Retrospective,Propensity Score-Matched Cohort Study Examining Timing of Fracture Fixation for Traumatic Thoracolumbar Fractures

Abstract

The timing of surgery in patients with traumatic thoracic/thoracolumbar fractures, with or without spinal cord injury, remains controversial. The objective of this study was to determine the importance of the timing of surgery for complications and resource utilization following fixation of traumatic thoracic/thoracolumbar fractures. In this retrospective cohort study, the 2003–2008 California Inpatient Databases were searched for patients receiving traumatic thoracic/thoracolumbar fracture fixation. Patients were classified as having early (<72 h) or late (>72 h) surgery. Propensity score modeling produced a matched cohort balanced on age, comorbidity, trauma severity, and other factors. Complications, mortality, length of stay, and hospital charges were assessed. Multivariate logistic regression was used to determine the impact of delayed surgery on in-hospital complications after balancing and controlling for other important factors. Early surgery (<72 h) for traumatic thoracic/thoracolumbar fractures was associated with a significantly lower overall complication rate (including cardiac, thromboembolic, and respiratory complications), and decreased hospital stay. In-hospital charges were significantly lower ($38,120 difference) in the early surgery group. Multivariate analysis identified time to surgery as the strongest predictor of in-hospital complications, although age, medical comorbidities, and injury severity score were also independently associated with increased complications. We reinforce the beneficial impact of early spinal surgery (prior to 72 h) in traumatic thoracic/thoracolumbar fractures to reduce in-hospital complications, hospital stay, and resource utilization. These results provide further support to the emerging literature and professional consensus regarding the importance of early thoracic/thoracolumbar spine stabilization of traumatic fractures to improve patient outcomes and limit hospitalization costs.

Introduction

T he timing of surgery in patients with traumatic thoracic/thoracolumbar fractures, with or without spinal cord injury, remains controversial. Early surgery (within 72 h) can be associated with increased operative blood loss and hypotension, which may compromise an at-risk spinal cord, and consequently increase neurologic complications. Theoretically, delaying surgery for 72 h avoids these complications and allows for surgical intervention under more controlled conditions. Early surgery, however, has the advantage of quicker patient mobilization with shorter hospital stays and lower rates of pulmonary and thromboembolic complications (Croce et al., 2001; Chipman et al., 2004). Current studies are inadequate to determine whether the timing of surgery affects surgical outcomes. A number of single institution and administrative database studies (Croce et al., 2001; Chipman et al., 2004; Kerwin et al., 2005,2007) concluded that early surgery was associated with lower in-hospital complication rates and shorter hospital length of stay. Generalization of these retrospective studies has been limited by the small numbers of patients, with the largest study consisting of less than 200 subjects. In addition, these studies included comparisons of non-matched or inadequately matched cohorts, and did not include comorbidities, age, or injury score. Recent systematic reviews concluded that early surgery was beneficial with respect to fewer complications and decreased length of stay (Bellabarba et al., 2010; Dimar et al., 2010; Thomas et al., 2006). However, these studies were based on and limited by the above-described study design flaws. Likewise, to date there are limited data on the impact of timing of surgery on in-hospital costs.

Administrative databases such as the California State Inpatient Database (CA SID; Healthcare Cost and Utilization Project, 2003–2004) provide a means of looking at in-hospital complications of a large number of patients segregated by timing of operation (i.e., those receiving an operation within 72 h of admission and those outside this 72-h window). In this study we used CA SID to identify a cohort of patients with traumatic thoracic/thoracolumbar fractures, with or without spinal cord injury, who underwent spinal fusion. Our goal was to compare in-hospital complication rates and length of stay by time to surgery, and assess the impact of early surgery on hospitalization costs. The cohort of patients who underwent early surgery (within 72 h) was compared with an age, comorbidity, and trauma injury score–matched cohort of patients who underwent surgery after 72 h. Secondary predictors of in-hospital complications were identified by multivariate analysis.

Methods

Data source and inclusion and exclusion criteria

The Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project obtains, processes, and distributes the CA SID, which contains discharge data on over 95% of hospitalizations at community and non-community (e.g., federal) hospitals in California. The reliability of the CA SID has previously been demonstrated (Romano and Mark, 1994). The 2003 through 2008 CA SID was searched for adult (age ≥18 years) patients with a primary International Classification of Diseases, Ninth Revision (ICD-9) diagnosis code of open or closed traumatic fracture of the lower thoracic or lumbar vertebrae, with or without known spinal cord injury (805.2–805.5, 806.25–806.29, 806.35–806.39, 806.4, and 806.5), and a primary ICD-9 procedure code of thoracic/thoracolumbar fusion surgery (anterior, 81.04; posterior, 81.05). Only cases with a known time to surgery occurring on or after the day of admission were included in this study.

Recorded data

Age, race, sex, insurance status (private insurance versus Medicare versus all others, including Medicaid and uninsured), admission source (emergency department versus all other), diagnosis and procedure codes (up to 25 each), time to surgery (early [i.e., within 72 hours] versus late [i.e., beyond 72 h]), discharge status (routine [i.e., to home] versus non-routine [i.e., to anywhere other than home]), length of stay, hospital charges, and inpatient mortality, were recorded for each patient. Comorbidities were assessed using the Elixhauser method, a well-established technique for quantifying comorbidity burden using ICD-9 diagnosis codes (Elixhauser et al., 1998; Stukenborg et al., 2001). A comorbidity score was produced by adding one point for each comorbidity (maximum possible, 27; paralysis was excluded as a comorbidity category). The degree of trauma was quantified using the International Classification of Disease-9 Based Injury Severity Score (ICISS; Osler et al., 1996), and survival risk ratios published by the American College of Surgeons (American College of Surgeons, Committee on Trauma, 2010), to estimate the probability of survival for each patient given the set of assigned diagnosis codes. The ICISS score was subtracted from 1 and multiplied by 100 to produce a 0 (certain survival) to 100 (certain death) probability of death corresponding to injury severity (i.e., the injury severity score). In-hospital postoperative complications were identified by ICD-9 diagnosis code as follows: renal (584 and 997.5), cardiac (997.1 and 410.0–410.91), neurologic (997.00–997.09), deep-vein thrombosis or pulmonary embolism (415.1, 415.11, 415.19, 451.1, 451.11, 451.19, 451.2, 451.81, 451.9, 453.4, 453.40, 453.41, 453.42, 453.8, and 453.9), pulmonary (507.0, 518.4, 518.5, 518.81, 518.82, 997.3, 997.31, and 997.39), infection (038, 320, 510, 513.1, 519.2, 590.1, 590.80, and 683), and wound complications (998.1–998.9). Where appropriate, the CA SID's “diagnosis present on admission” indicator was used to distinguish comorbidity from complication (e.g., pulmonary embolism; Zhan et al., 2007). Of note, the CA SID does not report neurological outcomes.

Statistical analysis

Retrospective observational studies comparing outcomes between two interventions (e.g., early versus late fusion) are complicated by the fact that patients are not randomized to one group or the other, and thus important differences between treatment groups may exist that can confound the outcomes of interest. Propensity Score Matching (PSM) can be applied to retrospective observational studies to mitigate imbalances in variables known to be associated with assignment to a particular intervention, thereby nullifying these observed differences between treatment groups and improving comparison of outcomes. We performed univariate analysis to identify variables associated with early versus late surgery; p<0.15 was used as a cut-off. We derived an optimal 1:1 PSM matching algorithm from the work of Kleinman and Horton (2010), and Kosanke and Bergstralg (2010), to balance age, sex, comorbidity burden, admission from emergency department, injury severity score, and insurance status between time to surgery groups. The maximum allowable propensity score distance between each matched early/late surgery pair was set to 0.01. In this manner, a cohort of 1126 PSM-matched patients was derived from an initial cohort of 1506 patients. The quality of the PSM match was assessed in two ways. First, we visually assessed propensity score histograms for each time to surgery group before and after matching. Second, standardized differences for each balancing variable were computed; 0.10 was considered well-balanced (Austin and Mamdani, 2006; Rosenbaum and Rubin, 1985).

Patient characteristics and outcomes were compared between time-to-surgery groups by the Rao-Scott chi square test, Fischer exact test, or t-test, as appropriate. Univariate analysis was performed to identify potential predictors of in-hospital complications; variables with p<0.15 were kept. A multivariate model was fit separately for both the PSM-matched and overall cohorts to identify significant predictors of in-hospital complications; p<0.05 was considered statistically significant. All calculations were performed using SAS software (version 9.2; SAS Institute, Inc, Cary, NC) running on Windows XP Pro.

Results

In the initial cohort of 1506 patients, there were statistically significant differences between early and late surgery patients across several variables, including admission source (55% of early versus 76% of late surgery patients were admitted from an emergency department), age (85% of early versus 80% of late surgery patients were 65 or older), comorbidity burden (54% of early versus 64% of late surgery patients had at least one comorbidity), insurance type (50% of early versus 42% of late surgery patients had private insurance), and mean injury severity score (3.5% for early versus 5.4% chance of death for late surgery patients; Table 1). PSM matching was performed over admission source, age, comorbidity score, insurance type, sex, and injury severity score. Standardized differences between time-to-surgery groups for the PSM-matched cohort are listed in Table 2; all values were less than 0.10, indicating that balance was achieved between the groups. Characteristics of the 1126 PSM-matched patients are presented in Table 3; as expected following PSM-matching, there were no significant differences between groups for any of the measured variables.

Table 1.

Patient Characteristics of the Traumatic Thoracolumbar Fracture Fixation Cohort

	Overall		Treated within 72 h		Treated after 72 h
	Cases	(%)	Cases	(%)	Cases	(%)	p Value
Overall	1506	(100)	877	(100)	629	(100)	<0.0001^*
Admitted from emergency department							<0.0001^*
Yes	963	(63.9)	486	(55.4)	477	(75.8)
No	543	(36.1)	391	(44.6)	152	(24.2)
Age							0.0095^*
Under 65	1249	(82.9)	131	(14.9)	126	(20.0)
65 and Older	257	(17.1)	746	(85.1)	503	(80.0)
Elixhauser comorbidity group							<0.0001^*
Zero	630	(41.8)	404	(46.1)	226	(35.9)
One	394	(26.2)	233	(26.6)	161	(25.6)
Two	231	(15.3)	120	(13.7)	111	(17.7)
Three	140	(9.3)	71	(8.1)	69	(11.0)
Four or More	111	(7.4)	49	(5.6)	62	(9.9)
Expected payer							0.0083^*
Private	698	(46.4)	436	(49.7)	262	(41.7)
Medicare	265	(17.6)	142	(16.2)	123	(19.6)
Other including Medicaid	542	(36.0)	299	(34.1)	243	(38.7)
Race/ethnicity							0.6990
Asian or Pacific Islander	71	(4.7)	38	(4.3)	33	(5.3)
Black	33	(2.2)	17	(1.9)	16	(2.5)
Hispanic	291	(19.3)	165	(18.8)	126	(20.0)
White	995	(66.1)	588	(67.1)	407	(64.7)
Other including Native Americans	50	(3.3)	27	(3.1)	23	(3.7)
Unknown	66	(4.4)	42	(4.8)	24	(3.8)
Sex							0.0998
Female	515	(34.3)	315	(36.0)	200	(32.0)
Male	985	(65.7)	559	(64.0)	426	(68.1)
Injury severity score^a	4%	[0, 76]	3.5%	[0, 67]	5.4%	[0, 76]	<0.0001^*

p<0.05 is statistically significant.

An estimate of injury severity derived from the International Classification of Disease-9 Based Injury Severity Score; ranges from 0 (no chance of death) to 100 (certain death). Presented as mean [minimum, maximum].

Table 2.

Standardized Differences of Balancing Variables in the PSM-Matched Cohort

Balancing variable	Standardized difference ^a
Admitted from emergency department	0.008
Age group	0.010
Elixhauser comorbidity score	0.004
Expected payer	0.021
Injury severity score	0.015
Sex	0.035

Standardized differences for each balancing variable used during propensity score modeling. Values less than 0.10 indicate balance was achieved between treatment groups.

PSM, Propensity Score Modeling.

Table 3.

Traumatic Thoracolumbar Fracture Fixation Patient Characteristics of the PSM-Matched Cohort

	Overall		Treated within 72 h		Treated after 72 h
	Cases	(%)	Cases	(%)	Cases	(%)	p Value
Overall	1,126	(100)	563	(100)	563	(100)
Admitted from emergency department							0.8930
Yes	824	(73.2)	411	(73.0)	413	(73.4)
No	302	(26.8)	152	(27.0)	150	(26.6)
Age							0.8678
Under 65	956	(84.9)	479	(85.1)	477	(84.7)
65 and Older	170	(15.1)	84	(14.9)	86	(15.3)
Elixhauser comorbidity group							0.9666
Zero	444	(39.4)	223	(39.6)	221	(39.3)
One	316	(28.1)	156	(27.7)	160	(28.4)
Two	189	(16.8)	96	(17.1)	93	(16.5)
Three	105	(9.3)	50	(8.9)	55	(9.8)
Four or More	72	(6.4)	38	(6.8)	34	(6.0)
Expected payer							0.9089
Private	494	(43.9)	245	(43.5)	249	(44.2)
Medicare	175	(15.5)	86	(15.3)	89	(15.8)
Other including Medicaid	457	(40.6)	232	(41.2)	225	(40.0)
Race/ethnicity							0.9526
Asian or Pacific Islander	54	(4.8)	28	(5.0)	26	(4.6)
Black	28	(2.5)	13	(2.3)	15	(2.7)
Hispanic	238	(21.1)	120	(21.3)	118	(21.0)
White	726	(64.5)	361	(64.1)	365	(64.8)
Other including Native Americans	41	(3.6)	19	(3.4)	22	(3.9)
Unknown	39	(3.5)	22	(3.9)	17	(3.0)
Sex							0.5594
Female	341	(30.3)	166	(29.5)	175	(31.1)
Male	785	(69.7)	397	(70.5)	388	(68.9)
Injury severity score^a	5%	[0, 76%]	4%	[0, 67]	5%	[0, 76]	0.7968

p<0.05 is statistically significant.

PSM, Propensity Score Modeling.

Outcomes for the PSM-matched cohort are shown in Table 4. There were statistically significant differences in in-hospital complication rate, length of stay, and total hospital charges. The overall complication rate was 22.3% (18.7% for early and 25.9% for late surgery patients; p=0.0033). Cardiac, thromboembolic, and pulmonary complication rates were all significantly increased in the late surgery group. The mean [minimum, maximum] and median length of stay in days was 10.0 [2, 113] and 7.0 (Q1:6, Q3:11), respectively, for patients treated within 72 h, and 15.0 [3, 117] and 12.0 (Q1:9, Q3:18), respectively, for patients treated beyond 72 h (p<0.0001). There was no statistically significant difference for in-hospital mortality or rate of non-routine discharge by time to surgery in the PSM-matched group.

Table 4.

Outcomes of the PSM-Matched Cohort for Patients Undergoing Traumatic Thoracolumbar Fracture Fixation

	Overall	Treated within 72 h	Treated after 72 h	p Value
In-hospital complications (%)	22.3	18.7	25.9	0.0033^*
Cardiac	1.6	0.4	2.8	0.0009^*
Infection	2.0	1.2	2.8	0.0579
Neurologic	0.8	0.9	0.7	1.0000
Pulmonary	13.7	10.1	17.2	0.0005^*
Renal	3.2	3.6	2.8	0.4980
Thromboembolic	3.7	2.3	5.2	0.0119^*
Wound	5.2	5.2	5.3	0.8936
In-hospital mortality (%)	1.0	0.5	1.4	0.1305
Length of stay, mean (days)	12.5	10.0	15.0	<0.0001^*
Non-routine discharge (%)	57.0	57.3	56.7	0.8298
Total charges, median ($US)	$232,251	$213,031	$251,151	<0.0001^*

p<0.05 is statistically significant.

PSM, Propensity Score Modeling.

On univariate analysis, admission source, age, insurance, comorbidity burden, injury severity score, and time to surgery, were all associated with in-hospital complications. On multivariate analysis, time to surgery was the strongest predictor of in-hospital complications, with a delay in surgery beyond 72 h associated with a 59% increase in the odds of complication after controlling for other factors (OR [95% CI]=1.592 [1.180, 2.148]). Admission from an emergency department, increasing comorbidity burden, and injury severity score, were also independently associated with an increased risk of complications. The full model is presented in Table 5. This model was also fit to the overall (i.e., pre-PSM matching) cohort; results were similar with the exception that admission from emergency department was not an independent predictor of complications in the un-matched cohort.

Table 5.

Logistic Regression for the Presence of In-Hospital Complications after Traumatic Thoracolumbar Fracture Fixation for the PSM-Matched Cohort

	Odds ratio	95% Confidence interval
Admitted from emergency department
Yes v. No	0.667	[0.476, 0.936]^a
Age
Single year increase	1.011	[1.000, 1.023]
Expected payer
Medicare versus private insurance	1.041	[0.625, 1.732]
All other including Medicaid versus private insurance	1.115	[0.793, 1.567]
Elixhauser comorbidity score
Single point increase	1.388	[1.232, 1.564]^a
Injury severity score
Single percentage point increase	1.047	[1.031, 1.064]^a
Time to surgery
Over 72 h versus within 72 h	1.592	[1.180, 2.148]^a

Statistically significant when the odds ratio does not include 1.000.

PSM, Propensity Score Modeling.

Discussion

The main finding of our study is that early surgery for traumatic thoracic/thoracolumbar fractures was associated with a significantly lower overall complication rate (including cardiac, thromboembolic, and respiratory complications), and decreased hospital length of stay. In addition, in-hospital charges were significantly lower ($38,120 difference) in the early (<72 h) surgery group compared to the delayed surgery group. To our knowledge, this is the largest study to report the impact of timing of surgery on in-hospitalization charges. Multivariate analysis identified time to surgery as the strongest predictor of in-hospital complications, more so than age, medical comorbidities, or injury severity score, each of which was also independently associated with increased complications. These results confirm those of previous smaller single-institution and database studies (Croce et al., 2001; Chipman et al., 2004; Kerwin et al., 2005,2007).

In non-matched cohorts, Croce and associates (2001) retrospectively compared outcomes of early versus late stabilization in 291 patients, of which 79 had thoracic fractures. They found statistically significant differences in several outcomes, including fewer ventilator days (1.2 versus 7.6), fewer days in the intensive care unit (ICU; 7.0 versus 15.1), fewer hospital days (13.0 versus 30.3), and fewer cases of pneumonia (3% versus 37%). This study also reported lower hospital charges in the early surgery group. This was confirmed by the review by Chipman and associates (2004) of 146 patients, who reported shorter hospitalization and fewer complications in the early surgery cohort. Kerwin and co-workers (2005) retrospectively reviewed data of 299 patients from the National Trauma Database who underwent surgical stabilization for spine fractures. Of these, 90 patients had thoracic fractures and 49 underwent early stabilization (within 72 h). Groups were matched for age, injury score, and Glasgow Coma Scale score, but not for comorbidities. Patients who underwent early stabilization likewise had significantly shorter hospital length of stay (10.1 versus 30.5 days), ICU length of stay (2.3 versus 13.1 days), and a lower incidence of pneumonia (6.5% versus 33.3%). Single-institution studies are summarized in recent comprehensive reviews. Dimar and associates (2010) concluded that “early stabilization consistently leads to shorter hospital stays, shorter intensive care unit stays, fewer days on mechanical ventilation, and lower pulmonary complications.” Bellabarba and colleagues (2010) reported similar results from their systematic review, and recommended early stabilization (<72 hours) in all patients with unstable thoracic fractures.

Much of the controversy surrounding the timing of surgery in traumatic thoracic/thoracolumbar fractures concerns the possible deleterious effect of early surgery on neurological outcomes. Unfortunately, the CA SID does not report neurological outcomes. Regardless, the controversy over neurological outcomes and timing of surgery is likely to continue until the results of randomized clinical trials are available. The ongoing Surgical Treatment of Acute Spinal Cord Injury Study (STASCIS) trial is the main prospective, multicenter clinical trial designed to compare neurological outcomes from early and delayed surgery in acute spinal cord-injured patients (Fehlings et al., 2009). Preliminary results suggest that early surgery improves clinical outcomes in cervical spinal cord injury (Furlan et al., 2011). In animals, definitive evidence exists that early decompression improves outcomes (Carlson et al., 2003; Delamarter et al., 1995; Fehlings and Wilson, 2010b). One concern regarding early stabilization involves the availability of systemic resources for consistent achievement of early surgery. Unavailability of 24-hour CT scanning and MRI scanning on the weekends can limit the ability to operate within 24 h of injury. In a worldwide survey of 2000 neurosurgeons, only 48.3% believed surgical decompression could be accomplished within 24 h of admission at their hospitals. Despite that, the majority (80%) recommended early stabilization for thoracolumbar spinal fractures (Fehlings et al., 2010a).

Despite the lack of results from prospective, randomized data, there is an emerging consensus toward early stabilization (Bellabarba et al., 2010; Dimar et al., 2010; Fehlings and Wilson, 2010; Fehlings et al., 2010a,2010b; Furlan et al., 2011). The Spine Trauma Study Group concluded, “based on the emerging evidence, it is recommended that patients with major spine trauma or spinal cord injury should undergo surgical treatment when deemed appropriate, within 24 hours of their injury” (Fehlings and Wilson, 2010b). Our results add to this growing body of evidence strengthening arguments for early stabilization.

Limitations of our study include the absence of true randomization, though this was compensated for by the use of PSM, which is a well-validated method of creating comparable non-randomized matched cohorts. The use of ICD-9 codes in selecting cohorts is a well-known weakness of administrative database research due to potential coding errors. Of note, our results agree well with the results from the independent National Trauma Database. Another limitation is that the CA SID does not allow us to determine time from injury to surgery (which may be more important than time from admission to surgery as a predictor of complication rate, hospital stay, and discharge), particularly if patients are transferred from outside hospitals after initial stabilization, or if there are other causes of surgical delay. It is important to note that the CA SID does not provide information about why surgeries were postponed. The strengths of our study include the large number of patients in each cohort, and the use of PSM on age, injury severity score, comorbidity, and other factors we identified as predictors of early or late surgery.

In summary, by the use of a propensity score-matched, retrospective cohort study of a large administrative database, we found arguments that favor early spinal surgery (within 72 h) in those with traumatic thoracic/thoracolumbar fractures to reduce in-hospital complications, hospital length of stay, and resource utilization. These results provide further support to the emerging literature and professional consensus regarding the importance of early stabilization of traumatic thoracic/thoracolumbar fractures to improve patient outcomes and limit hospitalization costs.

Footnotes

Author Disclosure Statement

No conflicting financial interests exist.

References

American College of Surgeons, Committee on Trauma. 2010. http://www.facs.org/trauma/ntdb/srr.zip. 2010 September 15.

Austin

P.C.

, Mamdani

M.M.

2006. A comparison of propensity score methods: A case-study estimating the effectiveness of post-AMI statin use. Stat. Med., 25:2084–2106.

Bellabarba

, Fisher

, Chapman

J.R.

, Dettori

J.R.

, Norvell

D.C.

2010. Does early fracture fixation of thoracolumbar spine fractures decrease morbidity or mortality? Spine (Phila. Pa. 1976), 35:S138–S145.

Carlson

G.D.

, Gorden

C.D.

, Oliff

H.S.

, Pillai

J.J.

, LaManna

J.C.

2003. Sustained spinal cord compression: Part I: Time-dependent effect on long-term pathophysiology. J. Bone Joint Surg. Am., 85:A86–A94.

Chipman

J.G.

, Deuser

W.E.

, Beilman

G.J.

2004. Early surgery for thoracolumbar spine injuries decreases complications. J. Trauma, 56:52–57.

Croce

M.A.

, Bee

T.K.

, Pritchard

, Miller

P.R.

, Fabian

T.C.

2001. Does optimal timing for spine fracture fixation exist? Ann. Surg., 233:851–858.

Delamarter

R.B.

, Sherman

, Carr

J.B.

1995. Pathophysiology of spinal cord injury. recovery after immediate and delayed decompression. J. Bone Joint Surg. Am., 77:1042–1049.

Dimar

J.R.

, Carreon

L.Y.

, Riina

, Schwartz

D.G.

, Harris

M.B.

2010. Early versus late stabilization of the spine in the polytrauma patient. Spine (Phila. Pa. 1976), 35:S187–S192.

Elixhauser

, Steiner

, Harris

D.R.

, Coffey

R.M.

1998. Comorbidity measures for use with administrative data. Med. Care, 36:8–27.

10.

Fehlings

M.G.

, Wilson

J.R.

2010. Timing of surgical intervention in spinal trauma: What does the evidence indicate? Spine (Phila. Pa. 1976), 35:S159–S160.

11.

Fehlings

M.G.

, Rabin

, Sears

, Cadotte

D.W.

, Aarabi

2010a. Current practice in the timing of surgical intervention in spinal cord injury. Spine (Phila. Pa. 1976), 35:S166–S173.

12.

Fehlings

M.G.

, Vaccaro

, Aarabi

, Shaffrey

, Dvorak

, Fisher

, Arnold

P.M.

, Masicotte

, Rampersaud

, Lewis

2009. One year outcomes of the STASCIS study: A prospective, multicenter trial to evaluate the role and timing of decompression in patients with cervical spinal cord injury. J. Neurosurg. Spine, 110:A1044.

13.

Fehlings

M.G.

, Wilson

J.R.

, Dvorak

M.F.

, Vaccaro

, Fisher

C.G.

2010b. The challenges of managing spine and spinal cord injuries: An evolving consensus and opportunities for change. Spine (Phila. Pa. 1976), 35:S161–S165.

14.

Furlan

J.C.

, Noonan

, Cadotte

D.W.

, Fehlings

M.G.

2011. Timing of decompressive surgery of spinal cord after traumatic spinal cord injury: An evidence-based examination of pre-clinical and clinical studies. J. Neurotrauma, 28:1371–1399.

15.

Healthcare Cost and Utilization Project (HCUP). 2003–2004. HCUP State Inpatient Databases (SID) Agency for Healthcare Research and Quality: Rockville, MD www.hcup-us.ahrq.gov/sidoverview.jsp.

16.

Kerwin

A.J.

, Frykberg

E.R.

, Schinco

M.A.

, Griffen

M.M.

, Arce

C.A.

, Nguyen

T.Q.

, Tepas

J.J.

3rd . 2007. The effect of early surgical treatment of traumatic spine injuries on patient mortality. J. Trauma, 63:1308–1313.

17.

Kerwin

A.J.

, Frykberg

E.R.

, Schinco

M.A.

, Griffen

M.M.

, Murphy

, Tepas

J.J.

3rd . 2005. The effect of early spine fixation on non-neurologic outcome. J. Trauma, 58:15–21.

18.

Kleinman

, Horton

2010. Example 7.35: Propensity score matching. http://sas-and-r.blogspot.com/2010/05/example-735-propensity-score-matchingn.html. 2010 August 31.

19.

Kosanke

, Bergstralg

2010. Vmatch. http://mayoresearch.mayo.edu/biostat/sasmacros.cfm. 2010 August 31.

20.

Osler

, Rutledge

, Deis

, Bedrick

1996. ICISS: An international classification of disease-9 based injury severity score. J. Trauma, 41:380–388.

21.

Romano

P.S.

, Mark

D.H.

1994. Bias in the coding of hospital discharge data and its implications for quality assessment. Med. Care, 32:81–90.

22.

Rosenbaum

P.R.

, Rubin

D.B.

1985. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. American Statistician, 39:33–39.

23.

Stukenborg

G.J.

, Wagner

D.P.

, Connors

A.F.

Jr.

2001. Comparison of the performance of two comorbidity measures, with and without information from prior hospitalizations. Med. Care, 39:727–739.

24.

Thomas

K.C.

, Bailey

C.S.

, Dvorak

M.F.

, Kwon

, Fisher

2006. Comparison of operative and nonoperative treatment for thoracolumbar burst fractures in patients without neurological deficit: A systematic review. J. Neurosurg. Spine, 4:351–358.

25.

Zhan

, Battles

, Chiang

Y.P.

, Hunt

2007. The validity of ICD-9-CM codes in identifying postoperative deep vein thrombosis and pulmonary embolism. Jt. Comm. J. Qual. Patient Saf., 33:326–331.