increasing drug overdose deaths in Ohio.16

FIG. 2.
FIG. 2.

Yearly admission of patients with SEA either without (black line, n = 226) or with (gray line, n = 98) a history of IVDU, expressed as a percentage of each cohort.

Baseline demographic and clinical characteristics are outlined in Table 1. The mean age of the two cohorts was significantly different: non-IVDU, 60.2 ± 12.3 years, range 22.9–92.5 years; and IVDU, 44.9 ± 11.3 years, range 21.5–70.0 years (p < 0.001). The majority of patients in both groups were male (53.1% vs 61.2%, p = 0.18) and White (85.8% vs 87.8%, p = 0.71). Employment status was significantly different between the two cohorts, with 40.3% versus 88.8% unemployed or disabled (p < 0.001).

TABLE 1.

Baseline demographic and clinical characteristics of patients without or with IVDU-associated SEA

VariableNon-IVDU SEAIVDU SEAp Value
No. of patients22698
Age in yrs60.2 ± 12.3 (22.9–92.5)44.9 ± 11.3 (21.5–70.0)<0.001
Male120 (53.1)60 (61.2)0.18
Race0.71
 White194 (85.8)86 (87.8)
 African American27 (11.9)9 (9.2)
 Other5 (2.2)3 (3.1)
Employment<0.001
 Unemployed/disabled91 (40.3)87 (88.8)
 Full-time/part-time39 (17.3)9 (9.2)
 Retired94 (41.6)2 (2.0)
 Unknown2 (0.9)0 (0.0)
Habit
 Opioid use139 (61.5)45 (45.9)<0.01
 Stimulant use1 (0.4)41 (41.8)<0.001
 Smoker120 (53.1)91 (92.9)<0.001
Medical history
 DM98 (43.4)16 (16.3)<0.001
 HIV4 (1.8)1 (1.0)0.62
 Hepatitis B/C19 (8.4)49 (50.0)<0.001
 Chemo19 (8.4)5 (5.1)0.30
 Organ transplant3 (1.3)0 (0)0.25
Examination
 Fever80 (35.4)30 (30.6)0.40
 Motor deficit55 (24.3)23 (23.5)0.87
 Bladder dysfunction36 (15.9)16 (16.3)0.93
Treatment0.19
 Antibiotics alone102 (45.1)52 (53.1)
 Antibiotics & surgery124 (54.9)46 (46.9)
LOS in days16.2 ± 12.6 (2.0–73.0)17.8 ± 12.6 (1.0–62.0)0.29
Loss to FU36 (15.9)21 (21.4)0.23
Treatment failure*0.58
 SEA progression34 (17.9)22 (28.6)
 Kyphotic deformity12 (6.3)4 (5.2)
 Hardware failure5 (2.6)3 (3.9)
Mortality*
 30-day8 (4.2)2 (2.6)0.73
 60-day13 (6.8)3 (3.9)0.27
 1-yr22 (11.6)6 (7.8)0.25

Chemo = chemotherapy; FU = follow-up; LOS = length of stay.

Values expressed as mean ± standard deviation (range) or number (%), unless indicated otherwise. Boldface type indicates statistical significance.

Excludes patients lost to follow-up.

Subjects with non-IVDU SEA were significantly more likely to have current prescribed oral opioid use (61.5% vs 45.9%, p < 0.01) but less likely to have a history of stimulant use (0.4% vs 41.8%, p < 0.001) or smoking (53.1% vs 92.9%, p < 0.001). Medical history variables such as DM (43.4% vs 16.3%, p < 0.001) and hepatitis B and/or C (8.4% vs 50.0%, p < 0.001) were significantly different between the cohorts, whereas the relative incidence of HIV positivity, recent chemotherapy, and organ transplant history did not demonstrate a statistically significant difference. Further etiological analysis of the non-IVDU SEA cohort based on infectious disease specialist evaluations (not shown in Table 1) revealed that 46 patients (20.4%) had no identifiable risk factors, 35 (15.5%) had a general medical history–based risk factor (such as DM, pancreatitis, and/or alcoholism), 33 (14.6%) had recent spinal surgery, 24 (10.6%) had urinary tract infection– or pneumonia-associated bacteremia, 24 (10.6%) had chronic superficial nonhealing wounds or cellulitis, 21 (9.3%) had a recent diagnostic or interventional procedure (most commonly an epidural spinal injection), 20 (8.8%) had infected central lines or hemodialysis catheters, 15 (6.6%) were on immunosuppressants, 6 (2.7%) had tooth abscesses, poor dentition, or recent tooth extractions, and 2 (0.9%) had an esophageal perforation.

The relative incidence of admission clinical assessment findings was not significantly different between the two cohorts, including in terms of the presence of fever, motor deficit, or bladder dysfunction. Furthermore, treatment paradigm (antibiotics without or with surgery on admission), length of stay (days), or subtype treatment failure rates (overall 51/190 [26.8%] vs 29/77 [37.7%]) did not differ between the two cohorts. A wide variety of antibiotic regimens were used in the non-IVDU and IVDU SEA cohorts (not statistically different, p = 0.88), including vancomycin (19.5% vs 21.4%, respectively), nafcillin (16.0% vs 22.4%), vancomycin and third-generation cephalosporin (9.7% vs 6.1%), daptomycin (7.5% vs 7.1%), ceftriaxone (7.1% vs 6.1%), and penicillin (5.8% vs 5.1%). Loss to follow-up (15.9% vs 21.4%, p = 0.23) was not significantly different between the two cohorts. Finally, mortality rates at 30 days, 60 days, and 1 year were not significantly different between the two cohorts.

Table 2 summarizes the various baseline laboratory (including pooled culture data) and imaging characteristics of patients with either non-IVDU or IVDU SEA. Whereas mean WBC, ESR, CRP, and albumin levels did not differ between the two cohorts, MRSA, Pseudomonas, and fungal infections (15.0% vs 28.6%, p < 0.005; 0.4% vs 3.1%, p < 0.05; 0% vs 5.1%, p = 0.001) were less common, and Streptococcus species (9.7% vs 3.1%, p = 0.04) were more common in patients with non-IVDU SEA. The incidences of bacteremia and endocarditis on admission were not significantly different between the two cohorts. In contrast, several imaging parameters differed significantly between the two cohorts, including a cervicothoracic SEA location, SEA length, and remote level bone density as measured in Hounsfield units, all greater in the IVDU SEA cohort (3.1% vs 9.2%, p < 0.05; 50.2 ± 39.6 vs 66.9 ± 47.1 mm, p < 0.005; 196.3 ± 82.7 vs 228.8 ± 97.7 mm, p < 0.05). Lumbosacral and anterior SEA locations were most common for both non-IVDU and IVDU patients. Discitis and/or osteomyelitis were more commonly noted in both cohorts than either associated bony endplate destruction or compression deformity.

TABLE 2.

Baseline laboratory and imaging measures of patients without or with IVDU-associated SEA

VariableNon-IVDU SEAIVDU SEAp Value
No. of patients22698
WBC in ×109 cells/L12.8 ± 8.5 (1.5–97.2)13.5 ± 7.2 (4.0–36.4)0.46
ESR in mm/hr77.1 ± 35.1 (9.0–140.0)75.1 ± 32.1 (11.0–140.0)0.62
CRP in mg/L138.6 ± 107.4 (1.0–526.1)127.2 ± 102.0 (1.0–451.5)0.38
Albumin in g/dl2.8 ± 0.7 (1.0–4.4)2.9 ± 0.7 (1.3–4.2)0.89
Organism
 No growth63 (27.9)21 (21.4)0.22
 MSSA52 (23.0)24 (24.5)0.77
 MRSA34 (15.0)28 (28.6)<0.005
Streptococcus 22 (9.7)3 (3.1)<0.05
Staphylococcus epidermidis 15 (6.6)2 (2.0)0.09
 Other Staphylococcus species3 (1.3)0 (0)0.25
Enterococcus 4 (1.8)1 (1.0)0.62
Escherichia coli 6 (2.7)1 (1.0)0.35
Pseudomonas 1 (0.4)3 (3.1)<0.05
Klebsiella 3 (1.3)1 (1.0)0.82
Mycobacterium 4 (1.8)0 (0)0.19
 Polymicrobial16 (7.1)6 (6.1)0.75
 Fungal0 (0)5 (5.1)0.001
 Other3 (1.3)3 (3.1)0.29
Bacteremia114 (50.4)57 (58.2)0.20
Endocarditis14 (6.2)11 (11.2)0.12
Imaging
 Cervical30 (13.3)14 (14.3)0.81
 Cervicothoracic7 (3.1)9 (9.2)<0.05
 Thoracic72 (31.9)25 (25.5)0.25
 Thoracolumbar16 (7.1)7 (7.1)0.98
 Lumbosacral97 (42.9)42 (42.9)0.99
 Multifocal4 (1.8)1 (1.0)0.62
 Anterior134 (59.3)60 (61.2)0.74
 Posterior53 (23.5)24 (24.5)0.84
 Circumferential39 (17.3)14 (14.3)0.51
 Abscess length in mm50.2 ± 39.6 (6.2–269.3)66.9 ± 47.1 (5.0–319.0)<0.005
 Abscess width in mm6.6 ± 2.6 (1.5–19.9)6.9 ± 3.3 (1.9–19.2)0.37
 Discitis147 (65.0)62 (63.3)0.76
 Osteomyelitis145 (64.2)64 (65.3)0.84
 Endplate destruction94 (41.6)40 (40.8)0.90
 Local level bone density in HU270.4 ± 111.7 (74.3–564.0)307.6 ± 129.2 (88.7–655.0)0.05
 Remote level bone density in HU196.3 ± 82.7 (34.0–489.0)228.8 ± 97.7 (101.0–542.0)<0.05
 Compression deformity: yes/no84 (37.2)41 (41.8)0.43
 Compression deformity in °4.3 ± 7.16.0 ± 9.10.08

HU = Hounsfield unit; MSSA = methicillin-sensitive Staphylococcus aureus.

Values expressed as mean ± standard deviation (range) or number (%), unless indicated otherwise. Boldface type indicates statistical significance.

Of note, even though the noted loss to follow-up rates were within the range expected for such patient populations,4 we undertook statistical analyses comparing the incidence of several relevant baseline parameters in patients who were not lost to follow-up with that in patients who were lost to follow-up to assess for potential selection bias. For the non-IVDU SEA cohort, no significant differences were noted in any of the relevant baseline parameters (see below) including DM (p = 0.07), hepatitis B/C (p = 0.99), osteomyelitis (p = 0.43), and compression deformity (p = 0.13). Similarly, for the IVDU SEA cohort, no significant differences were noted for various relevant factors including albumin (p = 0.70), endocarditis (p = 0.78), or endplate fracture (p = 0.78).

Univariate Analyses

Given the above set of cohort-specific characteristics, we performed separate univariate analyses to determine the impact of various baseline parameters in predicting multimodal treatment failure for patients with either non-IVDU or IVDU SEA (Supplemental Table 1). Several cohort-specific associations were noted for clinical, laboratory, and radiological parameters. For patients with non-IVDU SEA, opioid use (p < 0.05), DM (p = 0.001), MRSA infection (p = 0.05), posterior SEA location (p < 0.05), discitis (p < 0.001), compression deformity (p < 0.001), and compression deformity severity (p < 0.001) were all predictive of treatment failure on univariate analyses. In contrast, for patients with IVDU SEA, albumin level (p < 0.005) and the presence of endocarditis on admission (p < 0.05) were predictive of treatment failure. Hepatitis B/C (p < 0.001 for each group), osteomyelitis (p < 0.001 and < 0.05, respectively), and endplate destruction (p < 0.001 and < 0.01, respectively) were predictive of treatment failure for both cohorts on univariate analyses.

Multivariate Analyses

Several unique and shared variables were identified as predictive of multimodal treatment failure for patients with either non-IVDU or IVDU SEA (with p < 0.05 on univariate analysis). For each cohort, tests of collinearity were separately undertaken for the variables included in the multivariate model to confirm that the predictor variables were not colinear, that is, that they were not strongly related to one another as indicated by variance inflation factors < 1.3 in all cases. Therefore, we constructed independent multivariate logistic regression models based on these identified parameters for patients with either non-IVDU or IVDU SEA (Table 3). For patients with non-IVDU SEA, we identified four variables as independent factors: DM (OR 3.470, p = 0.002), hepatitis B/C (OR 14.434, p = 0.002), osteomyelitis (OR 42.127, p = 0.001), and compression deformity severity (OR 1.06, p = 0.02). In contrast, for patients with IVDU SEA, we identified three variables as independent factors: albumin level (OR 0.198, p = 0.015), endocarditis (OR 11.171, p = 0.025), and endplate destruction (OR 14.109, p = 0.004). The Hosmer-Lemeshow goodness-of-fit tests were not significant for either model. We further validated the two models by testing for significant differences between included patients and those lost to follow-up in terms of the variables used in the multivariate analyses. Only DM was found to be statistically significantly different in the subsample of non-IVDU SEA patients lost to follow-up, but this variable was already included in the analysis. For the IVDU SEA subsample lost to follow-up, only compression deformity and hepatitis were found to be significantly different. However, compression deformity was not significant in the univariate analysis and therefore was already removed from the multivariate analysis. With respect to hepatitis, inclusion of this variable did not result in any significant performance improvement of the multivariate model.

TABLE 3.

Multivariate analysis of baseline parameters for patients with either non-IVDU or IVDU SEA

VariableBSE BWald Statisticp ValueOR95% CI OR
Non-IVDU SEA
 DM1.2440.4099.2590.0023.4701.557–7.734
 Hepatitis B/C2.6700.8609.6370.00214.4342.675–77.876
 Osteomyelitis3.7411.10111.5420.00142.1274.868–364.577
 Compression deformity severity0.060.035.100.021.061.01–1.12
 Intercept−5.3381.11822.8160.0000.005
IVDU SEA
 Albumin (g/dl)−1.6180.6645.9350.0150.1980.054–0.729
 Endocarditis2.4131.0775.0220.02511.1711.353–92.205
 Endplate destruction2.6470.9088.4990.00414.1092.381–83.624
 Intercept2.4081.8221.7480.18611.116

B = unstandardized beta; SE B = standard error of unstandardized beta.

Boldface type indicates statistical significance.

As a next step, ROC and corresponding area under the curve (AUC) analyses were undertaken for the multivariate models predicting the likelihood of treatment failure in patients with non-IVDU SEA (AUC = 0.88) and with IVDU SEA (AUC = 0.89), demonstrating that the derived models could adequately predict the risk of multimodal treatment failure (Fig. 3). The resultant risk factor scale, whose point system ranges from 0 to 32 points for patients with non-IVDU SEA and from 0 to 11 for patients with IVDU SEA, is shown in Table 4. The respective point ranges and corresponding predicted relative treatment failure risks for the two cohorts are shown in Table 5.

FIG. 3.
FIG. 3.

ROC and corresponding AUC analysis for the multivariate models predicting the likelihood of treatment failure in patients with non-IVDU SEA (left) and with IVDU SEA (right).

TABLE 4.

Risk factor point scale for the prediction of multimodality treatment failure in patients with non-IVDU or IVDU SEA

Non-IVDU SEAIVDU SEA
Risk FactorPointsRisk FactorPoints
DMAlbumin (g/dl)
 No0 >3.40
 Yes4 3.0–3.41
Hepatitis B/C 2.5–2.92
 No0 2.0–2.43
 Yes9 1.5–1.94
Osteomyelitis <1.55
 No0Endocarditis
 Yes13 No0
Compression deformity severity Yes3
 None0Endplate destruction
 Mild (5.0°–14.9°)2 No0
 Moderate (15°–24.9°)4 Yes3
 Severe (>25°)6
TABLE 5.

Predicted risk of multimodal treatment failure for patients with non-IVDU and IVDU SEA

Non-IVDU SEAIVDU SEA
Point TotalRiskPoint TotalRisk
00.00500.024
20.01010.054
40.01820.114
60.03230.225
80.05740.395
90.07650.595
100.10060.767
130.21570.881
150.33480.943
170.47790.974
190.624100.988
210.752110.994
220.803
230.846
240.881
260.931
280.961
320.987

At a cutoff value of 15 for the non-IVDU SEA cohort, the model correctly classified failed treatment with 86.3% accuracy (sensitivity) and no treatment failure with 73.4% accuracy (specificity). The calculated maximum Youden Index is 0.60 for this model. At a cutoff value of 4 for the IVDU SEA cohort, the model correctly classified failed treatment with 90% accuracy (sensitivity) and no treatment failure with 81.6% accuracy (specificity). The calculated maximum Youden Index is 0.71 for this model.

Discussion

SEA is a potentially devastating entity that can elude detection until the onset of neurological compromise. Early diagnosis and treatment are of paramount importance prior to abscess enlargement potentially causing neurological compromise from mechanical compression and/or vascular infarction. Despite aggressive multimodal treatment protocols at tertiary care centers involving prolonged antibiotic administration and surgical intervention, treatment failure remains common, often leading to permanent disability and even death.10

Our baseline demographic data confirm the traditional impression that SEA more often presents in older patients with chronic conditions, including DM (Table 1). Nevertheless, the relative proportion of SEA patients with concomitant IVDU markedly increased in the final 2 years of our analysis (Fig. 2), consistent with the expanding opioid crisis in Ohio that in turn has led to rapidly increasing illicit drug use and overdose deaths.16 Overall, US deaths related to heroin overdose began to rise in 2011 and plateaued in 2016 at a rate of 6 deaths per 100,000 population. The increase in heroin deaths likely reflects overall increases in heroin use, which in turn explains the rise in hospital admissions due to IVDU SEA at institutions like ours during this period.

In our study, patients with IVDU SEA were younger than those with non-IVDU SEA by an average of 15 years, just under 90% were unemployed or on disability, and 50% were also diagnosed with hepatitis B or C. Average WBC counts were only mildly elevated, typically ranging between 12 and 13 × 109 cells/L, thus highlighting the relative lack of diagnostic sensitivity of this measure for SEA. Although ESR and CRP levels were not statistically different between the two cohorts, both values were usually elevated on presentation and thus represent critical tests in suspected SEA cases. Cultures were negative in 27.9% and 21.4% of non-IVDU and IVDU SEA cases, respectively. Although antibiotics administered prior to culture analysis likely confounded these data, significantly higher rates of MRSA, Pseudomonas, and fungal infections were noted in patients with IVDU SEA. The majority of SEAs in our cohort involved either the thoracic or lumbosacral spine, and just under 15% occurred in the cervical spine.

These findings from a large single-center cohort are generally consistent with national registry and other single-center data,4,5,10,11,17–19 supporting the overall generalizability of our findings. However, IVDU SEA in our study was not associated with statistically significant increases in length of stay, loss to follow-up, or treatment failure, which is in contrast to results in prior studies.4,11 Our findings, at least in part, reflect the advanced multidisciplinary care offered at our academic center along with the relative lack of other options available to these patients because of insurance-related constraints (particularly for the IVDU SEA cohort).

Patients with IVDU SEA often present with unique social stressors, undergo frequent hospitalizations, and may pursue discharge against medical advice.18 In part related to these challenges, no study to date has specifically evaluated the comparative admission variables that stratify the risk for treatment failure following standard-of-care multimodal therapy in patients with non-IVDU and IVDU SEA. Given the large single-center nature of our cohort with adequate follow-up rates, we were able for the first time to assess in parallel the differential predictive ability of various baseline demographic, clinical, laboratory, and imaging measures by using multivariate logistic regression analysis.

Thus, we identified a distinct set of four baseline parameters for patients with non-IVDU SEA (DM, hepatitis B/C, osteomyelitis, and compression deformity severity) that accurately predict multimodal treatment failure (AUC = 0.88, calculated maximum Youden Index = 0.60). To our knowledge, only DM in this set of parameters has been linked to medical treatment failure.12,20 While the influence of acute or chronic hepatitis B/C infection may be related to the virus’s complex effects on patients’ immune function,21 underlying osteomyelitis and compression deformity severity are likely more related to the interplay between the infecting organism’s virulence and an array of patient-specific spine biomechanical and bone metabolic characteristics. Our statistical analysis showed that the combination of these other three medical history and radiological factors along with DM in a 32-point risk factor scale (Tables 4 and 5) can serve as a reliable prognostication tool to aid physicians in deriving tailored treatment and follow-up regimens.

Prior studies have confirmed that a history of IVDU renders patients more susceptible to developing spinal infections with drug-resistant bacteria, making their treatment course more difficult and thus potentially leading to poorer outcomes.10,12 MRSA in particular has been identified as an independent risk factor for treatment failure in patients with SEA.12 Pseudomonas has also been shown to develop hypermutability that can make it resistant to antibiotics and markedly increase its virulence.17,22 However, infection with neither of these organisms significantly contributed to treatment failure prediction for patients with IVDU SEA even in our univariate analysis. Rather, multivariate logistic regression supported the inclusion of three independent admission variables for the IVDU SEA cohort (albumin level, endocarditis, and endplate destruction) to accurately predict multimodal treatment failure (AUC = 0.89, calculated maximum Youden Index = 0.71).

Recent studies using both institutional and registry-based data sets have identified low admission albumin levels as an independent predictor of 30- and 90-day postdischarge complications and death.23,24 Interestingly, in our study, while admission albumin values were low in both cohorts, only the levels for patients with IVDU SEA were highly predictive of treatment failure in the multivariate model, potentially indicating the prime importance of a patient’s nutritional and catabolic status in determining recovery in this younger group. IVDU-associated endocarditis (identified in 11.2% of patients in our cohort) has increased in incidence over the past decade25 and represents another newly identified independent risk factor for treatment failure in patients with IVDU SEA. While these high-risk patients frequently undergo combinatorial treatment regimens including surgery, our data indicated that concomitant cardiac involvement likely predisposes to refractory spinal disease. Finally, inclusion of bony endplate destruction, instead of the more widely recognized findings of either discitis or osteomyelitis assessed in prior studies,11 yielded the strongest prediction model for this cohort. Thus, endplate fractures may represent an imaging marker that is more reflective of an organism’s virulence in patients with IVDU SEA. Thus, considering admission albumin levels in combination with the presence of endocarditis and endplate destruction can serve as a reliable prognostication tool using an 11-point scale to aid physicians in deriving tailored management regimens for this particularly challenging group of patients.

Conclusions

Taken together, our logistic regression analyses strongly support the integration of distinct sets of admission parameters into clinical models for patients with either non-IVDU SEA and IVDU SEA to help predict failure after initial multimodal treatment paradigms. The derived risk factor point scales for the two cohorts can assist in patient counseling, clinical decision-making, and tailoring follow-up plans for patients after their initial course of multimodal treatments. Further prospective and larger-scale multiinstitutional studies are required to more precisely define these admission parameters as reliable predictors of treatment failure.

Disclosures

Dr. Viljoen is a consultant for and a member of the speakers bureau for Johnson & Johnson and is a consultant for Medtronic. Dr. Farhadi has received support from DePuy Synthes, Implanet America, and Nexxt Spine for non–study-related clinical or research effort.

Author Contributions

Conception and design: Farhadi, Viljoen, Minnema, Shaikhouni, Grossbach, Nimjee. Acquisition of data: Farhadi, Baum, Gifford. Analysis and interpretation of data: Farhadi, Viljoen, Gifford, Shaikhouni, Grossbach, Nimjee. Drafting the article: Baum, Gifford. Critically revising the article: Farhadi, Viljoen, Shaikhouni. Reviewed submitted version of manuscript: Shaikhouni, Grossbach, Nimjee. Statistical analysis: Baum, Gifford. Study supervision: Farhadi.

Supplemental Information

Online-Only Content

Supplemental material is available with the online version of the article.

References

  • 1

    Darouiche RO . Spinal epidural abscess. N Engl J Med. 2006;355(19):20122020.

  • 2

    Baker AS, Ojemann RG, Swartz MN, Richardson EP Jr . Spinal epidural abscess. N Engl J Med. 1975;293(10):463468.

  • 3

    Artenstein AW, Friderici J, Holers A, Lewis D, Fitzgerald J, Visintainer P . Spinal epidural abscess in adults: a 10-year clinical experience at a tertiary care academic medical center. Open Forum Infect Dis. 2016;3(4):ofw191.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 4

    Toppo AJ, Rogerson A, Oh DHW, Tybor DJ, Wurcel AG, Salzler MJ . Injection drug use in patients with spinal epidural abscess: nationwide data, 2000 to 2013.Spine ( Phila Pa 1976).2020;45(12):843850.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 5

    Adogwa O, Karikari IO, Carr KR, Krucoff M, Ajay D, Fatemi P, et al. Spontaneous spinal epidural abscess in patients 50 years of age and older: a 15-year institutional perspective and review of the literature: clinical article. J Neurosurg Spine. 2014;20(3):344349.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 6

    Mampalam TJ, Rosegay H, Andrews BT, Rosenblum ML, Pitts LH . Nonoperative treatment of spinal epidural infections. J Neurosurg. 1989;71(2):208210.

  • 7

    Siddiq F, Chowfin A, Tight R, Sahmoun AE, Smego RA Jr . Medical vs surgical management of spinal epidural abscess. Arch Intern Med. 2004;164(22):24092412.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 8

    Connor DE Jr, Chittiboina P, Caldito G, Nanda A . Comparison of operative and nonoperative management of spinal epidural abscess: a retrospective review of clinical and laboratory predictors of neurological outcome. J Neurosurg Spine. 2013;19(1):119127.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 9

    Ghobrial GM, Beygi S, Viereck MJ, Maulucci CM, Sharan A, Heller J, et al. Timing in the surgical evacuation of spinal epidural abscesses. Neurosurg Focus. 2014;37(2):E1.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 10

    Stratton A, Gustafson K, Thomas K, James MT . Incidence and risk factors for failed medical management of spinal epidural abscess: a systematic review and meta-analysis. J Neurosurg Spine. 2017;26(1):8189.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 11

    Shah AA, Yang H, Ogink PT, Schwab JH . Independent predictors of spinal epidural abscess recurrence. Spine J. 2018;18(10):18371844.

  • 12

    Kim SD, Melikian R, Ju KL, Zurakowski D, Wood KB, Bono CM, Harris MB . Independent predictors of failure of nonoperative management of spinal epidural abscesses. Spine J. 2014;14(8):16731679.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 13

    Hanley JA, McNeil BJ . The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):2936.

  • 14

    Sullivan LM, Massaro JM, D’Agostino RB Sr . Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med. 2004;23(10):16311660.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 15

    Schisterman EF, Perkins NJ, Liu A, Bondell H . Optimal cut-point and its corresponding Youden Index to discriminate individuals using pooled blood samples. Epidemiology. 2005;16(1):7381.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 16

    Ohio: Opioid-involved deaths and related harms. National Institute on Drug Abuse. Updated April 3, 2020.Accessed August 12, 2021. https://www.drugabuse.gov/drug-topics/opioids/opioid-summaries-by-state/ohio-opioid-involved-deaths-related-harms

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 17

    Chuo CY, Fu YC, Lu YM, Chen JC, Shen WJ, Yang CH, Chen CY . Spinal infection in intravenous drug abusers. J Spinal Disord Tech. 2007;20(4):324328.

  • 18

    Wang Z, Lenehan B, Itshayek E, Boyd M, Dvorak M, Fisher C, et al. Primary pyogenic infection of the spine in intravenous drug users: a prospective observational study. Spine (Phila Pa 1976).2012;37(8):685692.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 19

    Shah AA, Karhade AV, Bono CM, Harris MB, Nelson SB, Schwab JH . Development of a machine learning algorithm for prediction of failure of nonoperative management in spinal epidural abscess. Spine J. 2019;19(10):16571665.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 20

    Patel AR, Alton TB, Bransford RJ, Lee MJ, Bellabarba CB, Chapman JR . Spinal epidural abscesses: risk factors, medical versus surgical management, a retrospective review of 128 cases. Spine J. 2014;14(2):326330.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 21

    Shin EC, Sung PS, Park SH . Immune responses and immunopathology in acute and chronic viral hepatitis. Nat Rev Immunol. 2016;16(8):509523.

  • 22

    Hadjipavlou AG, Mader JT, Necessary JT, Muffoletto AJ . Hematogenous pyogenic spinal infections and their surgical management. Spine (Phila Pa 1976).2000;25(13):16681679.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 23

    Karhade AV, Shah AA, Lin KY, Ogink PT, Shah KC, Nelson SB, Schwab JH . Albumin and spinal epidural abscess: derivation and validation in two independent data sets. World Neurosurg. 2019;123:e416e426.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 24

    Goh BC, Ferrone ML, Barghi A, Liu CY, Cronin PK, Blucher JA, et al. The prognostic value of laboratory markers and ambulatory function at presentation for post-treatment morbidity and mortality following epidural abscess. Spine (Phila Pa 1976).2020;45(15):E959E966.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 25

    Wang A, Gaca JG, Chu VH . Management considerations in infective endocarditis: a review. JAMA. 2018;320(1):7283.

Supplementary Materials

  • Collapse
  • Expand
Illustration from Levi and Schwab (pp 653–659). Copyright Roberto Suazo. Published with permission.
  • FIG. 1.

    Representative spinal images obtained in patients with SEA, illustrating various baseline imaging parameters. A: SEA length and maximal width measurements (arrows) in a patient with lumbar discitis and osteomyelitis but relative bony endplate preservation. B: SEA with severe lumbar endplate destruction (arrows). C: Sample average Hounsfield unit calculation at the level of a lumbar SEA. D–F: Compression deformities of mild, moderate, and severe levels as assessed by vertebral body Cobb angle calculations (12°, 19°, and 30°, respectively). Figure is available in color online only.

  • FIG. 2.

    Yearly admission of patients with SEA either without (black line, n = 226) or with (gray line, n = 98) a history of IVDU, expressed as a percentage of each cohort.

  • FIG. 3.

    ROC and corresponding AUC analysis for the multivariate models predicting the likelihood of treatment failure in patients with non-IVDU SEA (left) and with IVDU SEA (right).

  • 1

    Darouiche RO . Spinal epidural abscess. N Engl J Med. 2006;355(19):20122020.

  • 2

    Baker AS, Ojemann RG, Swartz MN, Richardson EP Jr . Spinal epidural abscess. N Engl J Med. 1975;293(10):463468.

  • 3

    Artenstein AW, Friderici J, Holers A, Lewis D, Fitzgerald J, Visintainer P . Spinal epidural abscess in adults: a 10-year clinical experience at a tertiary care academic medical center. Open Forum Infect Dis. 2016;3(4):ofw191.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 4

    Toppo AJ, Rogerson A, Oh DHW, Tybor DJ, Wurcel AG, Salzler MJ . Injection drug use in patients with spinal epidural abscess: nationwide data, 2000 to 2013.Spine ( Phila Pa 1976).2020;45(12):843850.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 5

    Adogwa O, Karikari IO, Carr KR, Krucoff M, Ajay D, Fatemi P, et al. Spontaneous spinal epidural abscess in patients 50 years of age and older: a 15-year institutional perspective and review of the literature: clinical article. J Neurosurg Spine. 2014;20(3):344349.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 6

    Mampalam TJ, Rosegay H, Andrews BT, Rosenblum ML, Pitts LH . Nonoperative treatment of spinal epidural infections. J Neurosurg. 1989;71(2):208210.

  • 7

    Siddiq F, Chowfin A, Tight R, Sahmoun AE, Smego RA Jr . Medical vs surgical management of spinal epidural abscess. Arch Intern Med. 2004;164(22):24092412.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 8

    Connor DE Jr, Chittiboina P, Caldito G, Nanda A . Comparison of operative and nonoperative management of spinal epidural abscess: a retrospective review of clinical and laboratory predictors of neurological outcome. J Neurosurg Spine. 2013;19(1):119127.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 9

    Ghobrial GM, Beygi S, Viereck MJ, Maulucci CM, Sharan A, Heller J, et al. Timing in the surgical evacuation of spinal epidural abscesses. Neurosurg Focus. 2014;37(2):E1.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 10

    Stratton A, Gustafson K, Thomas K, James MT . Incidence and risk factors for failed medical management of spinal epidural abscess: a systematic review and meta-analysis. J Neurosurg Spine. 2017;26(1):8189.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 11

    Shah AA, Yang H, Ogink PT, Schwab JH . Independent predictors of spinal epidural abscess recurrence. Spine J. 2018;18(10):18371844.

  • 12

    Kim SD, Melikian R, Ju KL, Zurakowski D, Wood KB, Bono CM, Harris MB . Independent predictors of failure of nonoperative management of spinal epidural abscesses. Spine J. 2014;14(8):16731679.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 13

    Hanley JA, McNeil BJ . The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology. 1982;143(1):2936.

  • 14

    Sullivan LM, Massaro JM, D’Agostino RB Sr . Presentation of multivariate data for clinical use: the Framingham Study risk score functions. Stat Med. 2004;23(10):16311660.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 15

    Schisterman EF, Perkins NJ, Liu A, Bondell H . Optimal cut-point and its corresponding Youden Index to discriminate individuals using pooled blood samples. Epidemiology. 2005;16(1):7381.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 16

    Ohio: Opioid-involved deaths and related harms. National Institute on Drug Abuse. Updated April 3, 2020.Accessed August 12, 2021. https://www.drugabuse.gov/drug-topics/opioids/opioid-summaries-by-state/ohio-opioid-involved-deaths-related-harms

    • PubMed
    • Search Google Scholar
    • Export Citation
  • 17

    Chuo CY, Fu YC, Lu YM, Chen JC, Shen WJ, Yang CH, Chen CY . Spinal infection in intravenous drug abusers. J Spinal Disord Tech. 2007;20(4):324328.

  • 18

    Wang Z, Lenehan B, Itshayek E, Boyd M, Dvorak M, Fisher C, et al. Primary pyogenic infection of the spine in intravenous drug users: a prospective observational study. Spine (Phila Pa 1976).2012;37(8):685692.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 19

    Shah AA, Karhade AV, Bono CM, Harris MB, Nelson SB, Schwab JH . Development of a machine learning algorithm for prediction of failure of nonoperative management in spinal epidural abscess. Spine J. 2019;19(10):16571665.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 20

    Patel AR, Alton TB, Bransford RJ, Lee MJ, Bellabarba CB, Chapman JR . Spinal epidural abscesses: risk factors, medical versus surgical management, a retrospective review of 128 cases. Spine J. 2014;14(2):326330.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 21

    Shin EC, Sung PS, Park SH . Immune responses and immunopathology in acute and chronic viral hepatitis. Nat Rev Immunol. 2016;16(8):509523.

  • 22

    Hadjipavlou AG, Mader JT, Necessary JT, Muffoletto AJ . Hematogenous pyogenic spinal infections and their surgical management. Spine (Phila Pa 1976).2000;25(13):16681679.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 23

    Karhade AV, Shah AA, Lin KY, Ogink PT, Shah KC, Nelson SB, Schwab JH . Albumin and spinal epidural abscess: derivation and validation in two independent data sets. World Neurosurg. 2019;123:e416e426.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 24

    Goh BC, Ferrone ML, Barghi A, Liu CY, Cronin PK, Blucher JA, et al. The prognostic value of laboratory markers and ambulatory function at presentation for post-treatment morbidity and mortality following epidural abscess. Spine (Phila Pa 1976).2020;45(15):E959E966.

    • Crossref
    • PubMed
    • Search Google Scholar
    • Export Citation
  • 25

    Wang A, Gaca JG, Chu VH . Management considerations in infective endocarditis: a review. JAMA. 2018;320(1):7283.

Metrics

All Time Past Year Past 30 Days
Abstract Views 793 0 0
Full Text Views 787 366 59
PDF Downloads 504 180 8
EPUB Downloads 0 0 0