The Spine Journal 15 (2015) 95–101
Clinical Study
Predicting patients with concurrent noncontiguous spinal epidural abscess lesions Kevin L. Ju, MD*, Sang Do Kim, MD, MS, Rojeh Melikian, MD, Christopher M. Bono, MD, Mitchel B. Harris, MD Department of Orthopaedic Surgery, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115, USA Received 10 January 2014; revised 30 April 2014; accepted 8 June 2014
Abstract
BACKGROUND CONTEXT: Spinal epidural abscess (SEA) is a serious condition that can lead to significant morbidity and mortality if not expeditiously diagnosed and appropriately treated. However, the nonspecific findings that accompany SEAs often make its diagnosis difficult. Concurrent noncontiguous SEAs are even more challenging to diagnose because whole-spine imaging is not routinely performed unless the patient demonstrates neurologic findings that are inconsistent with the identified lesion. Failure to recognize a separate SEA can subject patients to a second operation, continued sepsis, paralysis, or even death. PURPOSE: To formulate a set of clinical and laboratory predictors for identifying patients with concurrent noncontiguous SEAs. STUDY DESIGN: A retrospective, case-control study. PATIENT SAMPLE: Patients aged 18 years or older admitted to our institution during the study period who underwent entire spinal imaging and were diagnosed with one or more SEAs. OUTCOME MEASURES: The presence or absence of concurrent noncontiguous SEAs on magnetic resonance imaging or computed tomography (CT)-myelogram. METHODS: A retrospective review was performed on 233 adults with SEAs who presented to our health-care system from 1993 to 2011 and underwent entire spinal imaging. The clinical and radiographic features of patients with concurrent noncontiguous SEAs, defined as at least two lesions in different anatomical regions of the spine (ie, cervical, thoracic, or lumbar), were compared with those with a single SEA. Multivariate logistic regression identified independent predictors for the presence of a skip SEA, and a prediction algorithm based on these independent predictors was constructed. Institutional review board committee approval was obtained before initiating the study. RESULTS: Univariate and multivariate analyses comparing patients with skip SEA lesions (n522) with those with single lesions (n5211) demonstrated significant differences in three factors: delay in presentation (defined as symptoms for $7 days), a concomitant area of infection outside the spine and paraspinal region, and an erythrocyte sedimentation rate of O95 mm/h at presentation. The predicted probability for the presence of a skip lesion was 73% for patients possessing all three predictors, 13% for two, 2% for one, and 0% for zero predictors. Receiver operating characteristic curve analysis, used to evaluate the predictive accuracy of the model, revealed a steep shoulder with an area under the curve of 0.936 (p!.001).
FDA device/drug status: Not applicable. Author disclosures: KLJ: Nothing to disclose. SDK: Nothing to disclose. RM: Nothing to disclose. CMB: Royalties: Wolters Kluwer (A); Consulting: United Health Care (B), Intrinsic Therapeutics, DSMB (B). MBH: Nothing to disclose. The disclosure key can be found on the Table of Contents and at www. TheSpineJournalOnline.com. http://dx.doi.org/10.1016/j.spinee.2014.06.008 1529-9430/Ó 2015 Elsevier Inc. All rights reserved.
There was no external funding source for this study, and the institutional funding did not influence the investigation. * Corresponding author. Department of Orthopaedic Surgery, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115, USA. Tel.: (617) 600-8498. E-mail address: [email protected] (K.L. Ju)
96
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
CONCLUSIONS: The proposed set of three predictors may be a useful tool in predicting the risk of a skip SEA lesion and, consequently, which patients would benefit from entire spinal imaging. Ó 2015 Elsevier Inc. All rights reserved. Keywords:
Epidural abscess; Skip lesion; Concurrent; Noncontiguous; Prediction; Algorithm
Introduction Spinal epidural abscess (SEA) is a serious disease that can rapidly lead to neurologic deficit, sepsis, or even death if not promptly diagnosed and appropriately treated. Although SEA is relatively uncommon, with previous estimated incidences ranging from 0.2 to 1.2 cases per 10,000 hospital admissions [1], more recent data indicate the incidence has increased to 12.5 cases per 10,000 admissions [2]. This rise is likely a reflection of an aging population, increasing prevalence of predisposing comorbidities such as diabetes mellitus and immunosuppression, and more widespread intravenous drug use (IVDU) and alcoholism [3,4]. Despite advances in imaging techniques, antibiotic therapy, and surgical techniques, the overall mortality of SEA is still about 5% [3], and up to 22% of patients suffer irreversible paralysis [3,5]. Timely diagnosis and initiation of appropriate treatment is key to reducing the morbidity and mortality associated with this potentially devastating disease [6–10]. Identifying patients with a single SEA lesion can be challenging as the most common presenting complaint is nonspecific back or neck pain [3,7]. In a metaanalysis of 915 patients with SEAs, about three-quarters had back or neck pain, half had a fever, and only a quarter had neurologic deficits [10]. The classic clinical triad of back pain, fever, and neurologic deficit is only present in a minority of patients with SEA and exhibits a diagnostic sensitivity of only 8% [7]. Simultaneous noncontiguous SEAs in different anatomical regions of the spine (ie, cervical, thoracic, or lumbar), also known as skip lesions, pose an even greater diagnostic challenge. Unless a patient exhibits signs and symptoms that are not concordant with a detected, single SEA, practitioners do not reflexively order additional imaging of other spinal regions [11,12]. Failing to detect a skip SEA can lead to continued sepsis or neurologic decline. Further complicating the diagnostic dilemma is the fact that additional SEAs may initially be asymptomatic or present with pain only. Pfister et al. [11] described a patient with skip SEAs (cervical and lumbar) in whom the lumbar SEA was missed on the initial work-up. It was not until the patient had undergone a cervical decompression and continued to deteriorate neurologically that a whole spine magnetic resonance imaging (MRI) was performed and the lumbar SEA lesion was identified. With the exception of case reports and small case series, there is a dearth of clinical data concerning skip SEAs [11,13,14]. The purpose of our study was to formulate a set of clinical and laboratory predictors to help
identify patients with skip SEAs in different regions of the spine based on a large retrospectively-collected database. Materials and methods Study design Cases of SEA treated within a large health-care system between 1993 and 2011 were retrospectively reviewed. Inclusion criteria were patients aged 18 years or older with an entire spinal imaging using MRI with IV contrast or computed tomography (CT) with myelogram and IV contrast and found to have a spinal epidural abscess with a fluid collection. Patients with spondylodiscitis or osteomyelitis with granulation tissue (ie, phlegmon) but without a clear epidural fluid collection were excluded. Secondary SEAs occurring in patients who previously had a lumbar puncture, epidural injection, laminectomy, or laminotomy at the level or adjacent levels to the SEA were excluded. Patients with insufficient information were also excluded. In total, 233 adults were included in the study. Two hundred thirty-one patients underwent entire spine MRIs with IV contrast, whereas two patients underwent CTmyelograms with IV contrast of the entire spine (one had a pacemaker and one could not physically fit inside the MRI machine). Two hundred eleven patients had a single lesion and 22 patients had concurrent noncontiguous epidural abscess lesions in different anatomical regions of the spine (ie, skip lesions). Complete medical records were obtained on these 233 individuals and the following data were collected for each patient: demographics (age and gender), medical history (diabetes mellitus, immunosuppression, alcohol use, tobacco use, and IVDU), maximal temperature in the hospital before any medical or surgical treatment, laboratory test values at the time of presentation before any medical or surgical treatment (white blood cell count and erythrocyte sedimentation rate [ESR]), location of epidural abscess lesion(s), any delay in presentation of 7 days or more from the time of symptom onset to hospital presentation, and presence of other sites of infection distant to the spine and paraspinal areas (eg, septic knee, pneumonia, subcutaneous abscess, etc). The presence of a lower urinary tract infection (ie, cystitis) did not qualify as a separate site of infection, given the frequency of urinary catheter use and their related infections. Fever was defined as a maximal temperature of O100.5 F before any medical or surgical treatment. There is great variability in the SEA
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
literature for the temperature threshold used to represent a ‘‘fever,’’ if a specific threshold is even mentioned in the article at all. Our literature review demonstrated that temperature thresholds ranged from 100.0 F to 101.0 F [2,7,9,15], thus we chose 100.5 F. Heavy alcohol intake was defined as greater than one drink a day for women and two drinks a day for men [16]. Immunosuppression included patients with hematological malignancies, human immunodeficiency virus, and patients on chemotherapy or long-term corticosteroids. Patients with skip SEAs were first compared with those with a single SEA via univariate analysis. Multivariate logistic regression with backward stepwise selection was then used to identify independent predictors of skip SEA lesions. Finally, a prediction algorithm based on these independent predictors was constructed for identifying patients at high risk of having skip SEAs. Statistical methods Univariate analysis was used to compare patients with a single SEA with those with noncontiguous SEA lesions with respect to the following variables: age, temperature of O100.5 F, delayed presentation of 7 or more days from the time of symptom onset to hospital presentation, presence of a concomitant infection distant to the spine, diabetes, active IVDU, alcohol abuse, tobacco use, immunosuppression, serum white blood cell count, and ESR. Categorical variables were analyzed with the two-tailed Fisher exact test and continuous variables with the two-sample Student t test. Multivariate logistic regression was then applied to identify significant independent predictors of skip epidural abscess lesions by considering candidate variables with p values of !.05 in the univariate analysis [17]. A backward stepwise selection procedure was used to establish the final multivariate model, and the likelihood ratio test was used to assess the significance of the results [18]. The optimal cutoff value (ie, threshold) for the ESR predictor was chosen with the use of the Youden index, which identifies the point on a receiver operating characteristic curve that is farthest from the line of nondiscrimination [19]. Thus, the identified cutoff value for ESR was chosen to maximize both sensitivity and specificity, as described in detail by Pepe [20]. Next multivariate logistic regression was used to calculate the probability of a skip epidural abscess lesion (and the associated 95% confidence interval [CI]) for each of the eight possible combinations of positive and negative predictors [21]. A more simplified model for estimating the probability of a skip lesion, in which the logistic regression was based on the number of positive predictors rather than on each possible combination, was also created. Finally, a receiver operating characteristic curve was created for our simplified model and the area under the curve was calculated. Statistical analyses were performed with SPSS software (version 20.0; SPSS/IBM, Chicago, IL, USA). A two-tailed p value of !.05 was considered significant.
97
Context Little surgical literature is capable of informing patient care in the setting of epidural abscess. In particular, few studies address the issue of non-contiguous abscesses presenting in the same individual. The presence of a noncontiguous abscess complicates care and may elevate the risk of mortality. Ju et al sought to study this issue, and factors associated with the need for increased surveillance, using data collected from two academic centers over the course of two decades. Contribution Delay in presentation, concomitant extra-spinal infection and elevated ESR were identified as risk factors associated with the presence of noncontiguous abscess. The authors present a predictive model intended to inform the need for pan-spinal imaging in individuals at elevated risk for noncontiguous epidural abscesses. Implications This study adds to the literature in a remarkably underexplored area. That being said, the investigation is still limited by its retrospective design and the possibility that individuals with noncontiguous abscess escaped detection due to the study’s methodology. While highlighting useful clinical characteristics that may increase scrutiny for noncontiguous abscess on the part of treating physicians, independent testing of the authors’ predictive rules is necessary among a larger sample. These issues are more completely elucidated in the companion commentary to this article: ‘‘Predictive Modeling for Epidural Abscess: What we can, can’t and should do about it.’’ —The Editors
Institutional review board committee approval was obtained before initiating the study. There was no external funding source for this study, and the institutional funding did not influence the investigation.
Results Our study cohort included 233 adults who all received imaging of the entire spinal column and were diagnosed with SEAs (Table 1). Two hundred eleven patients had a single SEA (mean and median age was 60 years) and 22 patients (9%) had concurrent noncontiguous epidural abscess lesions involving different anatomical regions of the spine (mean age of 63 years, median age of 62 years). There was no statistically significant correlation between age and the presence of skip lesions. For patients with skip lesions,
98
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
Table 1 Univariate analysis for assessing associations with skip lesions Variables Delay in presentation of $7 d Concomitant area of infection outside the spine Intravenous drug use Heavy alcohol intake Smoker Diabetes Immunosuppression Fever Serum white blood cell count (103 cells/mL) Erythrocyte sedimentation rate (mm/h)
Table 2 Significant independent multivariate predictors of skip lesions: logistic regression
No skip lesion (N5211)
Skip lesion (N522)
p
84 (40%) 51 (24%)
21 (91%) 21 (91%)
!.001 !.001
40 (19%) 38 (18%) 54 (26%) 54 (26%) 35 (17%) 87 (41%) 15.169.3
3 (13%) 3 (13%) 6 (26%) 6 (26%) 4 (17%) 11 (48%) 15.767.8
.584 .774 1.000 1.000 1.000 .658 .757
78.6631.3
106.3626.7
!.001
Note: Plus-minus data are shown as mean6standard deviation.
the cervical and thoracic spine were involved in 2 patients (9%), cervical and lumbar spine in 15 patients (68%), thoracic and lumbar spine in 3 patients (14%), and all three regions were involved in 2 patients (9%). Of the total 65 patients with an SEA involving the cervical spine, 19 (29%) had a skip lesion in the thoracic and/or lumbar spine. In terms of the causative organism in the 22 cases of skip SEA lesions, 19 (86%) were because of Staphylococcus species (12/22 [55%] because of methicillin-sensitive S. aureus, 6/22 [27%] from methicillin-resistant S. aureus, and 1/22 [5%] from coagulase-negative Staphylococcus). The remaining three cases were caused by Group B Streptococcus (2/22 or 9%) and Mycobacterium avium complex (1/22 or 5%). Univariate analyses comparing patients with skip epidural abscess lesions with patients with a single lesion demonstrated statistically significant differences in three variables (Table 1): delay of 7 or more days from the time of symptom onset to hospital presentation, other concurrent sites of infection outside the spine and paraspinal region, and ESR. The optimal ESR cutoff value for maximizing sensitivity and specificity was calculated to be an ESR of O95 mm/h. Identifying this cutoff value converts a continuous variable into a binary variable and aids its use in a prediction algorithm. Ninety-one percent of patients with a skip lesion had a delay in presentation of 7 or more days compared with only 40% of patients with a single SEA lesion. A similar disparity existed for the presence of other sites of infection, since this was true for 91% of patients with skip lesions but only present in 24% of patients with a single SEA lesion. Finally, 83% of skip lesion patients had a presentation ESR of O95 mm/h, whereas this criteria was met by only 30% of single SEA lesion patients. Multivariate analysis confirmed that having another site of infection (p!.001), an ESR of O95 mm/h (p5.003), and delay in presentation (p5.007) were all significant independent multivariate predictors for the presence of a skip epidural abscess lesion (Table 2). An
Variables
Odds ratio
95% Confidence interval
p
Concurrent site of infection outside the spine ESRO95 mm/h Delay in presentation of $7 d
57.6
6.8–490
!.001
8.6 8.5
2.0–39.2 1.5–47.4
.003 .007
ESR, erythrocyte sedimentation rate.
algorithm was then constructed to determine the predicted probability of a skip lesion on the basis of the eight (23) possible combinations of the three binary multivariate predictors (Table 3). The predicted probability of a skip lesion ranged from 73% for patients who had all three predictors to 0% for those with none of the three predictors. The area under the receiver operating characteristic curve for this full algorithm encompassing all possible combinations was excellent at 0.952 (p!.001, 95% CI: 0.914–0.990). For the ease of clinical application, regression modeling was used to develop a more simplified algorithm that considered only the number of multivariate predictors an individual possessed rather than which one of the eight specific combinations the patient satisfied. According to this simplified algorithm, the predicted probability of a skip lesion was 73% for patients satisfying all three predictors, 13% for two, 2% for one, and 0% for zero predictors (Table 4). Receiver operating characteristic curve analysis was used to evaluate the predictive accuracy of this model, and it revealed a steep shoulder with an excellent area under the curve of 0.936 (p!.001, 95% CI: 0.880–0.992) (Figure).
Discussion and conclusion Spinal epidural abscess is a serious condition that can rapidly lead to sepsis, paralysis, and even death if not Table 3 Full algorithm for probability of skip lesions based on combinations of the three multivariate predictors Another site of infection
ESRO95 mm/h
Delay of $7 d
Probability of skip lesion (%)
95% CI
No No No Yes No Yes Yes Yes
No No Yes No Yes No Yes Yes
No Yes No No Yes Yes No Yes
0 1 1 4 5 24 25 73
0–1 0–5 0–5 1–19 1–27 7–53 8–60 50–88
ESR, erythrocyte sedimentation rate; CI, confidence interval; AUC, area under the curve. Note: AUC for all three predictors combined: AUC50.952, 95% CI: 0.914–0.990, p!.001.
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101 Table 4 Simplified algorithm for the risk of skip lesions based on the number of multivariate predictors Number of predictors
Probability of skip lesion (%)
95% CI
0 1 2 3
0 2 13 73
0–1 1–9 5–31 50–88
CI, confidence interval; AUC, area under the curve. Note: AUC for simplified algorithm: AUC50.936, 95% CI: 0.880– 0.992, p!.001.
promptly diagnosed and appropriately treated. However, diagnosing an SEA can be difficult because of its nonspecific presentation. In fact, it is misdiagnosed up to 75% of the time [7,22], often being mistaken for degenerative disc disease, osteomyelitis, meningitis, or endocarditis [3]. For patients who present with neck or back pain and are subsequently diagnosed with an SEA based on an MRI of the involved spinal region (cervical, thoracic, or lumbar), clinicians must decide whether to image the remaining spinal regions. This question is especially germane for patients transferred to a tertiary care center after the referring facility diagnoses an SEA based on a single region MRI. In an attempt to balance the health care expenditures of additional imaging with the downsides of missing a skip SEA, information that risk-stratifies the patients would be useful. Accordingly, we have constructed a simple set of predictors to identify patients at high risk for skip lesions and, thus, warrant comprehensive spinal imaging. Data regarding skip epidural abscesses are sparse. Previous reports have been mostly in association with nonpyogenic lesions, such as tuberculosis [23,24]. In our study, one
Figure. Receiver operating characteristic curve for the simplified skip lesion prediction algorithm. The AUC is 0.936, indicating excellent diagnostic performance using a set of three clinical predictors (another site of infection, erythrocyte sedimentation rate, and delay in presentation of $7 days). The 45 dashed line represents the line of nondiscrimination. AUC, area under the curve; CI, confidence interval.
99
patient with a skip SEA was infected with a related organism, M. avium complex. With regard to pyogenic spine infections,Shousha and Boehm [25] published a retrospective study on 30 patients with cervical spondylodiscitis, where 14 patients (47%) had concomitant noncontiguous discitis in the thoracic and/or lumbar spine. However, they did not specify how many of these 14 patients actually had an epidural abscess. In our study, 29% of patients with a cervical SEA had a simultaneous skip SEA lesion in the thoracic and/or lumbar spine. There are only a few reports in the medical literature describing this phenomenon of noniatrogenic skip epidural abscess lesions located in different anatomical regions of the spine. Post et al. [13] mention in their study on the diagnostic utility of spinal MRI and intraoperative spinal ultrasound for spinal infections that 3 of their 24 patients had infections at two noncontiguous locations and that the second location was not suspected on the basis of clinical findings. However, they do not provide any further details on the radiographic findings and thus, it is impossible to know if the noncontiguous locations were in different regions of the spine or within the same anatomical region. Mampalam et al. [14] published a case series on nonoperative management of SEAs in six neurologically intact patients. In their report, one patient was found to have concurrent noncontiguous SEAs in the cervical and lumbar spine. She was a 25year-old woman who presented with 2 weeks of spinal pain and was also found to have a pulmonary empyema. Although her ESR was not specifically reported, the authors state that all study patients had elevated ESRs except for one that did not have a recorded ESR value. A case report by Pfister et al. [11] included a patient with concurrent SEAs in the cervical and lumbar spine, but only the cervical lesion was identified initially. This patient presented with 4 weeks of progressive lumbar pain. On admission, her ESR was 116 mm/h and she was septic from S. aureus. When she developed tetraparesis, a cervical spine MRI demonstrated a C5–C6 spondylodiscitis with an associated SEA. She underwent a cervical decompressive surgery and was put on broad-spectrum antibiotics. She initially improved but developed complete paresis of the lower extremities on postoperative Day 5. This prompted an entire spine MRI that revealed a second SEA at L3–L5. Thus, the patient returned to the operating room for L3–L5 laminectomies. Three months after admission, she still had moderate spastic tetraparesis. In their conclusion, Pfister et al. [11] state that SEAs can occur at noncontiguous sites and thus, ‘‘MRI of the entire spine may be necessary in select cases.’’ The present study is, to our knowledge, the largest study cohort on concomitant noncontiguous SEAs. More importantly, our study uniquely defines three predictors for the presence of a skip SEA lesion. Being mindful of the consequences of missing an SEA lesion, the following management scheme may be appropriate. Patients who present with symptoms or neurologic findings attributable to
100
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
different regions of the spine should undergo an entire spine gadolinium-enhanced MRI (or CT-myelogram). For those patients with symptoms/findings attributable to one spinal region, our algorithm can help with the risk stratification. Low-risk patients (ie, those with zero or one predictor) likely do not require entire spinal column imaging at the time of presentation and, thus, can undergo imaging of the one symptomatic spinal region. On the other hand, those who possess all three predictors are at high risk of having a skip lesion and, thus, warrant imaging of the entire spine. Finally, considering the ramifications of missing an SEA lesion, patients who possess two predictors should also undergo entire spinal column imaging. Regardless of how much of the spine is imaged, it is paramount to closely monitor SEA patients with serial neurologic exams as they can deteriorate quickly. By applying our predictive algorithm to the patients in the two case reports would have demonstrated that each patient was at an increased risk of having a noncontiguous SEA. The patient in the case series by Mampalam et al. [14] had a delay in presentation of $7 days from the time of symptom onset and also another site of infection (pulmonary empyema). Unfortunately, her ESR was not provided. The patient in the case report by Pfister et al. [11] also had a $7 days delay in the presentation and had an ESR of O95 mm/h. Unfortunately, from the case report, we do not know if she had another site of infection, although this would not be unexpected given her S. aureus sepsis. Nonetheless both patients possessed at least two of our predictors, which corresponds to a 13% risk of having a skip SEA lesion and warrants an entire spinal imaging. However, this risk could easily have been higher if we had a more complete clinical picture that included the ESR for the first patient and whether another site of infection (eg, skin abscess, septic joint, etc) was present for the second patient. One of the limitations of our study is its retrospective case-control design, which is theoretically subject to bias. We attempted to minimize the information bias associated with analysis of incomplete data by excluding patients with incomplete medical records. Furthermore, we attempted to reduce selection bias associated with including patients with presumptive and inconsistent diagnoses by excluding patients who did not undergo imaging of the entire spine on presentation. This is an important consideration because without entire spinal imaging, one cannot be certain that a patient truly only has one SEA lesion. Given the retrospective nature of this study and the fact that subjects were recruited from multiple hospitals within a large healthcare system, it was difficult to establish the precise reason why certain patients with SEA received full spinal imaging, whereas others did not. A review of the medical records suggests that variability in the provider preference was the predominant reason. However, this inconsistency further highlights the benefits of our proposed spinal imaging algorithm for clinicians treating patients with SEAs. The number of patients with skip SEA lesions was not very high
in our study; however, this is a rare event and our report represents the largest study published to date on this phenomenon. Also, we were unable to incorporate C-reactive protein data into our analysis because C-reactive protein levels were not reliably drawn on patients before the initiation of medical or surgical treatment. Finally, although we included the most pertinent risk factors that are examined in the context of SEAs, we did not use a specific comorbidity score (eg, Charlson comorbidity index) in our analysis. Given the retrospective nature of our study, it would be difficult to determine whether individual patients possessed each of the specific comorbidities accounted for in a scoring system, which can lead to inaccurate or incomplete scores. Despite these limitations, the clear differences between our case and control groups, combined with the highly significant results from our statistical analyses, make it very unlikely that our findings are spurious. Nonetheless, as with any new prediction algorithm, follow-up prospective studies with larger sample sizes should be conducted to evaluate the accuracy and utility of the algorithm before it is put into clinical practice. Although relatively rare, SEA should be included in the differential diagnosis of patients presenting with acute back pain, even in the absence of fever or neurologic deficits. It bears keeping in mind that a subgroup of patients may have a concurrent, yet noncontiguous, epidural abscess in another distant region of the spine. This second lesion, if initially unrecognized, can prompt an additional trip to the operating room, be a source of continued sepsis, and even lead to worsening paralysis or death. Thus, in the era of evidence-based medicine and rising health-care costs, clinical prediction algorithms are a way for physicians to make more costeffective and informed decisions. In the absence of symptoms or neurologic findings attributable to different regions of the spine, our proposed set of clinical predictors provides excellent diagnostic performance in riskstratifying the patients and identifying those who should receive entire spine MRIs. References [1] Baker AS, Ojemann RG, Swartz MN, Richardson EP Jr. Spinal epidural abscess. N Engl J Med 1975;293:463–8. [2] Rigamonti D, Liem L, Sampath P, et al. Spinal epidural abscess: contemporary trends in etiology, evaluation, and management. Surg Neurol 1999;52:189–96; discussion 197. [3] Darouiche RO. Spinal epidural abscess. N Engl J Med 2006;355: 2012–20. [4] Darouiche RO, Hamill RJ, Greenberg SB, et al. Bacterial spinal epidural abscess. Review of 43 cases and literature survey. Medicine 1992;71:369–85. [5] Khanna RK, Malik GM, Rock JP, Rosenblum ML. Spinal epidural abscess: evaluation of factors influencing outcome. Neurosurgery 1996;39:958–64. [6] Danner RL, Hartman BJ. Update on spinal epidural abscess: 35 cases and review of the literature. Rev Infect Dis 1987;9:265–74. [7] Davis DP, Wold RM, Patel RJ, et al. The clinical presentation and impact of diagnostic delays on emergency department patients with spinal epidural abscess. J Emerg Med 2004;26:285–91.
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101 [8] Mackenzie AR, Laing RB, Smith CC, et al. Spinal epidural abscess: the importance of early diagnosis and treatment. J Neurol Neurosurg Psychiatry 1998;65:209–12. [9] Maslen DR, Jones SR, Crislip MA, et al. Spinal epidural abscess. Optimizing patient care. Arch Intern Med 1993;153:1713–21. [10] Reihsaus E, Waldbaur H, Seeling W. Spinal epidural abscess: a meta-analysis of 915 patients. Neurosurg Rev 2000;23:175–204; discussion 205. [11] Pfister HW, von Rosen F, Yousry T. MRI detection of epidural spinal abscesses at noncontiguous sites. J Neurol 1996;243:315–7. [12] Mellado JM, Perez del Palomar L, Camins A, et al. MR imaging of spinal infection: atypical features, interpretive pitfalls and potential mimickers. Eur Radiol 2004;14:1980–9. [13] Post MJ, Quencer RM, Montalvo BM, et al. Spinal infection: evaluation with MR imaging and intraoperative US. Radiology 1988;169: 765–71. [14] Mampalam TJ, Rosegay H, Andrews BT, et al. Nonoperative treatment of spinal epidural infections. J Neurosurg 1989;71:208–10. [15] Del Curling O Jr, Gower DJ, McWhorter JM. Changing concepts in spinal epidural abscess: a report of 29 cases. Neurosurgery 1990;27: 185–92. [16] U.S. Department of Agriculture and U.S. Department of Health and Human Services. Dietary Guidelines for Americans, 2010. 7th ed.
[17] [18] [19] [20] [21]
[22]
[23]
[24] [25]
101
Washington, DC: U.S. Government Printing Office, December 2010. Katz MH. Multivariable analysis: a practical guide for clinicians. 2nd ed. New York, NY: Cambridge University Press, 2006. Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic and Cox regression. Am J Epidemiol 2007;165:710–8. Perkins NJ, Schisterman EF. The Youden Index and the optimal cutpoint corrected for measurement error. Biom J 2005;47:428–41. Pepe MS. The statistical evaluation of medical tests for classification and prediction. New York, NY: Oxford University Press, 2004. Harrell FE. Regression modeling strategies with applications to linear models, logistic regression, and survival analysis. New York, NY: Springer, 2001. Tang HJ, Lin HJ, Liu YC, Li CM. Spinal epidural abscess–experience with 46 patients and evaluation of prognostic factors. J Infect 2002;45:76–81. Nussbaum ES, Rockswold GL, Bergman TA, et al. Spinal tuberculosis: a diagnostic and management challenge. J Neurosurg 1995;83: 243–7. Shanley DJ. Tuberculosis of the spine: imaging features. AJR Am J Roentgenol 1995;164:659–64. Shousha M, Boehm H. Surgical treatment of cervical spondylodiscitis: a review of 30 consecutive patients. Spine 2012;37:E30–6.
Clinical Study
Predicting patients with concurrent noncontiguous spinal epidural abscess lesions Kevin L. Ju, MD*, Sang Do Kim, MD, MS, Rojeh Melikian, MD, Christopher M. Bono, MD, Mitchel B. Harris, MD Department of Orthopaedic Surgery, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115, USA Received 10 January 2014; revised 30 April 2014; accepted 8 June 2014
Abstract
BACKGROUND CONTEXT: Spinal epidural abscess (SEA) is a serious condition that can lead to significant morbidity and mortality if not expeditiously diagnosed and appropriately treated. However, the nonspecific findings that accompany SEAs often make its diagnosis difficult. Concurrent noncontiguous SEAs are even more challenging to diagnose because whole-spine imaging is not routinely performed unless the patient demonstrates neurologic findings that are inconsistent with the identified lesion. Failure to recognize a separate SEA can subject patients to a second operation, continued sepsis, paralysis, or even death. PURPOSE: To formulate a set of clinical and laboratory predictors for identifying patients with concurrent noncontiguous SEAs. STUDY DESIGN: A retrospective, case-control study. PATIENT SAMPLE: Patients aged 18 years or older admitted to our institution during the study period who underwent entire spinal imaging and were diagnosed with one or more SEAs. OUTCOME MEASURES: The presence or absence of concurrent noncontiguous SEAs on magnetic resonance imaging or computed tomography (CT)-myelogram. METHODS: A retrospective review was performed on 233 adults with SEAs who presented to our health-care system from 1993 to 2011 and underwent entire spinal imaging. The clinical and radiographic features of patients with concurrent noncontiguous SEAs, defined as at least two lesions in different anatomical regions of the spine (ie, cervical, thoracic, or lumbar), were compared with those with a single SEA. Multivariate logistic regression identified independent predictors for the presence of a skip SEA, and a prediction algorithm based on these independent predictors was constructed. Institutional review board committee approval was obtained before initiating the study. RESULTS: Univariate and multivariate analyses comparing patients with skip SEA lesions (n522) with those with single lesions (n5211) demonstrated significant differences in three factors: delay in presentation (defined as symptoms for $7 days), a concomitant area of infection outside the spine and paraspinal region, and an erythrocyte sedimentation rate of O95 mm/h at presentation. The predicted probability for the presence of a skip lesion was 73% for patients possessing all three predictors, 13% for two, 2% for one, and 0% for zero predictors. Receiver operating characteristic curve analysis, used to evaluate the predictive accuracy of the model, revealed a steep shoulder with an area under the curve of 0.936 (p!.001).
FDA device/drug status: Not applicable. Author disclosures: KLJ: Nothing to disclose. SDK: Nothing to disclose. RM: Nothing to disclose. CMB: Royalties: Wolters Kluwer (A); Consulting: United Health Care (B), Intrinsic Therapeutics, DSMB (B). MBH: Nothing to disclose. The disclosure key can be found on the Table of Contents and at www. TheSpineJournalOnline.com. http://dx.doi.org/10.1016/j.spinee.2014.06.008 1529-9430/Ó 2015 Elsevier Inc. All rights reserved.
There was no external funding source for this study, and the institutional funding did not influence the investigation. * Corresponding author. Department of Orthopaedic Surgery, Brigham and Women’s Hospital, 75 Francis St, Boston, MA 02115, USA. Tel.: (617) 600-8498. E-mail address: [email protected] (K.L. Ju)
96
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
CONCLUSIONS: The proposed set of three predictors may be a useful tool in predicting the risk of a skip SEA lesion and, consequently, which patients would benefit from entire spinal imaging. Ó 2015 Elsevier Inc. All rights reserved. Keywords:
Epidural abscess; Skip lesion; Concurrent; Noncontiguous; Prediction; Algorithm
Introduction Spinal epidural abscess (SEA) is a serious disease that can rapidly lead to neurologic deficit, sepsis, or even death if not promptly diagnosed and appropriately treated. Although SEA is relatively uncommon, with previous estimated incidences ranging from 0.2 to 1.2 cases per 10,000 hospital admissions [1], more recent data indicate the incidence has increased to 12.5 cases per 10,000 admissions [2]. This rise is likely a reflection of an aging population, increasing prevalence of predisposing comorbidities such as diabetes mellitus and immunosuppression, and more widespread intravenous drug use (IVDU) and alcoholism [3,4]. Despite advances in imaging techniques, antibiotic therapy, and surgical techniques, the overall mortality of SEA is still about 5% [3], and up to 22% of patients suffer irreversible paralysis [3,5]. Timely diagnosis and initiation of appropriate treatment is key to reducing the morbidity and mortality associated with this potentially devastating disease [6–10]. Identifying patients with a single SEA lesion can be challenging as the most common presenting complaint is nonspecific back or neck pain [3,7]. In a metaanalysis of 915 patients with SEAs, about three-quarters had back or neck pain, half had a fever, and only a quarter had neurologic deficits [10]. The classic clinical triad of back pain, fever, and neurologic deficit is only present in a minority of patients with SEA and exhibits a diagnostic sensitivity of only 8% [7]. Simultaneous noncontiguous SEAs in different anatomical regions of the spine (ie, cervical, thoracic, or lumbar), also known as skip lesions, pose an even greater diagnostic challenge. Unless a patient exhibits signs and symptoms that are not concordant with a detected, single SEA, practitioners do not reflexively order additional imaging of other spinal regions [11,12]. Failing to detect a skip SEA can lead to continued sepsis or neurologic decline. Further complicating the diagnostic dilemma is the fact that additional SEAs may initially be asymptomatic or present with pain only. Pfister et al. [11] described a patient with skip SEAs (cervical and lumbar) in whom the lumbar SEA was missed on the initial work-up. It was not until the patient had undergone a cervical decompression and continued to deteriorate neurologically that a whole spine magnetic resonance imaging (MRI) was performed and the lumbar SEA lesion was identified. With the exception of case reports and small case series, there is a dearth of clinical data concerning skip SEAs [11,13,14]. The purpose of our study was to formulate a set of clinical and laboratory predictors to help
identify patients with skip SEAs in different regions of the spine based on a large retrospectively-collected database. Materials and methods Study design Cases of SEA treated within a large health-care system between 1993 and 2011 were retrospectively reviewed. Inclusion criteria were patients aged 18 years or older with an entire spinal imaging using MRI with IV contrast or computed tomography (CT) with myelogram and IV contrast and found to have a spinal epidural abscess with a fluid collection. Patients with spondylodiscitis or osteomyelitis with granulation tissue (ie, phlegmon) but without a clear epidural fluid collection were excluded. Secondary SEAs occurring in patients who previously had a lumbar puncture, epidural injection, laminectomy, or laminotomy at the level or adjacent levels to the SEA were excluded. Patients with insufficient information were also excluded. In total, 233 adults were included in the study. Two hundred thirty-one patients underwent entire spine MRIs with IV contrast, whereas two patients underwent CTmyelograms with IV contrast of the entire spine (one had a pacemaker and one could not physically fit inside the MRI machine). Two hundred eleven patients had a single lesion and 22 patients had concurrent noncontiguous epidural abscess lesions in different anatomical regions of the spine (ie, skip lesions). Complete medical records were obtained on these 233 individuals and the following data were collected for each patient: demographics (age and gender), medical history (diabetes mellitus, immunosuppression, alcohol use, tobacco use, and IVDU), maximal temperature in the hospital before any medical or surgical treatment, laboratory test values at the time of presentation before any medical or surgical treatment (white blood cell count and erythrocyte sedimentation rate [ESR]), location of epidural abscess lesion(s), any delay in presentation of 7 days or more from the time of symptom onset to hospital presentation, and presence of other sites of infection distant to the spine and paraspinal areas (eg, septic knee, pneumonia, subcutaneous abscess, etc). The presence of a lower urinary tract infection (ie, cystitis) did not qualify as a separate site of infection, given the frequency of urinary catheter use and their related infections. Fever was defined as a maximal temperature of O100.5 F before any medical or surgical treatment. There is great variability in the SEA
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
literature for the temperature threshold used to represent a ‘‘fever,’’ if a specific threshold is even mentioned in the article at all. Our literature review demonstrated that temperature thresholds ranged from 100.0 F to 101.0 F [2,7,9,15], thus we chose 100.5 F. Heavy alcohol intake was defined as greater than one drink a day for women and two drinks a day for men [16]. Immunosuppression included patients with hematological malignancies, human immunodeficiency virus, and patients on chemotherapy or long-term corticosteroids. Patients with skip SEAs were first compared with those with a single SEA via univariate analysis. Multivariate logistic regression with backward stepwise selection was then used to identify independent predictors of skip SEA lesions. Finally, a prediction algorithm based on these independent predictors was constructed for identifying patients at high risk of having skip SEAs. Statistical methods Univariate analysis was used to compare patients with a single SEA with those with noncontiguous SEA lesions with respect to the following variables: age, temperature of O100.5 F, delayed presentation of 7 or more days from the time of symptom onset to hospital presentation, presence of a concomitant infection distant to the spine, diabetes, active IVDU, alcohol abuse, tobacco use, immunosuppression, serum white blood cell count, and ESR. Categorical variables were analyzed with the two-tailed Fisher exact test and continuous variables with the two-sample Student t test. Multivariate logistic regression was then applied to identify significant independent predictors of skip epidural abscess lesions by considering candidate variables with p values of !.05 in the univariate analysis [17]. A backward stepwise selection procedure was used to establish the final multivariate model, and the likelihood ratio test was used to assess the significance of the results [18]. The optimal cutoff value (ie, threshold) for the ESR predictor was chosen with the use of the Youden index, which identifies the point on a receiver operating characteristic curve that is farthest from the line of nondiscrimination [19]. Thus, the identified cutoff value for ESR was chosen to maximize both sensitivity and specificity, as described in detail by Pepe [20]. Next multivariate logistic regression was used to calculate the probability of a skip epidural abscess lesion (and the associated 95% confidence interval [CI]) for each of the eight possible combinations of positive and negative predictors [21]. A more simplified model for estimating the probability of a skip lesion, in which the logistic regression was based on the number of positive predictors rather than on each possible combination, was also created. Finally, a receiver operating characteristic curve was created for our simplified model and the area under the curve was calculated. Statistical analyses were performed with SPSS software (version 20.0; SPSS/IBM, Chicago, IL, USA). A two-tailed p value of !.05 was considered significant.
97
Context Little surgical literature is capable of informing patient care in the setting of epidural abscess. In particular, few studies address the issue of non-contiguous abscesses presenting in the same individual. The presence of a noncontiguous abscess complicates care and may elevate the risk of mortality. Ju et al sought to study this issue, and factors associated with the need for increased surveillance, using data collected from two academic centers over the course of two decades. Contribution Delay in presentation, concomitant extra-spinal infection and elevated ESR were identified as risk factors associated with the presence of noncontiguous abscess. The authors present a predictive model intended to inform the need for pan-spinal imaging in individuals at elevated risk for noncontiguous epidural abscesses. Implications This study adds to the literature in a remarkably underexplored area. That being said, the investigation is still limited by its retrospective design and the possibility that individuals with noncontiguous abscess escaped detection due to the study’s methodology. While highlighting useful clinical characteristics that may increase scrutiny for noncontiguous abscess on the part of treating physicians, independent testing of the authors’ predictive rules is necessary among a larger sample. These issues are more completely elucidated in the companion commentary to this article: ‘‘Predictive Modeling for Epidural Abscess: What we can, can’t and should do about it.’’ —The Editors
Institutional review board committee approval was obtained before initiating the study. There was no external funding source for this study, and the institutional funding did not influence the investigation.
Results Our study cohort included 233 adults who all received imaging of the entire spinal column and were diagnosed with SEAs (Table 1). Two hundred eleven patients had a single SEA (mean and median age was 60 years) and 22 patients (9%) had concurrent noncontiguous epidural abscess lesions involving different anatomical regions of the spine (mean age of 63 years, median age of 62 years). There was no statistically significant correlation between age and the presence of skip lesions. For patients with skip lesions,
98
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
Table 1 Univariate analysis for assessing associations with skip lesions Variables Delay in presentation of $7 d Concomitant area of infection outside the spine Intravenous drug use Heavy alcohol intake Smoker Diabetes Immunosuppression Fever Serum white blood cell count (103 cells/mL) Erythrocyte sedimentation rate (mm/h)
Table 2 Significant independent multivariate predictors of skip lesions: logistic regression
No skip lesion (N5211)
Skip lesion (N522)
p
84 (40%) 51 (24%)
21 (91%) 21 (91%)
!.001 !.001
40 (19%) 38 (18%) 54 (26%) 54 (26%) 35 (17%) 87 (41%) 15.169.3
3 (13%) 3 (13%) 6 (26%) 6 (26%) 4 (17%) 11 (48%) 15.767.8
.584 .774 1.000 1.000 1.000 .658 .757
78.6631.3
106.3626.7
!.001
Note: Plus-minus data are shown as mean6standard deviation.
the cervical and thoracic spine were involved in 2 patients (9%), cervical and lumbar spine in 15 patients (68%), thoracic and lumbar spine in 3 patients (14%), and all three regions were involved in 2 patients (9%). Of the total 65 patients with an SEA involving the cervical spine, 19 (29%) had a skip lesion in the thoracic and/or lumbar spine. In terms of the causative organism in the 22 cases of skip SEA lesions, 19 (86%) were because of Staphylococcus species (12/22 [55%] because of methicillin-sensitive S. aureus, 6/22 [27%] from methicillin-resistant S. aureus, and 1/22 [5%] from coagulase-negative Staphylococcus). The remaining three cases were caused by Group B Streptococcus (2/22 or 9%) and Mycobacterium avium complex (1/22 or 5%). Univariate analyses comparing patients with skip epidural abscess lesions with patients with a single lesion demonstrated statistically significant differences in three variables (Table 1): delay of 7 or more days from the time of symptom onset to hospital presentation, other concurrent sites of infection outside the spine and paraspinal region, and ESR. The optimal ESR cutoff value for maximizing sensitivity and specificity was calculated to be an ESR of O95 mm/h. Identifying this cutoff value converts a continuous variable into a binary variable and aids its use in a prediction algorithm. Ninety-one percent of patients with a skip lesion had a delay in presentation of 7 or more days compared with only 40% of patients with a single SEA lesion. A similar disparity existed for the presence of other sites of infection, since this was true for 91% of patients with skip lesions but only present in 24% of patients with a single SEA lesion. Finally, 83% of skip lesion patients had a presentation ESR of O95 mm/h, whereas this criteria was met by only 30% of single SEA lesion patients. Multivariate analysis confirmed that having another site of infection (p!.001), an ESR of O95 mm/h (p5.003), and delay in presentation (p5.007) were all significant independent multivariate predictors for the presence of a skip epidural abscess lesion (Table 2). An
Variables
Odds ratio
95% Confidence interval
p
Concurrent site of infection outside the spine ESRO95 mm/h Delay in presentation of $7 d
57.6
6.8–490
!.001
8.6 8.5
2.0–39.2 1.5–47.4
.003 .007
ESR, erythrocyte sedimentation rate.
algorithm was then constructed to determine the predicted probability of a skip lesion on the basis of the eight (23) possible combinations of the three binary multivariate predictors (Table 3). The predicted probability of a skip lesion ranged from 73% for patients who had all three predictors to 0% for those with none of the three predictors. The area under the receiver operating characteristic curve for this full algorithm encompassing all possible combinations was excellent at 0.952 (p!.001, 95% CI: 0.914–0.990). For the ease of clinical application, regression modeling was used to develop a more simplified algorithm that considered only the number of multivariate predictors an individual possessed rather than which one of the eight specific combinations the patient satisfied. According to this simplified algorithm, the predicted probability of a skip lesion was 73% for patients satisfying all three predictors, 13% for two, 2% for one, and 0% for zero predictors (Table 4). Receiver operating characteristic curve analysis was used to evaluate the predictive accuracy of this model, and it revealed a steep shoulder with an excellent area under the curve of 0.936 (p!.001, 95% CI: 0.880–0.992) (Figure).
Discussion and conclusion Spinal epidural abscess is a serious condition that can rapidly lead to sepsis, paralysis, and even death if not Table 3 Full algorithm for probability of skip lesions based on combinations of the three multivariate predictors Another site of infection
ESRO95 mm/h
Delay of $7 d
Probability of skip lesion (%)
95% CI
No No No Yes No Yes Yes Yes
No No Yes No Yes No Yes Yes
No Yes No No Yes Yes No Yes
0 1 1 4 5 24 25 73
0–1 0–5 0–5 1–19 1–27 7–53 8–60 50–88
ESR, erythrocyte sedimentation rate; CI, confidence interval; AUC, area under the curve. Note: AUC for all three predictors combined: AUC50.952, 95% CI: 0.914–0.990, p!.001.
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101 Table 4 Simplified algorithm for the risk of skip lesions based on the number of multivariate predictors Number of predictors
Probability of skip lesion (%)
95% CI
0 1 2 3
0 2 13 73
0–1 1–9 5–31 50–88
CI, confidence interval; AUC, area under the curve. Note: AUC for simplified algorithm: AUC50.936, 95% CI: 0.880– 0.992, p!.001.
promptly diagnosed and appropriately treated. However, diagnosing an SEA can be difficult because of its nonspecific presentation. In fact, it is misdiagnosed up to 75% of the time [7,22], often being mistaken for degenerative disc disease, osteomyelitis, meningitis, or endocarditis [3]. For patients who present with neck or back pain and are subsequently diagnosed with an SEA based on an MRI of the involved spinal region (cervical, thoracic, or lumbar), clinicians must decide whether to image the remaining spinal regions. This question is especially germane for patients transferred to a tertiary care center after the referring facility diagnoses an SEA based on a single region MRI. In an attempt to balance the health care expenditures of additional imaging with the downsides of missing a skip SEA, information that risk-stratifies the patients would be useful. Accordingly, we have constructed a simple set of predictors to identify patients at high risk for skip lesions and, thus, warrant comprehensive spinal imaging. Data regarding skip epidural abscesses are sparse. Previous reports have been mostly in association with nonpyogenic lesions, such as tuberculosis [23,24]. In our study, one
Figure. Receiver operating characteristic curve for the simplified skip lesion prediction algorithm. The AUC is 0.936, indicating excellent diagnostic performance using a set of three clinical predictors (another site of infection, erythrocyte sedimentation rate, and delay in presentation of $7 days). The 45 dashed line represents the line of nondiscrimination. AUC, area under the curve; CI, confidence interval.
99
patient with a skip SEA was infected with a related organism, M. avium complex. With regard to pyogenic spine infections,Shousha and Boehm [25] published a retrospective study on 30 patients with cervical spondylodiscitis, where 14 patients (47%) had concomitant noncontiguous discitis in the thoracic and/or lumbar spine. However, they did not specify how many of these 14 patients actually had an epidural abscess. In our study, 29% of patients with a cervical SEA had a simultaneous skip SEA lesion in the thoracic and/or lumbar spine. There are only a few reports in the medical literature describing this phenomenon of noniatrogenic skip epidural abscess lesions located in different anatomical regions of the spine. Post et al. [13] mention in their study on the diagnostic utility of spinal MRI and intraoperative spinal ultrasound for spinal infections that 3 of their 24 patients had infections at two noncontiguous locations and that the second location was not suspected on the basis of clinical findings. However, they do not provide any further details on the radiographic findings and thus, it is impossible to know if the noncontiguous locations were in different regions of the spine or within the same anatomical region. Mampalam et al. [14] published a case series on nonoperative management of SEAs in six neurologically intact patients. In their report, one patient was found to have concurrent noncontiguous SEAs in the cervical and lumbar spine. She was a 25year-old woman who presented with 2 weeks of spinal pain and was also found to have a pulmonary empyema. Although her ESR was not specifically reported, the authors state that all study patients had elevated ESRs except for one that did not have a recorded ESR value. A case report by Pfister et al. [11] included a patient with concurrent SEAs in the cervical and lumbar spine, but only the cervical lesion was identified initially. This patient presented with 4 weeks of progressive lumbar pain. On admission, her ESR was 116 mm/h and she was septic from S. aureus. When she developed tetraparesis, a cervical spine MRI demonstrated a C5–C6 spondylodiscitis with an associated SEA. She underwent a cervical decompressive surgery and was put on broad-spectrum antibiotics. She initially improved but developed complete paresis of the lower extremities on postoperative Day 5. This prompted an entire spine MRI that revealed a second SEA at L3–L5. Thus, the patient returned to the operating room for L3–L5 laminectomies. Three months after admission, she still had moderate spastic tetraparesis. In their conclusion, Pfister et al. [11] state that SEAs can occur at noncontiguous sites and thus, ‘‘MRI of the entire spine may be necessary in select cases.’’ The present study is, to our knowledge, the largest study cohort on concomitant noncontiguous SEAs. More importantly, our study uniquely defines three predictors for the presence of a skip SEA lesion. Being mindful of the consequences of missing an SEA lesion, the following management scheme may be appropriate. Patients who present with symptoms or neurologic findings attributable to
100
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101
different regions of the spine should undergo an entire spine gadolinium-enhanced MRI (or CT-myelogram). For those patients with symptoms/findings attributable to one spinal region, our algorithm can help with the risk stratification. Low-risk patients (ie, those with zero or one predictor) likely do not require entire spinal column imaging at the time of presentation and, thus, can undergo imaging of the one symptomatic spinal region. On the other hand, those who possess all three predictors are at high risk of having a skip lesion and, thus, warrant imaging of the entire spine. Finally, considering the ramifications of missing an SEA lesion, patients who possess two predictors should also undergo entire spinal column imaging. Regardless of how much of the spine is imaged, it is paramount to closely monitor SEA patients with serial neurologic exams as they can deteriorate quickly. By applying our predictive algorithm to the patients in the two case reports would have demonstrated that each patient was at an increased risk of having a noncontiguous SEA. The patient in the case series by Mampalam et al. [14] had a delay in presentation of $7 days from the time of symptom onset and also another site of infection (pulmonary empyema). Unfortunately, her ESR was not provided. The patient in the case report by Pfister et al. [11] also had a $7 days delay in the presentation and had an ESR of O95 mm/h. Unfortunately, from the case report, we do not know if she had another site of infection, although this would not be unexpected given her S. aureus sepsis. Nonetheless both patients possessed at least two of our predictors, which corresponds to a 13% risk of having a skip SEA lesion and warrants an entire spinal imaging. However, this risk could easily have been higher if we had a more complete clinical picture that included the ESR for the first patient and whether another site of infection (eg, skin abscess, septic joint, etc) was present for the second patient. One of the limitations of our study is its retrospective case-control design, which is theoretically subject to bias. We attempted to minimize the information bias associated with analysis of incomplete data by excluding patients with incomplete medical records. Furthermore, we attempted to reduce selection bias associated with including patients with presumptive and inconsistent diagnoses by excluding patients who did not undergo imaging of the entire spine on presentation. This is an important consideration because without entire spinal imaging, one cannot be certain that a patient truly only has one SEA lesion. Given the retrospective nature of this study and the fact that subjects were recruited from multiple hospitals within a large healthcare system, it was difficult to establish the precise reason why certain patients with SEA received full spinal imaging, whereas others did not. A review of the medical records suggests that variability in the provider preference was the predominant reason. However, this inconsistency further highlights the benefits of our proposed spinal imaging algorithm for clinicians treating patients with SEAs. The number of patients with skip SEA lesions was not very high
in our study; however, this is a rare event and our report represents the largest study published to date on this phenomenon. Also, we were unable to incorporate C-reactive protein data into our analysis because C-reactive protein levels were not reliably drawn on patients before the initiation of medical or surgical treatment. Finally, although we included the most pertinent risk factors that are examined in the context of SEAs, we did not use a specific comorbidity score (eg, Charlson comorbidity index) in our analysis. Given the retrospective nature of our study, it would be difficult to determine whether individual patients possessed each of the specific comorbidities accounted for in a scoring system, which can lead to inaccurate or incomplete scores. Despite these limitations, the clear differences between our case and control groups, combined with the highly significant results from our statistical analyses, make it very unlikely that our findings are spurious. Nonetheless, as with any new prediction algorithm, follow-up prospective studies with larger sample sizes should be conducted to evaluate the accuracy and utility of the algorithm before it is put into clinical practice. Although relatively rare, SEA should be included in the differential diagnosis of patients presenting with acute back pain, even in the absence of fever or neurologic deficits. It bears keeping in mind that a subgroup of patients may have a concurrent, yet noncontiguous, epidural abscess in another distant region of the spine. This second lesion, if initially unrecognized, can prompt an additional trip to the operating room, be a source of continued sepsis, and even lead to worsening paralysis or death. Thus, in the era of evidence-based medicine and rising health-care costs, clinical prediction algorithms are a way for physicians to make more costeffective and informed decisions. In the absence of symptoms or neurologic findings attributable to different regions of the spine, our proposed set of clinical predictors provides excellent diagnostic performance in riskstratifying the patients and identifying those who should receive entire spine MRIs. References [1] Baker AS, Ojemann RG, Swartz MN, Richardson EP Jr. Spinal epidural abscess. N Engl J Med 1975;293:463–8. [2] Rigamonti D, Liem L, Sampath P, et al. Spinal epidural abscess: contemporary trends in etiology, evaluation, and management. Surg Neurol 1999;52:189–96; discussion 197. [3] Darouiche RO. Spinal epidural abscess. N Engl J Med 2006;355: 2012–20. [4] Darouiche RO, Hamill RJ, Greenberg SB, et al. Bacterial spinal epidural abscess. Review of 43 cases and literature survey. Medicine 1992;71:369–85. [5] Khanna RK, Malik GM, Rock JP, Rosenblum ML. Spinal epidural abscess: evaluation of factors influencing outcome. Neurosurgery 1996;39:958–64. [6] Danner RL, Hartman BJ. Update on spinal epidural abscess: 35 cases and review of the literature. Rev Infect Dis 1987;9:265–74. [7] Davis DP, Wold RM, Patel RJ, et al. The clinical presentation and impact of diagnostic delays on emergency department patients with spinal epidural abscess. J Emerg Med 2004;26:285–91.
K.L. Ju et al. / The Spine Journal 15 (2015) 95–101 [8] Mackenzie AR, Laing RB, Smith CC, et al. Spinal epidural abscess: the importance of early diagnosis and treatment. J Neurol Neurosurg Psychiatry 1998;65:209–12. [9] Maslen DR, Jones SR, Crislip MA, et al. Spinal epidural abscess. Optimizing patient care. Arch Intern Med 1993;153:1713–21. [10] Reihsaus E, Waldbaur H, Seeling W. Spinal epidural abscess: a meta-analysis of 915 patients. Neurosurg Rev 2000;23:175–204; discussion 205. [11] Pfister HW, von Rosen F, Yousry T. MRI detection of epidural spinal abscesses at noncontiguous sites. J Neurol 1996;243:315–7. [12] Mellado JM, Perez del Palomar L, Camins A, et al. MR imaging of spinal infection: atypical features, interpretive pitfalls and potential mimickers. Eur Radiol 2004;14:1980–9. [13] Post MJ, Quencer RM, Montalvo BM, et al. Spinal infection: evaluation with MR imaging and intraoperative US. Radiology 1988;169: 765–71. [14] Mampalam TJ, Rosegay H, Andrews BT, et al. Nonoperative treatment of spinal epidural infections. J Neurosurg 1989;71:208–10. [15] Del Curling O Jr, Gower DJ, McWhorter JM. Changing concepts in spinal epidural abscess: a report of 29 cases. Neurosurgery 1990;27: 185–92. [16] U.S. Department of Agriculture and U.S. Department of Health and Human Services. Dietary Guidelines for Americans, 2010. 7th ed.
[17] [18] [19] [20] [21]
[22]
[23]
[24] [25]
101
Washington, DC: U.S. Government Printing Office, December 2010. Katz MH. Multivariable analysis: a practical guide for clinicians. 2nd ed. New York, NY: Cambridge University Press, 2006. Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic and Cox regression. Am J Epidemiol 2007;165:710–8. Perkins NJ, Schisterman EF. The Youden Index and the optimal cutpoint corrected for measurement error. Biom J 2005;47:428–41. Pepe MS. The statistical evaluation of medical tests for classification and prediction. New York, NY: Oxford University Press, 2004. Harrell FE. Regression modeling strategies with applications to linear models, logistic regression, and survival analysis. New York, NY: Springer, 2001. Tang HJ, Lin HJ, Liu YC, Li CM. Spinal epidural abscess–experience with 46 patients and evaluation of prognostic factors. J Infect 2002;45:76–81. Nussbaum ES, Rockswold GL, Bergman TA, et al. Spinal tuberculosis: a diagnostic and management challenge. J Neurosurg 1995;83: 243–7. Shanley DJ. Tuberculosis of the spine: imaging features. AJR Am J Roentgenol 1995;164:659–64. Shousha M, Boehm H. Surgical treatment of cervical spondylodiscitis: a review of 30 consecutive patients. Spine 2012;37:E30–6.
